{
  "best_metric": null,
  "best_model_checkpoint": null,
  "epoch": 0.5,
  "global_step": 25177,
  "is_hyper_param_search": false,
  "is_local_process_zero": true,
  "is_world_process_zero": true,
  "log_history": [
    {
      "epoch": 0.0,
      "learning_rate": 9.92063492063492e-07,
      "loss": 11.3405,
      "theoretical_loss": 20.81281629304291,
      "tokens_seen": 65536
    },
    {
      "epoch": 0.0,
      "learning_rate": 1.984126984126984e-06,
      "loss": 11.3718,
      "theoretical_loss": 17.566199595824404,
      "tokens_seen": 131072
    },
    {
      "epoch": 0.0,
      "learning_rate": 2.9761904761904763e-06,
      "loss": 11.2959,
      "theoretical_loss": 15.939475584332328,
      "tokens_seen": 196608
    },
    {
      "epoch": 0.0,
      "learning_rate": 3.968253968253968e-06,
      "loss": 11.0056,
      "theoretical_loss": 14.89231524748433,
      "tokens_seen": 262144
    },
    {
      "epoch": 0.0,
      "learning_rate": 4.96031746031746e-06,
      "loss": 10.7017,
      "theoretical_loss": 14.136215429258735,
      "tokens_seen": 327680
    },
    {
      "epoch": 0.0,
      "learning_rate": 5.9523809523809525e-06,
      "loss": 10.7198,
      "theoretical_loss": 13.552559964045983,
      "tokens_seen": 393216
    },
    {
      "epoch": 0.0,
      "learning_rate": 6.944444444444444e-06,
      "loss": 10.4947,
      "theoretical_loss": 13.08180749289695,
      "tokens_seen": 458752
    },
    {
      "epoch": 0.0,
      "learning_rate": 7.936507936507936e-06,
      "loss": 10.1018,
      "theoretical_loss": 12.690128116979084,
      "tokens_seen": 524288
    },
    {
      "epoch": 0.0,
      "learning_rate": 8.928571428571428e-06,
      "loss": 10.1205,
      "theoretical_loss": 12.356590955369384,
      "tokens_seen": 589824
    },
    {
      "epoch": 0.0,
      "learning_rate": 9.92063492063492e-06,
      "loss": 10.1014,
      "theoretical_loss": 12.067411098530838,
      "tokens_seen": 655360
    },
    {
      "epoch": 0.0,
      "learning_rate": 1.0912698412698412e-05,
      "loss": 10.0515,
      "theoretical_loss": 11.813064722597437,
      "tokens_seen": 720896
    },
    {
      "epoch": 0.0,
      "learning_rate": 1.1904761904761905e-05,
      "loss": 9.8482,
      "theoretical_loss": 11.58671770020249,
      "tokens_seen": 786432
    },
    {
      "epoch": 0.0,
      "learning_rate": 1.2896825396825396e-05,
      "loss": 9.5829,
      "theoretical_loss": 11.383312631682548,
      "tokens_seen": 851968
    },
    {
      "epoch": 0.0,
      "learning_rate": 1.3888888888888888e-05,
      "loss": 9.5705,
      "theoretical_loss": 11.199010193607633,
      "tokens_seen": 917504
    },
    {
      "epoch": 0.0,
      "learning_rate": 1.4880952380952381e-05,
      "loss": 9.3347,
      "theoretical_loss": 11.030832409473673,
      "tokens_seen": 983040
    },
    {
      "epoch": 0.0,
      "learning_rate": 1.5873015873015872e-05,
      "loss": 9.3549,
      "theoretical_loss": 10.87642657795271,
      "tokens_seen": 1048576
    },
    {
      "epoch": 0.0,
      "learning_rate": 1.6865079365079364e-05,
      "loss": 9.4759,
      "theoretical_loss": 10.733904231558485,
      "tokens_seen": 1114112
    },
    {
      "epoch": 0.0,
      "learning_rate": 1.7857142857142855e-05,
      "loss": 9.6373,
      "theoretical_loss": 10.60172836772604,
      "tokens_seen": 1179648
    },
    {
      "epoch": 0.0,
      "learning_rate": 1.884920634920635e-05,
      "loss": 9.6978,
      "theoretical_loss": 10.478632663852402,
      "tokens_seen": 1245184
    },
    {
      "epoch": 0.0,
      "learning_rate": 1.984126984126984e-05,
      "loss": 9.3671,
      "theoretical_loss": 10.36356243525909,
      "tokens_seen": 1310720
    },
    {
      "epoch": 0.0,
      "learning_rate": 2.0833333333333333e-05,
      "loss": 9.1034,
      "theoretical_loss": 10.255630712392506,
      "tokens_seen": 1376256
    },
    {
      "epoch": 0.0,
      "learning_rate": 2.1825396825396824e-05,
      "loss": 9.3646,
      "theoretical_loss": 10.154085044765779,
      "tokens_seen": 1441792
    },
    {
      "epoch": 0.0,
      "learning_rate": 2.2817460317460315e-05,
      "loss": 9.5759,
      "theoretical_loss": 10.058282053228359,
      "tokens_seen": 1507328
    },
    {
      "epoch": 0.0,
      "learning_rate": 2.380952380952381e-05,
      "loss": 9.4747,
      "theoretical_loss": 9.967667670336038,
      "tokens_seen": 1572864
    },
    {
      "epoch": 0.0,
      "objective/train/advantage_avg": 0.49520397186279297,
      "objective/train/docs_used": 13112,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 8.179563522338867,
      "objective/train/original_loss": 8.179564476013184,
      "objective/train/theoretical_loss": 9.881761617888868,
      "objective/train/tokens_used": 22098400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24528734385967255,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507673025131226,
      "objective/train/weighted_lm_loss": 8.594050407409668,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0363596677780151,
      "theoretical_loss": 9.881761617888868,
      "tokens_seen": 1638400
    },
    {
      "epoch": 0.0,
      "learning_rate": 2.48015873015873e-05,
      "loss": 9.2302,
      "theoretical_loss": 9.881761617888868,
      "tokens_seen": 1638400
    },
    {
      "epoch": 0.0,
      "learning_rate": 2.5793650793650793e-05,
      "loss": 8.7651,
      "theoretical_loss": 9.80014508303632,
      "tokens_seen": 1703936
    },
    {
      "epoch": 0.0,
      "learning_rate": 2.6785714285714284e-05,
      "loss": 9.1043,
      "theoretical_loss": 9.722450838403205,
      "tokens_seen": 1769472
    },
    {
      "epoch": 0.0,
      "learning_rate": 2.7777777777777776e-05,
      "loss": 9.157,
      "theoretical_loss": 9.648355250577305,
      "tokens_seen": 1835008
    },
    {
      "epoch": 0.0,
      "learning_rate": 2.876984126984127e-05,
      "loss": 9.4713,
      "theoretical_loss": 9.5775717626414,
      "tokens_seen": 1900544
    },
    {
      "epoch": 0.0,
      "learning_rate": 2.9761904761904762e-05,
      "loss": 8.199,
      "theoretical_loss": 9.509845538260612,
      "tokens_seen": 1966080
    },
    {
      "epoch": 0.0,
      "learning_rate": 3.075396825396825e-05,
      "loss": 9.0097,
      "theoretical_loss": 9.444949029127695,
      "tokens_seen": 2031616
    },
    {
      "epoch": 0.0,
      "learning_rate": 3.1746031746031745e-05,
      "loss": 9.3904,
      "theoretical_loss": 9.382678282406216,
      "tokens_seen": 2097152
    },
    {
      "epoch": 0.0,
      "learning_rate": 3.273809523809524e-05,
      "loss": 9.0986,
      "theoretical_loss": 9.32284984572974,
      "tokens_seen": 2162688
    },
    {
      "epoch": 0.0,
      "learning_rate": 3.373015873015873e-05,
      "loss": 9.4393,
      "theoretical_loss": 9.265298158156035,
      "tokens_seen": 2228224
    },
    {
      "epoch": 0.0,
      "learning_rate": 3.472222222222222e-05,
      "loss": 8.6919,
      "theoretical_loss": 9.209873338940517,
      "tokens_seen": 2293760
    },
    {
      "epoch": 0.0,
      "learning_rate": 3.571428571428571e-05,
      "loss": 8.7293,
      "theoretical_loss": 9.156439304004051,
      "tokens_seen": 2359296
    },
    {
      "epoch": 0.0,
      "learning_rate": 3.670634920634921e-05,
      "loss": 8.6656,
      "theoretical_loss": 9.10487215390911,
      "tokens_seen": 2424832
    },
    {
      "epoch": 0.0,
      "learning_rate": 3.76984126984127e-05,
      "loss": 8.6025,
      "theoretical_loss": 9.055058788029493,
      "tokens_seen": 2490368
    },
    {
      "epoch": 0.0,
      "learning_rate": 3.8690476190476195e-05,
      "loss": 8.7398,
      "theoretical_loss": 9.006895708139588,
      "tokens_seen": 2555904
    },
    {
      "epoch": 0.0,
      "learning_rate": 3.968253968253968e-05,
      "loss": 8.3473,
      "theoretical_loss": 8.960287981405116,
      "tokens_seen": 2621440
    },
    {
      "epoch": 0.0,
      "learning_rate": 4.067460317460318e-05,
      "loss": 8.7144,
      "theoretical_loss": 8.91514833813637,
      "tokens_seen": 2686976
    },
    {
      "epoch": 0.0,
      "learning_rate": 4.1666666666666665e-05,
      "loss": 8.896,
      "theoretical_loss": 8.871396383973984,
      "tokens_seen": 2752512
    },
    {
      "epoch": 0.0,
      "learning_rate": 4.265873015873016e-05,
      "loss": 8.8065,
      "theoretical_loss": 8.828957909649258,
      "tokens_seen": 2818048
    },
    {
      "epoch": 0.0,
      "learning_rate": 4.365079365079365e-05,
      "loss": 8.9836,
      "theoretical_loss": 8.787764284274173,
      "tokens_seen": 2883584
    },
    {
      "epoch": 0.0,
      "learning_rate": 4.464285714285714e-05,
      "loss": 8.8624,
      "theoretical_loss": 8.747751920407216,
      "tokens_seen": 2949120
    },
    {
      "epoch": 0.0,
      "learning_rate": 4.563492063492063e-05,
      "loss": 8.8054,
      "theoretical_loss": 8.708861801016592,
      "tokens_seen": 3014656
    },
    {
      "epoch": 0.0,
      "learning_rate": 4.6626984126984126e-05,
      "loss": 8.2214,
      "theoretical_loss": 8.671039060004606,
      "tokens_seen": 3080192
    },
    {
      "epoch": 0.0,
      "learning_rate": 4.761904761904762e-05,
      "loss": 8.0754,
      "theoretical_loss": 8.634232609231233,
      "tokens_seen": 3145728
    },
    {
      "epoch": 0.0,
      "learning_rate": 4.8611111111111115e-05,
      "loss": 8.3468,
      "theoretical_loss": 8.598394806032085,
      "tokens_seen": 3211264
    },
    {
      "epoch": 0.0,
      "objective/train/advantage_avg": 0.49455350637435913,
      "objective/train/docs_used": 13755,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 7.334794998168945,
      "objective/train/original_loss": 7.334794521331787,
      "objective/train/theoretical_loss": 8.563481156106828,
      "objective/train/tokens_used": 23736800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24466021358966827,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506988763809204,
      "objective/train/weighted_lm_loss": 7.706252574920654,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0363596677780151,
      "theoretical_loss": 8.563481156106828,
      "tokens_seen": 3276800
    },
    {
      "epoch": 0.0,
      "learning_rate": 4.96031746031746e-05,
      "loss": 8.2495,
      "theoretical_loss": 8.563481156106828,
      "tokens_seen": 3276800
    },
    {
      "epoch": 0.0,
      "learning_rate": 5.05952380952381e-05,
      "loss": 8.1712,
      "theoretical_loss": 8.529450047390876,
      "tokens_seen": 3342336
    },
    {
      "epoch": 0.0,
      "learning_rate": 5.1587301587301586e-05,
      "loss": 7.9962,
      "theoretical_loss": 8.496262511141762,
      "tokens_seen": 3407872
    },
    {
      "epoch": 0.0,
      "learning_rate": 5.257936507936508e-05,
      "loss": 8.3485,
      "theoretical_loss": 8.463882006992947,
      "tokens_seen": 3473408
    },
    {
      "epoch": 0.0,
      "learning_rate": 5.357142857142857e-05,
      "loss": 8.179,
      "theoretical_loss": 8.43227422916854,
      "tokens_seen": 3538944
    },
    {
      "epoch": 0.0,
      "learning_rate": 5.4563492063492063e-05,
      "loss": 8.1309,
      "theoretical_loss": 8.401406931426477,
      "tokens_seen": 3604480
    },
    {
      "epoch": 0.0,
      "learning_rate": 5.555555555555555e-05,
      "loss": 7.8855,
      "theoretical_loss": 8.37124976861597,
      "tokens_seen": 3670016
    },
    {
      "epoch": 0.0,
      "learning_rate": 5.6547619047619046e-05,
      "loss": 7.5677,
      "theoretical_loss": 8.341774153006837,
      "tokens_seen": 3735552
    },
    {
      "epoch": 0.0,
      "learning_rate": 5.753968253968254e-05,
      "loss": 8.0239,
      "theoretical_loss": 8.312953123781089,
      "tokens_seen": 3801088
    },
    {
      "epoch": 0.0,
      "learning_rate": 5.8531746031746036e-05,
      "loss": 8.2425,
      "theoretical_loss": 8.284761228276942,
      "tokens_seen": 3866624
    },
    {
      "epoch": 0.0,
      "learning_rate": 5.9523809523809524e-05,
      "loss": 8.1922,
      "theoretical_loss": 8.257174413747624,
      "tokens_seen": 3932160
    },
    {
      "epoch": 0.0,
      "learning_rate": 6.051587301587302e-05,
      "loss": 7.9839,
      "theoretical_loss": 8.230169928545873,
      "tokens_seen": 3997696
    },
    {
      "epoch": 0.0,
      "learning_rate": 6.15079365079365e-05,
      "loss": 8.4934,
      "theoretical_loss": 8.20372623177373,
      "tokens_seen": 4063232
    },
    {
      "epoch": 0.0,
      "learning_rate": 6.25e-05,
      "loss": 7.8083,
      "theoretical_loss": 8.177822910548805,
      "tokens_seen": 4128768
    },
    {
      "epoch": 0.0,
      "learning_rate": 6.349206349206349e-05,
      "loss": 7.8763,
      "theoretical_loss": 8.152440604135377,
      "tokens_seen": 4194304
    },
    {
      "epoch": 0.0,
      "learning_rate": 6.448412698412699e-05,
      "loss": 7.3497,
      "theoretical_loss": 8.127560934273259,
      "tokens_seen": 4259840
    },
    {
      "epoch": 0.0,
      "learning_rate": 6.547619047619048e-05,
      "loss": 7.8516,
      "theoretical_loss": 8.103166441111469,
      "tokens_seen": 4325376
    },
    {
      "epoch": 0.0,
      "learning_rate": 6.646825396825397e-05,
      "loss": 8.0264,
      "theoretical_loss": 8.0792405242184,
      "tokens_seen": 4390912
    },
    {
      "epoch": 0.0,
      "learning_rate": 6.746031746031745e-05,
      "loss": 7.4027,
      "theoretical_loss": 8.055767388197175,
      "tokens_seen": 4456448
    },
    {
      "epoch": 0.0,
      "learning_rate": 6.845238095238096e-05,
      "loss": 7.5691,
      "theoretical_loss": 8.032731992484768,
      "tokens_seen": 4521984
    },
    {
      "epoch": 0.0,
      "learning_rate": 6.944444444444444e-05,
      "loss": 7.7871,
      "theoretical_loss": 8.010120004957596,
      "tokens_seen": 4587520
    },
    {
      "epoch": 0.0,
      "learning_rate": 7.043650793650793e-05,
      "loss": 7.4956,
      "theoretical_loss": 7.987917759005139,
      "tokens_seen": 4653056
    },
    {
      "epoch": 0.0,
      "learning_rate": 7.142857142857142e-05,
      "loss": 7.6866,
      "theoretical_loss": 7.966112213767561,
      "tokens_seen": 4718592
    },
    {
      "epoch": 0.0,
      "learning_rate": 7.242063492063492e-05,
      "loss": 7.3368,
      "theoretical_loss": 7.944690917263747,
      "tokens_seen": 4784128
    },
    {
      "epoch": 0.0,
      "learning_rate": 7.341269841269842e-05,
      "loss": 7.6023,
      "theoretical_loss": 7.923641972163278,
      "tokens_seen": 4849664
    },
    {
      "epoch": 0.0,
      "objective/train/advantage_avg": 0.4955427050590515,
      "objective/train/docs_used": 15099,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 6.682462692260742,
      "objective/train/original_loss": 6.682462692260742,
      "objective/train/theoretical_loss": 7.902954003979827,
      "objective/train/tokens_used": 25375200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24560633301734924,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0508027076721191,
      "objective/train/weighted_lm_loss": 7.021694183349609,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 7.902954003979827,
      "tokens_seen": 4915200
    },
    {
      "epoch": 0.0,
      "learning_rate": 7.440476190476191e-05,
      "loss": 7.2096,
      "theoretical_loss": 7.902954003979827,
      "tokens_seen": 4915200
    },
    {
      "epoch": 0.0,
      "learning_rate": 7.53968253968254e-05,
      "loss": 7.2229,
      "theoretical_loss": 7.882616131484962,
      "tokens_seen": 4980736
    },
    {
      "epoch": 0.0,
      "learning_rate": 7.63888888888889e-05,
      "loss": 7.3944,
      "theoretical_loss": 7.862617939160387,
      "tokens_seen": 5046272
    },
    {
      "epoch": 0.0,
      "learning_rate": 7.738095238095239e-05,
      "loss": 7.2041,
      "theoretical_loss": 7.8429494515236975,
      "tokens_seen": 5111808
    },
    {
      "epoch": 0.0,
      "learning_rate": 7.837301587301588e-05,
      "loss": 7.2338,
      "theoretical_loss": 7.823601109178073,
      "tokens_seen": 5177344
    },
    {
      "epoch": 0.0,
      "learning_rate": 7.936507936507937e-05,
      "loss": 7.0693,
      "theoretical_loss": 7.804563746449924,
      "tokens_seen": 5242880
    },
    {
      "epoch": 0.0,
      "learning_rate": 8.035714285714287e-05,
      "loss": 7.4411,
      "theoretical_loss": 7.785828570490832,
      "tokens_seen": 5308416
    },
    {
      "epoch": 0.0,
      "learning_rate": 8.134920634920635e-05,
      "loss": 7.1094,
      "theoretical_loss": 7.767387141731124,
      "tokens_seen": 5373952
    },
    {
      "epoch": 0.0,
      "learning_rate": 8.234126984126984e-05,
      "loss": 7.0033,
      "theoretical_loss": 7.7492313555823795,
      "tokens_seen": 5439488
    },
    {
      "epoch": 0.0,
      "learning_rate": 8.333333333333333e-05,
      "loss": 7.3349,
      "theoretical_loss": 7.731353425295078,
      "tokens_seen": 5505024
    },
    {
      "epoch": 0.0,
      "learning_rate": 8.432539682539683e-05,
      "loss": 7.1884,
      "theoretical_loss": 7.713745865885679,
      "tokens_seen": 5570560
    },
    {
      "epoch": 0.0,
      "learning_rate": 8.531746031746032e-05,
      "loss": 7.2112,
      "theoretical_loss": 7.696401479054694,
      "tokens_seen": 5636096
    },
    {
      "epoch": 0.0,
      "learning_rate": 8.630952380952381e-05,
      "loss": 7.0943,
      "theoretical_loss": 7.67931333902394,
      "tokens_seen": 5701632
    },
    {
      "epoch": 0.0,
      "learning_rate": 8.73015873015873e-05,
      "loss": 6.9196,
      "theoretical_loss": 7.662474779227088,
      "tokens_seen": 5767168
    },
    {
      "epoch": 0.0,
      "learning_rate": 8.82936507936508e-05,
      "loss": 6.9028,
      "theoretical_loss": 7.645879379793039,
      "tokens_seen": 5832704
    },
    {
      "epoch": 0.0,
      "learning_rate": 8.928571428571429e-05,
      "loss": 7.0041,
      "theoretical_loss": 7.629520955766621,
      "tokens_seen": 5898240
    },
    {
      "epoch": 0.0,
      "learning_rate": 9.027777777777777e-05,
      "loss": 6.4774,
      "theoretical_loss": 7.613393546015455,
      "tokens_seen": 5963776
    },
    {
      "epoch": 0.0,
      "learning_rate": 9.126984126984126e-05,
      "loss": 6.974,
      "theoretical_loss": 7.59749140277604,
      "tokens_seen": 6029312
    },
    {
      "epoch": 0.0,
      "learning_rate": 9.226190476190476e-05,
      "loss": 7.118,
      "theoretical_loss": 7.581808981795648,
      "tokens_seen": 6094848
    },
    {
      "epoch": 0.0,
      "learning_rate": 9.325396825396825e-05,
      "loss": 6.9765,
      "theoretical_loss": 7.56634093303013,
      "tokens_seen": 6160384
    },
    {
      "epoch": 0.0,
      "learning_rate": 9.424603174603175e-05,
      "loss": 6.8346,
      "theoretical_loss": 7.5510820918607084,
      "tokens_seen": 6225920
    },
    {
      "epoch": 0.0,
      "learning_rate": 9.523809523809524e-05,
      "loss": 7.0484,
      "theoretical_loss": 7.536027470795679,
      "tokens_seen": 6291456
    },
    {
      "epoch": 0.0,
      "learning_rate": 9.623015873015874e-05,
      "loss": 7.1334,
      "theoretical_loss": 7.521172251625522,
      "tokens_seen": 6356992
    },
    {
      "epoch": 0.0,
      "learning_rate": 9.722222222222223e-05,
      "loss": 6.6474,
      "theoretical_loss": 7.506511778002256,
      "tokens_seen": 6422528
    },
    {
      "epoch": 0.0,
      "learning_rate": 9.821428571428572e-05,
      "loss": 6.7217,
      "theoretical_loss": 7.492041548416008,
      "tokens_seen": 6488064
    },
    {
      "epoch": 0.0,
      "objective/train/advantage_avg": 0.4945422410964966,
      "objective/train/docs_used": 15777,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 6.93560266494751,
      "objective/train/original_loss": 6.93560266494751,
      "objective/train/theoretical_loss": 7.477757209543791,
      "objective/train/tokens_used": 27013600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24466240406036377,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050697922706604,
      "objective/train/weighted_lm_loss": 7.286866188049316,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0408108234405518,
      "theoretical_loss": 7.477757209543791,
      "tokens_seen": 6553600
    },
    {
      "epoch": 0.0,
      "learning_rate": 9.92063492063492e-05,
      "loss": 6.7211,
      "theoretical_loss": 7.477757209543791,
      "tokens_seen": 6553600
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00010019841269841271,
      "loss": 6.819,
      "theoretical_loss": 7.463654549947222,
      "tokens_seen": 6619136
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.0001011904761904762,
      "loss": 6.6854,
      "theoretical_loss": 7.449729494097676,
      "tokens_seen": 6684672
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00010218253968253968,
      "loss": 7.0264,
      "theoretical_loss": 7.435978096708778,
      "tokens_seen": 6750208
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00010317460317460317,
      "loss": 6.8935,
      "theoretical_loss": 7.422396537357665,
      "tokens_seen": 6815744
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00010416666666666667,
      "loss": 6.8563,
      "theoretical_loss": 7.408981115377635,
      "tokens_seen": 6881280
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00010515873015873016,
      "loss": 6.908,
      "theoretical_loss": 7.395728245006104,
      "tokens_seen": 6946816
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00010615079365079365,
      "loss": 6.9078,
      "theoretical_loss": 7.382634450772793,
      "tokens_seen": 7012352
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00010714285714285714,
      "loss": 6.9631,
      "theoretical_loss": 7.369696363114134,
      "tokens_seen": 7077888
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00010813492063492064,
      "loss": 6.7264,
      "theoretical_loss": 7.356910714200849,
      "tokens_seen": 7143424
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00010912698412698413,
      "loss": 6.5155,
      "theoretical_loss": 7.344274333966448,
      "tokens_seen": 7208960
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00011011904761904761,
      "loss": 6.8308,
      "theoretical_loss": 7.331784146325278,
      "tokens_seen": 7274496
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.0001111111111111111,
      "loss": 6.516,
      "theoretical_loss": 7.319437165569436,
      "tokens_seen": 7340032
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.0001121031746031746,
      "loss": 6.7416,
      "theoretical_loss": 7.307230492934584,
      "tokens_seen": 7405568
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00011309523809523809,
      "loss": 6.9768,
      "theoretical_loss": 7.295161313325323,
      "tokens_seen": 7471104
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00011408730158730158,
      "loss": 6.6897,
      "theoretical_loss": 7.283226892191411,
      "tokens_seen": 7536640
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00011507936507936508,
      "loss": 6.4802,
      "theoretical_loss": 7.271424572546591,
      "tokens_seen": 7602176
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00011607142857142858,
      "loss": 6.7345,
      "theoretical_loss": 7.259751772122383,
      "tokens_seen": 7667712
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00011706349206349207,
      "loss": 6.8386,
      "theoretical_loss": 7.2482059806496295,
      "tokens_seen": 7733248
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00011805555555555556,
      "loss": 6.2228,
      "theoretical_loss": 7.236784757261022,
      "tokens_seen": 7798784
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00011904761904761905,
      "loss": 6.3882,
      "theoretical_loss": 7.225485728008257,
      "tokens_seen": 7864320
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00012003968253968255,
      "loss": 6.948,
      "theoretical_loss": 7.214306583487879,
      "tokens_seen": 7929856
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00012103174603174604,
      "loss": 6.4502,
      "theoretical_loss": 7.203245076570161,
      "tokens_seen": 7995392
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00012202380952380953,
      "loss": 6.6758,
      "theoretical_loss": 7.192299020225775,
      "tokens_seen": 8060928
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.000123015873015873,
      "loss": 6.649,
      "theoretical_loss": 7.181466285445275,
      "tokens_seen": 8126464
    },
    {
      "epoch": 0.0,
      "objective/train/advantage_avg": 0.4924044907093048,
      "objective/train/docs_used": 17166,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 6.253177642822266,
      "objective/train/original_loss": 6.253178119659424,
      "objective/train/theoretical_loss": 7.170744799246725,
      "objective/train/tokens_used": 28652000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24257515370845795,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0504733324050903,
      "objective/train/weighted_lm_loss": 6.56821346282959,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0425469875335693,
      "theoretical_loss": 7.170744799246725,
      "tokens_seen": 8192000
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.0001240079365079365,
      "loss": 6.7362,
      "theoretical_loss": 7.170744799246725,
      "tokens_seen": 8192000
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.000125,
      "loss": 6.5299,
      "theoretical_loss": 7.160132542767032,
      "tokens_seen": 8257536
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.0001259920634920635,
      "loss": 6.5561,
      "theoretical_loss": 7.149627549432898,
      "tokens_seen": 8323072
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00012698412698412698,
      "loss": 6.5383,
      "theoretical_loss": 7.139227903207399,
      "tokens_seen": 8388608
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00012797619047619048,
      "loss": 6.362,
      "theoretical_loss": 7.128931736908553,
      "tokens_seen": 8454144
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00012896825396825398,
      "loss": 6.3643,
      "theoretical_loss": 7.118737230596375,
      "tokens_seen": 8519680
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00012996031746031748,
      "loss": 6.6329,
      "theoretical_loss": 7.108642610025104,
      "tokens_seen": 8585216
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00013095238095238096,
      "loss": 6.5744,
      "theoretical_loss": 7.098646145157529,
      "tokens_seen": 8650752
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00013194444444444446,
      "loss": 6.659,
      "theoretical_loss": 7.088746148738453,
      "tokens_seen": 8716288
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00013293650793650793,
      "loss": 6.4896,
      "theoretical_loss": 7.0789409749245085,
      "tokens_seen": 8781824
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00013392857142857144,
      "loss": 6.4731,
      "theoretical_loss": 7.069229017967725,
      "tokens_seen": 8847360
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.0001349206349206349,
      "loss": 6.2886,
      "theoretical_loss": 7.059608710950327,
      "tokens_seen": 8912896
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.0001359126984126984,
      "loss": 6.4471,
      "theoretical_loss": 7.050078524568429,
      "tokens_seen": 8978432
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.0001369047619047619,
      "loss": 6.4571,
      "theoretical_loss": 7.040636965962385,
      "tokens_seen": 9043968
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00013789682539682541,
      "loss": 6.4965,
      "theoretical_loss": 7.031282577591693,
      "tokens_seen": 9109504
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.0001388888888888889,
      "loss": 6.1986,
      "theoretical_loss": 7.0220139361524385,
      "tokens_seen": 9175040
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.0001398809523809524,
      "loss": 6.4175,
      "theoretical_loss": 7.01282965153537,
      "tokens_seen": 9240576
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00014087301587301586,
      "loss": 6.6727,
      "theoretical_loss": 7.00372836582283,
      "tokens_seen": 9306112
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00014186507936507937,
      "loss": 6.588,
      "theoretical_loss": 6.994708752322817,
      "tokens_seen": 9371648
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00014285714285714284,
      "loss": 6.4597,
      "theoretical_loss": 6.985769514638539,
      "tokens_seen": 9437184
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00014384920634920634,
      "loss": 6.5805,
      "theoretical_loss": 6.976909385771949,
      "tokens_seen": 9502720
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00014484126984126984,
      "loss": 6.6703,
      "theoretical_loss": 6.968127127259775,
      "tokens_seen": 9568256
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00014583333333333335,
      "loss": 6.6715,
      "theoretical_loss": 6.959421528340654,
      "tokens_seen": 9633792
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00014682539682539685,
      "loss": 6.4198,
      "theoretical_loss": 6.950791405152069,
      "tokens_seen": 9699328
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00014781746031746032,
      "loss": 6.4149,
      "theoretical_loss": 6.942235599955788,
      "tokens_seen": 9764864
    },
    {
      "epoch": 0.0,
      "objective/train/advantage_avg": 0.49445855617523193,
      "objective/train/docs_used": 17861,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 5.295471668243408,
      "objective/train/original_loss": 5.295472145080566,
      "objective/train/theoretical_loss": 6.9337529803906595,
      "objective/train/tokens_used": 30290400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2445337325334549,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506888628005981,
      "objective/train/weighted_lm_loss": 5.56362771987915,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0451053380966187,
      "theoretical_loss": 6.9337529803906595,
      "tokens_seen": 9830400
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00014880952380952382,
      "loss": 5.9844,
      "theoretical_loss": 6.9337529803906595,
      "tokens_seen": 9830400
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.0001498015873015873,
      "loss": 6.3291,
      "theoretical_loss": 6.925342438751576,
      "tokens_seen": 9895936
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.0001507936507936508,
      "loss": 6.4475,
      "theoretical_loss": 6.917002891293558,
      "tokens_seen": 9961472
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00015178571428571427,
      "loss": 6.5167,
      "theoretical_loss": 6.908733277559907,
      "tokens_seen": 10027008
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.0001527777777777778,
      "loss": 6.3117,
      "theoretical_loss": 6.900532559733447,
      "tokens_seen": 10092544
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00015376984126984128,
      "loss": 6.1807,
      "theoretical_loss": 6.89239972200992,
      "tokens_seen": 10158080
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00015476190476190478,
      "loss": 5.9167,
      "theoretical_loss": 6.884333769992631,
      "tokens_seen": 10223616
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00015575396825396825,
      "loss": 6.3315,
      "theoretical_loss": 6.876333730107509,
      "tokens_seen": 10289152
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00015674603174603175,
      "loss": 6.4087,
      "theoretical_loss": 6.868398649037756,
      "tokens_seen": 10354688
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00015773809523809523,
      "loss": 6.2427,
      "theoretical_loss": 6.860527593177311,
      "tokens_seen": 10420224
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00015873015873015873,
      "loss": 6.1351,
      "theoretical_loss": 6.85271964810239,
      "tokens_seen": 10485760
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.0001597222222222222,
      "loss": 6.4149,
      "theoretical_loss": 6.844973918060402,
      "tokens_seen": 10551296
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00016071428571428573,
      "loss": 6.182,
      "theoretical_loss": 6.837289525475555,
      "tokens_seen": 10616832
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.0001617063492063492,
      "loss": 6.126,
      "theoretical_loss": 6.829665610470508,
      "tokens_seen": 10682368
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.0001626984126984127,
      "loss": 6.4237,
      "theoretical_loss": 6.822101330403479,
      "tokens_seen": 10747904
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00016369047619047618,
      "loss": 6.3283,
      "theoretical_loss": 6.814595859420155,
      "tokens_seen": 10813440
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00016468253968253969,
      "loss": 6.1484,
      "theoretical_loss": 6.807148388019941,
      "tokens_seen": 10878976
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00016567460317460316,
      "loss": 6.1467,
      "theoretical_loss": 6.799758122635906,
      "tokens_seen": 10944512
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00016666666666666666,
      "loss": 6.1517,
      "theoretical_loss": 6.792424285227984,
      "tokens_seen": 11010048
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00016765873015873016,
      "loss": 6.1389,
      "theoretical_loss": 6.785146112888908,
      "tokens_seen": 11075584
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00016865079365079366,
      "loss": 5.9013,
      "theoretical_loss": 6.777922857462398,
      "tokens_seen": 11141120
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00016964285714285717,
      "loss": 5.9434,
      "theoretical_loss": 6.7707537851731825,
      "tokens_seen": 11206656
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00017063492063492064,
      "loss": 6.0636,
      "theoretical_loss": 6.763638176268385,
      "tokens_seen": 11272192
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00017162698412698414,
      "loss": 6.1861,
      "theoretical_loss": 6.756575324669884,
      "tokens_seen": 11337728
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00017261904761904762,
      "loss": 6.283,
      "theoretical_loss": 6.749564537637246,
      "tokens_seen": 11403264
    },
    {
      "epoch": 0.0,
      "objective/train/advantage_avg": 0.49316972494125366,
      "objective/train/docs_used": 18493,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 6.047545909881592,
      "objective/train/original_loss": 6.047545433044434,
      "objective/train/theoretical_loss": 6.742605135440851,
      "objective/train/tokens_used": 31928800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24328872561454773,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050553560256958,
      "objective/train/weighted_lm_loss": 6.353062629699707,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0432155132293701,
      "theoretical_loss": 6.742605135440851,
      "tokens_seen": 11468800
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00017361111111111112,
      "loss": 6.1432,
      "theoretical_loss": 6.742605135440851,
      "tokens_seen": 11468800
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.0001746031746031746,
      "loss": 6.2202,
      "theoretical_loss": 6.735696451044834,
      "tokens_seen": 11534336
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.0001755952380952381,
      "loss": 6.1727,
      "theoretical_loss": 6.72883782979952,
      "tokens_seen": 11599872
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.0001765873015873016,
      "loss": 6.3492,
      "theoretical_loss": 6.7220286291429865,
      "tokens_seen": 11665408
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.0001775793650793651,
      "loss": 5.8535,
      "theoretical_loss": 6.715268218311449,
      "tokens_seen": 11730944
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00017857142857142857,
      "loss": 6.4073,
      "theoretical_loss": 6.708555978058172,
      "tokens_seen": 11796480
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00017956349206349207,
      "loss": 5.4498,
      "theoretical_loss": 6.701891300380584,
      "tokens_seen": 11862016
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00018055555555555555,
      "loss": 5.7002,
      "theoretical_loss": 6.6952735882553185,
      "tokens_seen": 11927552
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00018154761904761905,
      "loss": 6.1888,
      "theoretical_loss": 6.6887022553809405,
      "tokens_seen": 11993088
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00018253968253968252,
      "loss": 6.1907,
      "theoretical_loss": 6.682176725928034,
      "tokens_seen": 12058624
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00018353174603174602,
      "loss": 5.9728,
      "theoretical_loss": 6.675696434296475,
      "tokens_seen": 12124160
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00018452380952380953,
      "loss": 6.3165,
      "theoretical_loss": 6.669260824879574,
      "tokens_seen": 12189696
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00018551587301587303,
      "loss": 6.2443,
      "theoretical_loss": 6.662869351834917,
      "tokens_seen": 12255232
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.0001865079365079365,
      "loss": 6.1431,
      "theoretical_loss": 6.656521478861638,
      "tokens_seen": 12320768
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.0001875,
      "loss": 6.0853,
      "theoretical_loss": 6.650216678983931,
      "tokens_seen": 12386304
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.0001884920634920635,
      "loss": 6.0236,
      "theoretical_loss": 6.643954434340591,
      "tokens_seen": 12451840
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00018948412698412698,
      "loss": 6.2103,
      "theoretical_loss": 6.6377342359803855,
      "tokens_seen": 12517376
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00019047619047619048,
      "loss": 5.8085,
      "theoretical_loss": 6.631555583663063,
      "tokens_seen": 12582912
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00019146825396825398,
      "loss": 6.0944,
      "theoretical_loss": 6.625417985665809,
      "tokens_seen": 12648448
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00019246031746031748,
      "loss": 5.7959,
      "theoretical_loss": 6.619320958594983,
      "tokens_seen": 12713984
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00019345238095238096,
      "loss": 6.0324,
      "theoretical_loss": 6.6132640272029715,
      "tokens_seen": 12779520
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00019444444444444446,
      "loss": 5.8694,
      "theoretical_loss": 6.607246724209972,
      "tokens_seen": 12845056
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00019543650793650793,
      "loss": 5.9963,
      "theoretical_loss": 6.601268590130569,
      "tokens_seen": 12910592
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00019642857142857144,
      "loss": 5.9381,
      "theoretical_loss": 6.59532917310494,
      "tokens_seen": 12976128
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.0001974206349206349,
      "loss": 5.8007,
      "theoretical_loss": 6.589428028734545,
      "tokens_seen": 13041664
    },
    {
      "epoch": 0.0,
      "objective/train/advantage_avg": 0.4928921163082123,
      "objective/train/docs_used": 19731,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 5.5585432052612305,
      "objective/train/original_loss": 5.558542728424072,
      "objective/train/theoretical_loss": 6.583564719922174,
      "objective/train/tokens_used": 33567200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24302831292152405,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505244731903076,
      "objective/train/weighted_lm_loss": 5.8394341468811035,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0396549701690674,
      "theoretical_loss": 6.583564719922174,
      "tokens_seen": 13107200
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.0001984126984126984,
      "loss": 6.06,
      "theoretical_loss": 6.583564719922174,
      "tokens_seen": 13107200
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00019940476190476191,
      "loss": 6.2469,
      "theoretical_loss": 6.57773881671619,
      "tokens_seen": 13172736
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00020039682539682542,
      "loss": 6.0533,
      "theoretical_loss": 6.571949896158857,
      "tokens_seen": 13238272
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.0002013888888888889,
      "loss": 5.9509,
      "theoretical_loss": 6.566197542138623,
      "tokens_seen": 13303808
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.0002023809523809524,
      "loss": 6.1135,
      "theoretical_loss": 6.560481345246222,
      "tokens_seen": 13369344
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00020337301587301587,
      "loss": 5.9037,
      "theoretical_loss": 6.554800902634504,
      "tokens_seen": 13434880
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00020436507936507937,
      "loss": 5.9414,
      "theoretical_loss": 6.54915581788185,
      "tokens_seen": 13500416
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00020535714285714284,
      "loss": 6.1641,
      "theoretical_loss": 6.54354570085909,
      "tokens_seen": 13565952
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00020634920634920634,
      "loss": 6.1065,
      "theoretical_loss": 6.537970167599786,
      "tokens_seen": 13631488
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00020734126984126985,
      "loss": 5.9829,
      "theoretical_loss": 6.532428840173829,
      "tokens_seen": 13697024
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00020833333333333335,
      "loss": 5.8857,
      "theoretical_loss": 6.52692134656418,
      "tokens_seen": 13762560
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00020932539682539685,
      "loss": 5.9751,
      "theoretical_loss": 6.521447320546738,
      "tokens_seen": 13828096
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00021031746031746032,
      "loss": 6.1487,
      "theoretical_loss": 6.516006401573176,
      "tokens_seen": 13893632
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00021130952380952382,
      "loss": 6.0326,
      "theoretical_loss": 6.51059823465671,
      "tokens_seen": 13959168
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.0002123015873015873,
      "loss": 6.1904,
      "theoretical_loss": 6.505222470260669,
      "tokens_seen": 14024704
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.0002132936507936508,
      "loss": 5.8897,
      "theoretical_loss": 6.499878764189829,
      "tokens_seen": 14090240
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00021428571428571427,
      "loss": 6.0129,
      "theoretical_loss": 6.494566777484378,
      "tokens_seen": 14155776
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.0002152777777777778,
      "loss": 6.2993,
      "theoretical_loss": 6.489286176316504,
      "tokens_seen": 14221312
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00021626984126984128,
      "loss": 5.9129,
      "theoretical_loss": 6.48403663188946,
      "tokens_seen": 14286848
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00021726190476190478,
      "loss": 5.5438,
      "theoretical_loss": 6.478817820339071,
      "tokens_seen": 14352384
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00021825396825396825,
      "loss": 5.9031,
      "theoretical_loss": 6.473629422637629,
      "tokens_seen": 14417920
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00021924603174603176,
      "loss": 5.8432,
      "theoretical_loss": 6.468471124500068,
      "tokens_seen": 14483456
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00022023809523809523,
      "loss": 5.7172,
      "theoretical_loss": 6.463342616292376,
      "tokens_seen": 14548992
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00022123015873015873,
      "loss": 5.9131,
      "theoretical_loss": 6.458243592942189,
      "tokens_seen": 14614528
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.0002222222222222222,
      "loss": 5.8095,
      "theoretical_loss": 6.453173753851489,
      "tokens_seen": 14680064
    },
    {
      "epoch": 0.0,
      "objective/train/advantage_avg": 0.49456000328063965,
      "objective/train/docs_used": 20428,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 5.2190093994140625,
      "objective/train/original_loss": 5.219009876251221,
      "objective/train/theoretical_loss": 6.448132802811353,
      "objective/train/tokens_used": 35205600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24466407299041748,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506995916366577,
      "objective/train/weighted_lm_loss": 5.483295440673828,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0417574644088745,
      "theoretical_loss": 6.448132802811353,
      "tokens_seen": 14745600
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00022321428571428573,
      "loss": 5.6392,
      "theoretical_loss": 6.448132802811353,
      "tokens_seen": 14745600
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.0002242063492063492,
      "loss": 5.9487,
      "theoretical_loss": 6.443120447918698,
      "tokens_seen": 14811136
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.0002251984126984127,
      "loss": 6.1013,
      "theoretical_loss": 6.438136401494974,
      "tokens_seen": 14876672
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00022619047619047618,
      "loss": 5.6871,
      "theoretical_loss": 6.433180380006723,
      "tokens_seen": 14942208
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00022718253968253969,
      "loss": 6.1918,
      "theoretical_loss": 6.428252103987999,
      "tokens_seen": 15007744
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00022817460317460316,
      "loss": 5.9442,
      "theoretical_loss": 6.423351297964546,
      "tokens_seen": 15073280
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00022916666666666666,
      "loss": 5.8063,
      "theoretical_loss": 6.41847769037973,
      "tokens_seen": 15138816
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00023015873015873016,
      "loss": 5.6459,
      "theoretical_loss": 6.413631013522151,
      "tokens_seen": 15204352
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00023115079365079367,
      "loss": 5.7214,
      "theoretical_loss": 6.4088110034549,
      "tokens_seen": 15269888
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00023214285714285717,
      "loss": 5.982,
      "theoretical_loss": 6.404017399946415,
      "tokens_seen": 15335424
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00023313492063492064,
      "loss": 6.1643,
      "theoretical_loss": 6.399249946402892,
      "tokens_seen": 15400960
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00023412698412698414,
      "loss": 5.7842,
      "theoretical_loss": 6.394508389802212,
      "tokens_seen": 15466496
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00023511904761904762,
      "loss": 5.7139,
      "theoretical_loss": 6.3897924806293345,
      "tokens_seen": 15532032
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00023611111111111112,
      "loss": 5.7352,
      "theoretical_loss": 6.385101972813146,
      "tokens_seen": 15597568
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.0002371031746031746,
      "loss": 5.687,
      "theoretical_loss": 6.380436623664683,
      "tokens_seen": 15663104
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.0002380952380952381,
      "loss": 5.7728,
      "theoretical_loss": 6.3757961938167265,
      "tokens_seen": 15728640
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.0002390873015873016,
      "loss": 5.9507,
      "theoretical_loss": 6.3711804471647255,
      "tokens_seen": 15794176
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.0002400793650793651,
      "loss": 5.6308,
      "theoretical_loss": 6.366589150809007,
      "tokens_seen": 15859712
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00024107142857142857,
      "loss": 5.7513,
      "theoretical_loss": 6.36202207499824,
      "tokens_seen": 15925248
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00024206349206349207,
      "loss": 5.8209,
      "theoretical_loss": 6.357478993074131,
      "tokens_seen": 15990784
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00024305555555555555,
      "loss": 5.5718,
      "theoretical_loss": 6.352959681417312,
      "tokens_seen": 16056320
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00024404761904761905,
      "loss": 5.5157,
      "theoretical_loss": 6.348463919394389,
      "tokens_seen": 16121856
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00024503968253968255,
      "loss": 6.0747,
      "theoretical_loss": 6.343991489306125,
      "tokens_seen": 16187392
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.000246031746031746,
      "loss": 5.9865,
      "theoretical_loss": 6.33954217633673,
      "tokens_seen": 16252928
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.00024702380952380955,
      "loss": 6.0123,
      "theoretical_loss": 6.33511576850422,
      "tokens_seen": 16318464
    },
    {
      "epoch": 0.0,
      "objective/train/advantage_avg": 0.4939838945865631,
      "objective/train/docs_used": 21641,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 4.999059200286865,
      "objective/train/original_loss": 4.999059677124023,
      "objective/train/theoretical_loss": 6.330712056611843,
      "objective/train/tokens_used": 36844000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24411740899085999,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050639271736145,
      "objective/train/weighted_lm_loss": 5.251715183258057,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 6.330712056611843,
      "tokens_seen": 16384000
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.000248015873015873,
      "loss": 5.8227,
      "theoretical_loss": 6.330712056611843,
      "tokens_seen": 16384000
    },
    {
      "epoch": 0.0,
      "learning_rate": 0.0002490079365079365,
      "loss": 5.9736,
      "theoretical_loss": 6.32633083420051,
      "tokens_seen": 16449536
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.00025,
      "loss": 5.6796,
      "theoretical_loss": 6.32197189750223,
      "tokens_seen": 16515072
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0002509920634920635,
      "loss": 5.415,
      "theoretical_loss": 6.317635045394523,
      "tokens_seen": 16580608
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.000251984126984127,
      "loss": 5.7857,
      "theoretical_loss": 6.3133200793557815,
      "tokens_seen": 16646144
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.00025297619047619046,
      "loss": 5.3164,
      "theoretical_loss": 6.3090268034215455,
      "tokens_seen": 16711680
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.00025396825396825396,
      "loss": 5.7244,
      "theoretical_loss": 6.304755024141699,
      "tokens_seen": 16777216
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.00025496031746031746,
      "loss": 5.8446,
      "theoretical_loss": 6.300504550538534,
      "tokens_seen": 16842752
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.00025595238095238096,
      "loss": 5.7314,
      "theoretical_loss": 6.296275194065678,
      "tokens_seen": 16908288
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0002569444444444444,
      "loss": 5.7309,
      "theoretical_loss": 6.292066768567858,
      "tokens_seen": 16973824
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.00025793650793650796,
      "loss": 5.3834,
      "theoretical_loss": 6.28787909024149,
      "tokens_seen": 17039360
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.00025892857142857146,
      "loss": 5.6599,
      "theoretical_loss": 6.283711977596056,
      "tokens_seen": 17104896
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.00025992063492063497,
      "loss": 5.7938,
      "theoretical_loss": 6.279565251416267,
      "tokens_seen": 17170432
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0002609126984126984,
      "loss": 5.4578,
      "theoretical_loss": 6.275438734724988,
      "tokens_seen": 17235968
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0002619047619047619,
      "loss": 5.829,
      "theoretical_loss": 6.271332252746902,
      "tokens_seen": 17301504
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0002628968253968254,
      "loss": 5.7369,
      "theoretical_loss": 6.267245632872896,
      "tokens_seen": 17367040
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0002638888888888889,
      "loss": 5.7287,
      "theoretical_loss": 6.263178704625169,
      "tokens_seen": 17432576
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.00026488095238095237,
      "loss": 5.797,
      "theoretical_loss": 6.2591312996230055,
      "tokens_seen": 17498112
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.00026587301587301587,
      "loss": 5.4914,
      "theoretical_loss": 6.255103251549257,
      "tokens_seen": 17563648
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.00026686507936507937,
      "loss": 5.8389,
      "theoretical_loss": 6.251094396117449,
      "tokens_seen": 17629184
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.00026785714285714287,
      "loss": 5.7267,
      "theoretical_loss": 6.247104571039561,
      "tokens_seen": 17694720
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0002688492063492063,
      "loss": 5.6183,
      "theoretical_loss": 6.243133615994411,
      "tokens_seen": 17760256
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0002698412698412698,
      "loss": 5.7327,
      "theoretical_loss": 6.239181372596676,
      "tokens_seen": 17825792
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0002708333333333333,
      "loss": 5.6056,
      "theoretical_loss": 6.235247684366492,
      "tokens_seen": 17891328
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0002718253968253968,
      "loss": 5.6439,
      "theoretical_loss": 6.231332396699658,
      "tokens_seen": 17956864
    },
    {
      "epoch": 0.01,
      "objective/train/advantage_avg": 0.4953996241092682,
      "objective/train/docs_used": 22253,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 5.136692047119141,
      "objective/train/original_loss": 5.136691570281982,
      "objective/train/theoretical_loss": 6.227435356838402,
      "objective/train/tokens_used": 38482400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2454858124256134,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507879257202148,
      "objective/train/weighted_lm_loss": 5.397404670715332,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 6.227435356838402,
      "tokens_seen": 18022400
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0002728174603174603,
      "loss": 5.8516,
      "theoretical_loss": 6.227435356838402,
      "tokens_seen": 18022400
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0002738095238095238,
      "loss": 5.5735,
      "theoretical_loss": 6.223556413842715,
      "tokens_seen": 18087936
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0002748015873015873,
      "loss": 5.7966,
      "theoretical_loss": 6.219695418562217,
      "tokens_seen": 18153472
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.00027579365079365083,
      "loss": 5.592,
      "theoretical_loss": 6.215852223608581,
      "tokens_seen": 18219008
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.00027678571428571433,
      "loss": 5.7162,
      "theoretical_loss": 6.212026683328462,
      "tokens_seen": 18284544
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0002777777777777778,
      "loss": 5.277,
      "theoretical_loss": 6.208218653776937,
      "tokens_seen": 18350080
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0002787698412698413,
      "loss": 5.6509,
      "theoretical_loss": 6.2044279926914605,
      "tokens_seen": 18415616
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0002797619047619048,
      "loss": 5.7304,
      "theoretical_loss": 6.20065455946629,
      "tokens_seen": 18481152
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0002807539682539683,
      "loss": 5.7087,
      "theoretical_loss": 6.1968982151273995,
      "tokens_seen": 18546688
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.00028174603174603173,
      "loss": 5.5229,
      "theoretical_loss": 6.193158822307858,
      "tokens_seen": 18612224
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.00028273809523809523,
      "loss": 5.1858,
      "theoretical_loss": 6.1894362452236615,
      "tokens_seen": 18677760
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.00028373015873015873,
      "loss": 5.4294,
      "theoretical_loss": 6.185730349650021,
      "tokens_seen": 18743296
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.00028472222222222223,
      "loss": 5.5771,
      "theoretical_loss": 6.182041002898073,
      "tokens_seen": 18808832
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0002857142857142857,
      "loss": 5.5212,
      "theoretical_loss": 6.17836807379203,
      "tokens_seen": 18874368
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0002867063492063492,
      "loss": 5.5862,
      "theoretical_loss": 6.174711432646743,
      "tokens_seen": 18939904
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0002876984126984127,
      "loss": 5.3196,
      "theoretical_loss": 6.171070951245673,
      "tokens_seen": 19005440
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0002886904761904762,
      "loss": 5.5696,
      "theoretical_loss": 6.167446502819269,
      "tokens_seen": 19070976
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0002896825396825397,
      "loss": 5.5004,
      "theoretical_loss": 6.163837962023724,
      "tokens_seen": 19136512
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0002906746031746032,
      "loss": 5.9357,
      "theoretical_loss": 6.160245204920132,
      "tokens_seen": 19202048
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0002916666666666667,
      "loss": 5.5112,
      "theoretical_loss": 6.156668108954003,
      "tokens_seen": 19267584
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0002926587301587302,
      "loss": 5.6058,
      "theoretical_loss": 6.1531065529351565,
      "tokens_seen": 19333120
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0002936507936507937,
      "loss": 5.852,
      "theoretical_loss": 6.149560417017971,
      "tokens_seen": 19398656
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.00029464285714285714,
      "loss": 5.4236,
      "theoretical_loss": 6.146029582681981,
      "tokens_seen": 19464192
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.00029563492063492064,
      "loss": 5.6936,
      "theoretical_loss": 6.1425139327128235,
      "tokens_seen": 19529728
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.00029662698412698414,
      "loss": 5.5577,
      "theoretical_loss": 6.139013351183521,
      "tokens_seen": 19595264
    },
    {
      "epoch": 0.01,
      "objective/train/advantage_avg": 0.4941224157810211,
      "objective/train/docs_used": 23612,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 5.7014384269714355,
      "objective/train/original_loss": 5.701438903808594,
      "objective/train/theoretical_loss": 6.135527723436086,
      "objective/train/tokens_used": 40120800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24421964585781097,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506536960601807,
      "objective/train/weighted_lm_loss": 5.99001407623291,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0451053380966187,
      "theoretical_loss": 6.135527723436086,
      "tokens_seen": 19660800
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.00029761904761904765,
      "loss": 5.5959,
      "theoretical_loss": 6.135527723436086,
      "tokens_seen": 19660800
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0002986111111111111,
      "loss": 5.5736,
      "theoretical_loss": 6.132056936063465,
      "tokens_seen": 19726336
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0002996031746031746,
      "loss": 5.5708,
      "theoretical_loss": 6.128600876891781,
      "tokens_seen": 19791872
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0003005952380952381,
      "loss": 5.5428,
      "theoretical_loss": 6.125159434962898,
      "tokens_seen": 19857408
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0003015873015873016,
      "loss": 5.6828,
      "theoretical_loss": 6.121732500517281,
      "tokens_seen": 19922944
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.00030257936507936505,
      "loss": 5.5381,
      "theoretical_loss": 6.118319964977157,
      "tokens_seen": 19988480
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.00030357142857142855,
      "loss": 5.646,
      "theoretical_loss": 6.114921720929972,
      "tokens_seen": 20054016
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.00030456349206349205,
      "loss": 5.5318,
      "theoretical_loss": 6.111537662112119,
      "tokens_seen": 20119552
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0003055555555555556,
      "loss": 5.366,
      "theoretical_loss": 6.1081676833929555,
      "tokens_seen": 20185088
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.00030654761904761905,
      "loss": 5.1976,
      "theoretical_loss": 6.104811680759095,
      "tokens_seen": 20250624
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.00030753968253968255,
      "loss": 5.5778,
      "theoretical_loss": 6.101469551298964,
      "tokens_seen": 20316160
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.00030853174603174605,
      "loss": 5.6324,
      "theoretical_loss": 6.098141193187615,
      "tokens_seen": 20381696
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.00030952380952380956,
      "loss": 5.4753,
      "theoretical_loss": 6.094826505671813,
      "tokens_seen": 20447232
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.000310515873015873,
      "loss": 5.4323,
      "theoretical_loss": 6.091525389055353,
      "tokens_seen": 20512768
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0003115079365079365,
      "loss": 5.5334,
      "theoretical_loss": 6.088237744684644,
      "tokens_seen": 20578304
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0003125,
      "loss": 5.659,
      "theoretical_loss": 6.084963474934522,
      "tokens_seen": 20643840
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0003134920634920635,
      "loss": 5.4652,
      "theoretical_loss": 6.081702483194306,
      "tokens_seen": 20709376
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.000314484126984127,
      "loss": 5.3359,
      "theoretical_loss": 6.078454673854084,
      "tokens_seen": 20774912
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.00031547619047619046,
      "loss": 5.5398,
      "theoretical_loss": 6.075219952291231,
      "tokens_seen": 20840448
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.00031646825396825396,
      "loss": 5.5016,
      "theoretical_loss": 6.071998224857146,
      "tokens_seen": 20905984
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.00031746031746031746,
      "loss": 5.7141,
      "theoretical_loss": 6.068789398864208,
      "tokens_seen": 20971520
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.00031845238095238096,
      "loss": 5.3938,
      "theoretical_loss": 6.065593382572951,
      "tokens_seen": 21037056
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0003194444444444444,
      "loss": 5.5114,
      "theoretical_loss": 6.062410085179447,
      "tokens_seen": 21102592
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.00032043650793650796,
      "loss": 5.3493,
      "theoretical_loss": 6.059239416802894,
      "tokens_seen": 21168128
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.00032142857142857147,
      "loss": 5.6172,
      "theoretical_loss": 6.0560812884734085,
      "tokens_seen": 21233664
    },
    {
      "epoch": 0.01,
      "objective/train/advantage_avg": 0.4957156181335449,
      "objective/train/docs_used": 24330,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 5.916872978210449,
      "objective/train/original_loss": 5.916873455047607,
      "objective/train/theoretical_loss": 6.052935612120018,
      "objective/train/tokens_used": 41759200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24579912424087524,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.05082106590271,
      "objective/train/weighted_lm_loss": 6.21736478805542,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 6.052935612120018,
      "tokens_seen": 21299200
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.00032242063492063497,
      "loss": 5.6873,
      "theoretical_loss": 6.052935612120018,
      "tokens_seen": 21299200
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0003234126984126984,
      "loss": 5.4965,
      "theoretical_loss": 6.049802300558843,
      "tokens_seen": 21364736
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0003244047619047619,
      "loss": 5.4604,
      "theoretical_loss": 6.046681267481475,
      "tokens_seen": 21430272
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0003253968253968254,
      "loss": 5.5719,
      "theoretical_loss": 6.043572427443539,
      "tokens_seen": 21495808
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0003263888888888889,
      "loss": 5.5019,
      "theoretical_loss": 6.040475695853447,
      "tokens_seen": 21561344
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.00032738095238095237,
      "loss": 5.3761,
      "theoretical_loss": 6.037390988961312,
      "tokens_seen": 21626880
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.00032837301587301587,
      "loss": 5.7069,
      "theoretical_loss": 6.034318223848068,
      "tokens_seen": 21692416
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.00032936507936507937,
      "loss": 5.6658,
      "theoretical_loss": 6.031257318414738,
      "tokens_seen": 21757952
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.00033035714285714287,
      "loss": 5.6025,
      "theoretical_loss": 6.028208191371885,
      "tokens_seen": 21823488
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0003313492063492063,
      "loss": 5.5051,
      "theoretical_loss": 6.025170762229224,
      "tokens_seen": 21889024
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0003323412698412698,
      "loss": 5.3105,
      "theoretical_loss": 6.022144951285407,
      "tokens_seen": 21954560
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0003333333333333333,
      "loss": 5.3062,
      "theoretical_loss": 6.019130679617959,
      "tokens_seen": 22020096
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0003343253968253968,
      "loss": 5.4437,
      "theoretical_loss": 6.016127869073374,
      "tokens_seen": 22085632
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0003353174603174603,
      "loss": 5.5389,
      "theoretical_loss": 6.01313644225737,
      "tokens_seen": 22151168
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0003363095238095238,
      "loss": 5.6526,
      "theoretical_loss": 6.010156322525292,
      "tokens_seen": 22216704
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.00033730158730158733,
      "loss": 5.4174,
      "theoretical_loss": 6.007187433972668,
      "tokens_seen": 22282240
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.00033829365079365083,
      "loss": 5.2724,
      "theoretical_loss": 6.004229701425903,
      "tokens_seen": 22347776
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.00033928571428571433,
      "loss": 5.6003,
      "theoretical_loss": 6.001283050433128,
      "tokens_seen": 22413312
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0003402777777777778,
      "loss": 5.3554,
      "theoretical_loss": 5.998347407255185,
      "tokens_seen": 22478848
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0003412698412698413,
      "loss": 5.3899,
      "theoretical_loss": 5.995422698856746,
      "tokens_seen": 22544384
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0003422619047619048,
      "loss": 5.4873,
      "theoretical_loss": 5.992508852897577,
      "tokens_seen": 22609920
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0003432539682539683,
      "loss": 5.238,
      "theoretical_loss": 5.989605797723927,
      "tokens_seen": 22675456
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.00034424603174603173,
      "loss": 5.3523,
      "theoretical_loss": 5.986713462360052,
      "tokens_seen": 22740992
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.00034523809523809523,
      "loss": 5.2717,
      "theoretical_loss": 5.983831776499871,
      "tokens_seen": 22806528
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.00034623015873015873,
      "loss": 5.3051,
      "theoretical_loss": 5.980960670498742,
      "tokens_seen": 22872064
    },
    {
      "epoch": 0.01,
      "objective/train/advantage_avg": 0.49309515953063965,
      "objective/train/docs_used": 24983,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 5.524625778198242,
      "objective/train/original_loss": 5.5246262550354,
      "objective/train/theoretical_loss": 5.978100075365368,
      "objective/train/tokens_used": 43397600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24325785040855408,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505459308624268,
      "objective/train/weighted_lm_loss": 5.803831100463867,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0352214574813843,
      "theoretical_loss": 5.978100075365368,
      "tokens_seen": 22937600
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.00034722222222222224,
      "loss": 5.473,
      "theoretical_loss": 5.978100075365368,
      "tokens_seen": 22937600
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0003482142857142857,
      "loss": 5.411,
      "theoretical_loss": 5.975249922753816,
      "tokens_seen": 23003136
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0003492063492063492,
      "loss": 5.4239,
      "theoretical_loss": 5.972410144955672,
      "tokens_seen": 23068672
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0003501984126984127,
      "loss": 5.5323,
      "theoretical_loss": 5.969580674892294,
      "tokens_seen": 23134208
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0003511904761904762,
      "loss": 5.2138,
      "theoretical_loss": 5.966761446107192,
      "tokens_seen": 23199744
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0003521825396825397,
      "loss": 5.224,
      "theoretical_loss": 5.963952392758523,
      "tokens_seen": 23265280
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0003531746031746032,
      "loss": 5.0797,
      "theoretical_loss": 5.961153449611697,
      "tokens_seen": 23330816
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0003541666666666667,
      "loss": 5.1792,
      "theoretical_loss": 5.958364552032075,
      "tokens_seen": 23396352
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0003551587301587302,
      "loss": 5.511,
      "theoretical_loss": 5.955585635977804,
      "tokens_seen": 23461888
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0003561507936507937,
      "loss": 5.283,
      "theoretical_loss": 5.952816637992738,
      "tokens_seen": 23527424
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.00035714285714285714,
      "loss": 5.4284,
      "theoretical_loss": 5.950057495199465,
      "tokens_seen": 23592960
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.00035813492063492064,
      "loss": 5.1736,
      "theoretical_loss": 5.947308145292439,
      "tokens_seen": 23658496
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.00035912698412698415,
      "loss": 5.5475,
      "theoretical_loss": 5.944568526531218,
      "tokens_seen": 23724032
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.00036011904761904765,
      "loss": 5.4948,
      "theoretical_loss": 5.941838577733787,
      "tokens_seen": 23789568
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0003611111111111111,
      "loss": 5.3477,
      "theoretical_loss": 5.939118238269988,
      "tokens_seen": 23855104
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0003621031746031746,
      "loss": 5.5783,
      "theoretical_loss": 5.936407448055045,
      "tokens_seen": 23920640
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0003630952380952381,
      "loss": 5.2567,
      "theoretical_loss": 5.933706147543174,
      "tokens_seen": 23986176
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0003640873015873016,
      "loss": 5.4058,
      "theoretical_loss": 5.9310142777212995,
      "tokens_seen": 24051712
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.00036507936507936505,
      "loss": 5.5087,
      "theoretical_loss": 5.928331780102846,
      "tokens_seen": 24117248
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.00036607142857142855,
      "loss": 5.4261,
      "theoretical_loss": 5.925658596721627,
      "tokens_seen": 24182784
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.00036706349206349205,
      "loss": 5.5392,
      "theoretical_loss": 5.922994670125819,
      "tokens_seen": 24248320
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0003680555555555556,
      "loss": 5.3145,
      "theoretical_loss": 5.920339943372026,
      "tokens_seen": 24313856
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.00036904761904761905,
      "loss": 5.3914,
      "theoretical_loss": 5.917694360019418,
      "tokens_seen": 24379392
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.00037003968253968255,
      "loss": 5.3178,
      "theoretical_loss": 5.915057864123962,
      "tokens_seen": 24444928
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.00037103174603174606,
      "loss": 5.4064,
      "theoretical_loss": 5.912430400232731,
      "tokens_seen": 24510464
    },
    {
      "epoch": 0.01,
      "objective/train/advantage_avg": 0.49378445744514465,
      "objective/train/docs_used": 26439,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 4.7033820152282715,
      "objective/train/original_loss": 4.7033820152282715,
      "objective/train/theoretical_loss": 5.909811913378293,
      "objective/train/tokens_used": 45036000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24392488598823547,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506185293197632,
      "objective/train/weighted_lm_loss": 4.941839218139648,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0363596677780151,
      "theoretical_loss": 5.909811913378293,
      "tokens_seen": 24576000
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.00037202380952380956,
      "loss": 5.0631,
      "theoretical_loss": 5.909811913378293,
      "tokens_seen": 24576000
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.000373015873015873,
      "loss": 5.361,
      "theoretical_loss": 5.907202349073183,
      "tokens_seen": 24641536
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0003740079365079365,
      "loss": 5.1099,
      "theoretical_loss": 5.904601653304437,
      "tokens_seen": 24707072
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.000375,
      "loss": 5.2016,
      "theoretical_loss": 5.902009772528231,
      "tokens_seen": 24772608
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0003759920634920635,
      "loss": 5.5713,
      "theoretical_loss": 5.899426653664568,
      "tokens_seen": 24838144
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.000376984126984127,
      "loss": 5.3826,
      "theoretical_loss": 5.896852244092045,
      "tokens_seen": 24903680
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.00037797619047619046,
      "loss": 5.372,
      "theoretical_loss": 5.894286491642708,
      "tokens_seen": 24969216
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.00037896825396825396,
      "loss": 5.3521,
      "theoretical_loss": 5.891729344596959,
      "tokens_seen": 25034752
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.00037996031746031746,
      "loss": 5.2535,
      "theoretical_loss": 5.8891807516785395,
      "tokens_seen": 25100288
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.00038095238095238096,
      "loss": 5.5121,
      "theoretical_loss": 5.886640662049593,
      "tokens_seen": 25165824
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0003819444444444444,
      "loss": 5.2922,
      "theoretical_loss": 5.884109025305776,
      "tokens_seen": 25231360
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.00038293650793650797,
      "loss": 5.1638,
      "theoretical_loss": 5.881585791471453,
      "tokens_seen": 25296896
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.00038392857142857147,
      "loss": 5.5825,
      "theoretical_loss": 5.879070910994954,
      "tokens_seen": 25362432
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.00038492063492063497,
      "loss": 5.1319,
      "theoretical_loss": 5.876564334743884,
      "tokens_seen": 25427968
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0003859126984126984,
      "loss": 5.5137,
      "theoretical_loss": 5.874066014000512,
      "tokens_seen": 25493504
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0003869047619047619,
      "loss": 5.3543,
      "theoretical_loss": 5.871575900457216,
      "tokens_seen": 25559040
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0003878968253968254,
      "loss": 5.1918,
      "theoretical_loss": 5.86909394621199,
      "tokens_seen": 25624576
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0003888888888888889,
      "loss": 5.2748,
      "theoretical_loss": 5.8666201037640064,
      "tokens_seen": 25690112
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.00038988095238095237,
      "loss": 5.4944,
      "theoretical_loss": 5.864154326009246,
      "tokens_seen": 25755648
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.00039087301587301587,
      "loss": 5.1782,
      "theoretical_loss": 5.861696566236189,
      "tokens_seen": 25821184
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.00039186507936507937,
      "loss": 5.6189,
      "theoretical_loss": 5.859246778121547,
      "tokens_seen": 25886720
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0003928571428571429,
      "loss": 5.5505,
      "theoretical_loss": 5.856804915726074,
      "tokens_seen": 25952256
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0003938492063492063,
      "loss": 5.344,
      "theoretical_loss": 5.854370933490414,
      "tokens_seen": 26017792
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0003948412698412698,
      "loss": 5.2611,
      "theoretical_loss": 5.851944786231018,
      "tokens_seen": 26083328
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0003958333333333333,
      "loss": 5.3355,
      "theoretical_loss": 5.849526429136105,
      "tokens_seen": 26148864
    },
    {
      "epoch": 0.01,
      "objective/train/advantage_avg": 0.4930327534675598,
      "objective/train/docs_used": 27060,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 5.081445217132568,
      "objective/train/original_loss": 5.08144474029541,
      "objective/train/theoretical_loss": 5.847115817761683,
      "objective/train/tokens_used": 46674400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24318242073059082,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505393743515015,
      "objective/train/weighted_lm_loss": 5.338102340698242,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0363596677780151,
      "theoretical_loss": 5.847115817761683,
      "tokens_seen": 26214400
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0003968253968253968,
      "loss": 5.102,
      "theoretical_loss": 5.847115817761683,
      "tokens_seen": 26214400
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0003978174603174603,
      "loss": 5.1532,
      "theoretical_loss": 5.844712908027621,
      "tokens_seen": 26279936
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.00039880952380952383,
      "loss": 5.3383,
      "theoretical_loss": 5.842317656213765,
      "tokens_seen": 26345472
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.00039980158730158733,
      "loss": 5.5407,
      "theoretical_loss": 5.83993001895612,
      "tokens_seen": 26411008
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.00040079365079365083,
      "loss": 5.1595,
      "theoretical_loss": 5.83754995324307,
      "tokens_seen": 26476544
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.00040178571428571433,
      "loss": 5.2222,
      "theoretical_loss": 5.835177416411649,
      "tokens_seen": 26542080
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004027777777777778,
      "loss": 5.2372,
      "theoretical_loss": 5.8328123661438624,
      "tokens_seen": 26607616
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004037698412698413,
      "loss": 5.4145,
      "theoretical_loss": 5.830454760463059,
      "tokens_seen": 26673152
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004047619047619048,
      "loss": 5.1263,
      "theoretical_loss": 5.828104557730348,
      "tokens_seen": 26738688
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004057539682539683,
      "loss": 5.1369,
      "theoretical_loss": 5.825761716641054,
      "tokens_seen": 26804224
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.00040674603174603173,
      "loss": 5.1114,
      "theoretical_loss": 5.823426196221233,
      "tokens_seen": 26869760
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.00040773809523809523,
      "loss": 5.3269,
      "theoretical_loss": 5.821097955824221,
      "tokens_seen": 26935296
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.00040873015873015874,
      "loss": 5.1287,
      "theoretical_loss": 5.818776955127233,
      "tokens_seen": 27000832
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.00040972222222222224,
      "loss": 5.4583,
      "theoretical_loss": 5.816463154128003,
      "tokens_seen": 27066368
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004107142857142857,
      "loss": 5.0305,
      "theoretical_loss": 5.814156513141467,
      "tokens_seen": 27131904
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004117063492063492,
      "loss": 5.3215,
      "theoretical_loss": 5.811856992796489,
      "tokens_seen": 27197440
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004126984126984127,
      "loss": 5.4324,
      "theoretical_loss": 5.809564554032628,
      "tokens_seen": 27262976
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004136904761904762,
      "loss": 5.1104,
      "theoretical_loss": 5.80727915809695,
      "tokens_seen": 27328512
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004146825396825397,
      "loss": 5.254,
      "theoretical_loss": 5.805000766540871,
      "tokens_seen": 27394048
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004156746031746032,
      "loss": 5.1357,
      "theoretical_loss": 5.802729341217047,
      "tokens_seen": 27459584
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004166666666666667,
      "loss": 5.4149,
      "theoretical_loss": 5.800464844276306,
      "tokens_seen": 27525120
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004176587301587302,
      "loss": 5.0521,
      "theoretical_loss": 5.798207238164607,
      "tokens_seen": 27590656
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004186507936507937,
      "loss": 4.9521,
      "theoretical_loss": 5.795956485620051,
      "tokens_seen": 27656192
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.00041964285714285714,
      "loss": 5.2116,
      "theoretical_loss": 5.793712549669918,
      "tokens_seen": 27721728
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.00042063492063492065,
      "loss": 5.1607,
      "theoretical_loss": 5.791475393627746,
      "tokens_seen": 27787264
    },
    {
      "epoch": 0.01,
      "objective/train/advantage_avg": 0.49599596858024597,
      "objective/train/docs_used": 28453,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 4.815020561218262,
      "objective/train/original_loss": 4.8150200843811035,
      "objective/train/theoretical_loss": 5.789244981090448,
      "objective/train/tokens_used": 48312800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2460610270500183,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050850510597229,
      "objective/train/weighted_lm_loss": 5.0598039627075195,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 5.789244981090448,
      "tokens_seen": 27852800
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.00042162698412698415,
      "loss": 5.0717,
      "theoretical_loss": 5.789244981090448,
      "tokens_seen": 27852800
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.00042261904761904765,
      "loss": 5.4444,
      "theoretical_loss": 5.7870212759354605,
      "tokens_seen": 27918336
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004236111111111111,
      "loss": 5.4988,
      "theoretical_loss": 5.7848042423179304,
      "tokens_seen": 27983872
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004246031746031746,
      "loss": 4.834,
      "theoretical_loss": 5.782593844667936,
      "tokens_seen": 28049408
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004255952380952381,
      "loss": 5.1191,
      "theoretical_loss": 5.780390047687737,
      "tokens_seen": 28114944
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004265873015873016,
      "loss": 4.8911,
      "theoretical_loss": 5.778192816349071,
      "tokens_seen": 28180480
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.00042757936507936505,
      "loss": 5.1284,
      "theoretical_loss": 5.776002115890471,
      "tokens_seen": 28246016
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.00042857142857142855,
      "loss": 5.3885,
      "theoretical_loss": 5.773817911814618,
      "tokens_seen": 28311552
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.00042956349206349205,
      "loss": 4.8577,
      "theoretical_loss": 5.771640169885731,
      "tokens_seen": 28377088
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004305555555555556,
      "loss": 5.185,
      "theoretical_loss": 5.769468856126985,
      "tokens_seen": 28442624
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.00043154761904761905,
      "loss": 5.2508,
      "theoretical_loss": 5.76730393681796,
      "tokens_seen": 28508160
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.00043253968253968256,
      "loss": 4.8913,
      "theoretical_loss": 5.765145378492123,
      "tokens_seen": 28573696
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.00043353174603174606,
      "loss": 5.3771,
      "theoretical_loss": 5.762993147934338,
      "tokens_seen": 28639232
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.00043452380952380956,
      "loss": 4.9848,
      "theoretical_loss": 5.76084721217841,
      "tokens_seen": 28704768
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.000435515873015873,
      "loss": 5.4856,
      "theoretical_loss": 5.7587075385046536,
      "tokens_seen": 28770304
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004365079365079365,
      "loss": 5.3679,
      "theoretical_loss": 5.756574094437491,
      "tokens_seen": 28835840
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004375,
      "loss": 5.418,
      "theoretical_loss": 5.75444684774309,
      "tokens_seen": 28901376
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004384920634920635,
      "loss": 4.8353,
      "theoretical_loss": 5.752325766427008,
      "tokens_seen": 28966912
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.000439484126984127,
      "loss": 5.4687,
      "theoretical_loss": 5.750210818731889,
      "tokens_seen": 29032448
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.00044047619047619046,
      "loss": 5.6064,
      "theoretical_loss": 5.748101973135171,
      "tokens_seen": 29097984
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.00044146825396825396,
      "loss": 5.1614,
      "theoretical_loss": 5.745999198346823,
      "tokens_seen": 29163520
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.00044246031746031746,
      "loss": 5.0501,
      "theoretical_loss": 5.743902463307119,
      "tokens_seen": 29229056
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.00044345238095238096,
      "loss": 5.265,
      "theoretical_loss": 5.7418117371844275,
      "tokens_seen": 29294592
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004444444444444444,
      "loss": 5.3605,
      "theoretical_loss": 5.739726989373027,
      "tokens_seen": 29360128
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.00044543650793650797,
      "loss": 5.2102,
      "theoretical_loss": 5.737648189490956,
      "tokens_seen": 29425664
    },
    {
      "epoch": 0.01,
      "objective/train/advantage_avg": 0.4925776422023773,
      "objective/train/docs_used": 29115,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 5.341947555541992,
      "objective/train/original_loss": 5.341948509216309,
      "objective/train/theoretical_loss": 5.735575307377884,
      "objective/train/tokens_used": 49951200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2427273839712143,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0504915714263916,
      "objective/train/weighted_lm_loss": 5.611254692077637,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0425469875335693,
      "theoretical_loss": 5.735575307377884,
      "tokens_seen": 29491200
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.00044642857142857147,
      "loss": 5.0323,
      "theoretical_loss": 5.735575307377884,
      "tokens_seen": 29491200
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.00044742063492063497,
      "loss": 5.3969,
      "theoretical_loss": 5.733508313093,
      "tokens_seen": 29556736
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004484126984126984,
      "loss": 5.325,
      "theoretical_loss": 5.731447176912937,
      "tokens_seen": 29622272
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004494047619047619,
      "loss": 5.011,
      "theoretical_loss": 5.729391869329716,
      "tokens_seen": 29687808
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004503968253968254,
      "loss": 5.3939,
      "theoretical_loss": 5.727342361048713,
      "tokens_seen": 29753344
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004513888888888889,
      "loss": 5.3033,
      "theoretical_loss": 5.725298622986648,
      "tokens_seen": 29818880
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.00045238095238095237,
      "loss": 5.0085,
      "theoretical_loss": 5.723260626269605,
      "tokens_seen": 29884416
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.00045337301587301587,
      "loss": 5.0817,
      "theoretical_loss": 5.721228342231065,
      "tokens_seen": 29949952
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.00045436507936507937,
      "loss": 5.283,
      "theoretical_loss": 5.719201742409968,
      "tokens_seen": 30015488
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004553571428571429,
      "loss": 4.9355,
      "theoretical_loss": 5.717180798548798,
      "tokens_seen": 30081024
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004563492063492063,
      "loss": 5.4721,
      "theoretical_loss": 5.715165482591681,
      "tokens_seen": 30146560
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004573412698412698,
      "loss": 5.0003,
      "theoretical_loss": 5.713155766682521,
      "tokens_seen": 30212096
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004583333333333333,
      "loss": 4.6736,
      "theoretical_loss": 5.71115162316314,
      "tokens_seen": 30277632
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004593253968253968,
      "loss": 5.1334,
      "theoretical_loss": 5.709153024571448,
      "tokens_seen": 30343168
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.00046031746031746033,
      "loss": 5.199,
      "theoretical_loss": 5.707159943639638,
      "tokens_seen": 30408704
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.00046130952380952383,
      "loss": 5.0744,
      "theoretical_loss": 5.705172353292392,
      "tokens_seen": 30474240
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.00046230158730158733,
      "loss": 5.0145,
      "theoretical_loss": 5.703190226645112,
      "tokens_seen": 30539776
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.00046329365079365083,
      "loss": 5.2706,
      "theoretical_loss": 5.70121353700217,
      "tokens_seen": 30605312
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.00046428571428571433,
      "loss": 5.1444,
      "theoretical_loss": 5.699242257855182,
      "tokens_seen": 30670848
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004652777777777778,
      "loss": 5.2534,
      "theoretical_loss": 5.697276362881294,
      "tokens_seen": 30736384
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004662698412698413,
      "loss": 5.467,
      "theoretical_loss": 5.6953158259414955,
      "tokens_seen": 30801920
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004672619047619048,
      "loss": 5.3176,
      "theoretical_loss": 5.693360621078943,
      "tokens_seen": 30867456
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004682539682539683,
      "loss": 5.1162,
      "theoretical_loss": 5.691410722517309,
      "tokens_seen": 30932992
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.00046924603174603173,
      "loss": 5.0087,
      "theoretical_loss": 5.689466104659148,
      "tokens_seen": 30998528
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.00047023809523809523,
      "loss": 5.0309,
      "theoretical_loss": 5.6875267420842786,
      "tokens_seen": 31064064
    },
    {
      "epoch": 0.01,
      "objective/train/advantage_avg": 0.49429431557655334,
      "objective/train/docs_used": 30152,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 4.985562324523926,
      "objective/train/original_loss": 4.985561847686768,
      "objective/train/theoretical_loss": 5.685592609548186,
      "objective/train/tokens_used": 51589600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24438513815402985,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506716966629028,
      "objective/train/weighted_lm_loss": 5.2380242347717285,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0253151655197144,
      "theoretical_loss": 5.685592609548186,
      "tokens_seen": 31129600
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.00047123015873015874,
      "loss": 5.0097,
      "theoretical_loss": 5.685592609548186,
      "tokens_seen": 31129600
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.00047222222222222224,
      "loss": 5.0298,
      "theoretical_loss": 5.683663681980443,
      "tokens_seen": 31195136
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004732142857142857,
      "loss": 5.0499,
      "theoretical_loss": 5.68173993448314,
      "tokens_seen": 31260672
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004742063492063492,
      "loss": 5.0508,
      "theoretical_loss": 5.679821342329351,
      "tokens_seen": 31326208
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004751984126984127,
      "loss": 4.9167,
      "theoretical_loss": 5.677907880961591,
      "tokens_seen": 31391744
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004761904761904762,
      "loss": 5.0007,
      "theoretical_loss": 5.6759995259903135,
      "tokens_seen": 31457280
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004771825396825397,
      "loss": 5.0366,
      "theoretical_loss": 5.674096253192413,
      "tokens_seen": 31522816
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004781746031746032,
      "loss": 4.9913,
      "theoretical_loss": 5.672198038509743,
      "tokens_seen": 31588352
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004791666666666667,
      "loss": 5.3296,
      "theoretical_loss": 5.670304858047658,
      "tokens_seen": 31653888
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004801587301587302,
      "loss": 5.1249,
      "theoretical_loss": 5.66841668807356,
      "tokens_seen": 31719424
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004811507936507937,
      "loss": 5.1161,
      "theoretical_loss": 5.666533505015475,
      "tokens_seen": 31784960
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.00048214285714285715,
      "loss": 4.9459,
      "theoretical_loss": 5.664655285460632,
      "tokens_seen": 31850496
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.00048313492063492065,
      "loss": 5.2173,
      "theoretical_loss": 5.662782006154064,
      "tokens_seen": 31916032
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.00048412698412698415,
      "loss": 5.0951,
      "theoretical_loss": 5.660913643997225,
      "tokens_seen": 31981568
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.00048511904761904765,
      "loss": 4.93,
      "theoretical_loss": 5.659050176046617,
      "tokens_seen": 32047104
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004861111111111111,
      "loss": 5.0616,
      "theoretical_loss": 5.657191579512437,
      "tokens_seen": 32112640
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004871031746031746,
      "loss": 5.3957,
      "theoretical_loss": 5.655337831757233,
      "tokens_seen": 32178176
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004880952380952381,
      "loss": 5.0991,
      "theoretical_loss": 5.653488910294585,
      "tokens_seen": 32243712
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004890873015873016,
      "loss": 5.0489,
      "theoretical_loss": 5.651644792787781,
      "tokens_seen": 32309248
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004900793650793651,
      "loss": 5.0822,
      "theoretical_loss": 5.649805457048533,
      "tokens_seen": 32374784
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004910714285714286,
      "loss": 5.3497,
      "theoretical_loss": 5.647970881035681,
      "tokens_seen": 32440320
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.000492063492063492,
      "loss": 5.0531,
      "theoretical_loss": 5.646141042853927,
      "tokens_seen": 32505856
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004930555555555556,
      "loss": 4.9358,
      "theoretical_loss": 5.644315920752579,
      "tokens_seen": 32571392
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004940476190476191,
      "loss": 4.812,
      "theoretical_loss": 5.642495493124309,
      "tokens_seen": 32636928
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004950396825396826,
      "loss": 5.0656,
      "theoretical_loss": 5.640679738503916,
      "tokens_seen": 32702464
    },
    {
      "epoch": 0.01,
      "objective/train/advantage_avg": 0.49378877878189087,
      "objective/train/docs_used": 30758,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 5.122411727905273,
      "objective/train/original_loss": 5.122411727905273,
      "objective/train/theoretical_loss": 5.638868635567113,
      "objective/train/tokens_used": 53228000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2439243197441101,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506188869476318,
      "objective/train/weighted_lm_loss": 5.381689071655273,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 5.638868635567113,
      "tokens_seen": 32768000
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.000496031746031746,
      "loss": 5.3344,
      "theoretical_loss": 5.638868635567113,
      "tokens_seen": 32768000
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004970238095238095,
      "loss": 5.1051,
      "theoretical_loss": 5.6370621631293245,
      "tokens_seen": 32833536
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.000498015873015873,
      "loss": 4.8904,
      "theoretical_loss": 5.635260300144488,
      "tokens_seen": 32899072
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004990079365079365,
      "loss": 4.9756,
      "theoretical_loss": 5.633463025703879,
      "tokens_seen": 32964608
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0005,
      "loss": 5.1041,
      "theoretical_loss": 5.631670319034946,
      "tokens_seen": 33030144
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004999899699097292,
      "loss": 5.0526,
      "theoretical_loss": 5.6298821595001485,
      "tokens_seen": 33095680
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004999799398194584,
      "loss": 5.0709,
      "theoretical_loss": 5.628098526595821,
      "tokens_seen": 33161216
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004999699097291876,
      "loss": 5.013,
      "theoretical_loss": 5.626319399951038,
      "tokens_seen": 33226752
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004999598796389167,
      "loss": 4.9593,
      "theoretical_loss": 5.624544759326499,
      "tokens_seen": 33292288
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.000499949849548646,
      "loss": 5.1813,
      "theoretical_loss": 5.622774584613412,
      "tokens_seen": 33357824
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004999398194583751,
      "loss": 4.9004,
      "theoretical_loss": 5.621008855832411,
      "tokens_seen": 33423360
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004999297893681044,
      "loss": 5.154,
      "theoretical_loss": 5.619247553132459,
      "tokens_seen": 33488896
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004999197592778335,
      "loss": 4.9737,
      "theoretical_loss": 5.61749065678978,
      "tokens_seen": 33554432
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004999097291875627,
      "loss": 5.2621,
      "theoretical_loss": 5.615738147206796,
      "tokens_seen": 33619968
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004998996990972919,
      "loss": 5.1385,
      "theoretical_loss": 5.613990004911081,
      "tokens_seen": 33685504
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004998896690070211,
      "loss": 5.0918,
      "theoretical_loss": 5.612246210554302,
      "tokens_seen": 33751040
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004998796389167503,
      "loss": 5.0518,
      "theoretical_loss": 5.610506744911213,
      "tokens_seen": 33816576
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004998696088264795,
      "loss": 5.1352,
      "theoretical_loss": 5.608771588878618,
      "tokens_seen": 33882112
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004998595787362087,
      "loss": 5.2104,
      "theoretical_loss": 5.60704072347437,
      "tokens_seen": 33947648
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004998495486459378,
      "loss": 5.2858,
      "theoretical_loss": 5.605314129836366,
      "tokens_seen": 34013184
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.000499839518555667,
      "loss": 5.141,
      "theoretical_loss": 5.603591789221566,
      "tokens_seen": 34078720
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004998294884653962,
      "loss": 5.0401,
      "theoretical_loss": 5.601873683005008,
      "tokens_seen": 34144256
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004998194583751254,
      "loss": 4.9557,
      "theoretical_loss": 5.600159792678844,
      "tokens_seen": 34209792
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004998094282848546,
      "loss": 5.1873,
      "theoretical_loss": 5.598450099851373,
      "tokens_seen": 34275328
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004997993981945837,
      "loss": 5.2246,
      "theoretical_loss": 5.596744586246102,
      "tokens_seen": 34340864
    },
    {
      "epoch": 0.01,
      "objective/train/advantage_avg": 0.4945236146450043,
      "objective/train/docs_used": 31355,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 4.48709774017334,
      "objective/train/original_loss": 4.48709774017334,
      "objective/train/theoretical_loss": 5.595043233700797,
      "objective/train/tokens_used": 54866400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24461482465267181,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050695776939392,
      "objective/train/weighted_lm_loss": 4.7143988609313965,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 5.595043233700797,
      "tokens_seen": 34406400
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004997893681043129,
      "loss": 5.0142,
      "theoretical_loss": 5.595043233700797,
      "tokens_seen": 34406400
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004997793380140421,
      "loss": 5.2748,
      "theoretical_loss": 5.59334602416656,
      "tokens_seen": 34471936
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004997693079237714,
      "loss": 5.3186,
      "theoretical_loss": 5.591652939706902,
      "tokens_seen": 34537472
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004997592778335005,
      "loss": 5.1651,
      "theoretical_loss": 5.589963962496837,
      "tokens_seen": 34603008
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004997492477432298,
      "loss": 4.9555,
      "theoretical_loss": 5.588279074821972,
      "tokens_seen": 34668544
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004997392176529588,
      "loss": 5.1533,
      "theoretical_loss": 5.586598259077619,
      "tokens_seen": 34734080
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004997291875626881,
      "loss": 4.8208,
      "theoretical_loss": 5.58492149776791,
      "tokens_seen": 34799616
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004997191574724173,
      "loss": 5.0212,
      "theoretical_loss": 5.583248773504918,
      "tokens_seen": 34865152
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004997091273821465,
      "loss": 5.1963,
      "theoretical_loss": 5.581580069007791,
      "tokens_seen": 34930688
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004996990972918757,
      "loss": 4.9749,
      "theoretical_loss": 5.5799153671018935,
      "tokens_seen": 34996224
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004996890672016048,
      "loss": 5.1695,
      "theoretical_loss": 5.578254650717955,
      "tokens_seen": 35061760
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.000499679037111334,
      "loss": 5.2285,
      "theoretical_loss": 5.576597902891232,
      "tokens_seen": 35127296
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004996690070210632,
      "loss": 5.0775,
      "theoretical_loss": 5.574945106760667,
      "tokens_seen": 35192832
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004996589769307924,
      "loss": 4.9762,
      "theoretical_loss": 5.573296245568071,
      "tokens_seen": 35258368
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004996489468405216,
      "loss": 5.0611,
      "theoretical_loss": 5.571651302657301,
      "tokens_seen": 35323904
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004996389167502507,
      "loss": 5.0654,
      "theoretical_loss": 5.570010261473453,
      "tokens_seen": 35389440
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004996288866599799,
      "loss": 4.8661,
      "theoretical_loss": 5.568373105562058,
      "tokens_seen": 35454976
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004996188565697091,
      "loss": 4.9367,
      "theoretical_loss": 5.566739818568294,
      "tokens_seen": 35520512
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004996088264794383,
      "loss": 4.8894,
      "theoretical_loss": 5.565110384236193,
      "tokens_seen": 35586048
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004995987963891675,
      "loss": 5.0399,
      "theoretical_loss": 5.563484786407864,
      "tokens_seen": 35651584
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004995887662988968,
      "loss": 4.7906,
      "theoretical_loss": 5.56186300902273,
      "tokens_seen": 35717120
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004995787362086258,
      "loss": 4.9335,
      "theoretical_loss": 5.560245036116753,
      "tokens_seen": 35782656
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004995687061183551,
      "loss": 5.2027,
      "theoretical_loss": 5.55863085182169,
      "tokens_seen": 35848192
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004995586760280842,
      "loss": 5.1525,
      "theoretical_loss": 5.557020440364331,
      "tokens_seen": 35913728
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004995486459378135,
      "loss": 4.7596,
      "theoretical_loss": 5.555413786065771,
      "tokens_seen": 35979264
    },
    {
      "epoch": 0.01,
      "objective/train/advantage_avg": 0.49424242973327637,
      "objective/train/docs_used": 32385,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 4.772250652313232,
      "objective/train/original_loss": 4.772250652313232,
      "objective/train/theoretical_loss": 5.553810873340668,
      "objective/train/tokens_used": 56504800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24435585737228394,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050666332244873,
      "objective/train/weighted_lm_loss": 5.013638496398926,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0408108234405518,
      "theoretical_loss": 5.553810873340668,
      "tokens_seen": 36044800
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004995386158475427,
      "loss": 5.0525,
      "theoretical_loss": 5.553810873340668,
      "tokens_seen": 36044800
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004995285857572718,
      "loss": 4.9762,
      "theoretical_loss": 5.5522116866965145,
      "tokens_seen": 36110336
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.000499518555667001,
      "loss": 5.186,
      "theoretical_loss": 5.550616210732926,
      "tokens_seen": 36175872
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004995085255767302,
      "loss": 5.1386,
      "theoretical_loss": 5.549024430140915,
      "tokens_seen": 36241408
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004994984954864594,
      "loss": 4.6216,
      "theoretical_loss": 5.547436329702196,
      "tokens_seen": 36306944
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004994884653961886,
      "loss": 5.1468,
      "theoretical_loss": 5.545851894288477,
      "tokens_seen": 36372480
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004994784353059178,
      "loss": 5.321,
      "theoretical_loss": 5.544271108860773,
      "tokens_seen": 36438016
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004994684052156469,
      "loss": 4.7888,
      "theoretical_loss": 5.542693958468716,
      "tokens_seen": 36503552
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004994583751253761,
      "loss": 4.9266,
      "theoretical_loss": 5.5411204282498705,
      "tokens_seen": 36569088
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004994483450351053,
      "loss": 4.9023,
      "theoretical_loss": 5.539550503429072,
      "tokens_seen": 36634624
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004994383149448345,
      "loss": 5.0593,
      "theoretical_loss": 5.537984169317745,
      "tokens_seen": 36700160
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004994282848545637,
      "loss": 4.7155,
      "theoretical_loss": 5.536421411313254,
      "tokens_seen": 36765696
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004994182547642928,
      "loss": 4.8802,
      "theoretical_loss": 5.534862214898242,
      "tokens_seen": 36831232
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004994082246740221,
      "loss": 4.9123,
      "theoretical_loss": 5.533306565639981,
      "tokens_seen": 36896768
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004993981945837512,
      "loss": 4.8698,
      "theoretical_loss": 5.531754449189732,
      "tokens_seen": 36962304
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004993881644934805,
      "loss": 4.9766,
      "theoretical_loss": 5.530205851282113,
      "tokens_seen": 37027840
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004993781344032096,
      "loss": 4.8891,
      "theoretical_loss": 5.528660757734458,
      "tokens_seen": 37093376
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004993681043129389,
      "loss": 5.0524,
      "theoretical_loss": 5.527119154446201,
      "tokens_seen": 37158912
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.000499358074222668,
      "loss": 4.9397,
      "theoretical_loss": 5.525581027398248,
      "tokens_seen": 37224448
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004993480441323972,
      "loss": 4.9622,
      "theoretical_loss": 5.524046362652378,
      "tokens_seen": 37289984
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004993380140421264,
      "loss": 5.0195,
      "theoretical_loss": 5.522515146350619,
      "tokens_seen": 37355520
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004993279839518556,
      "loss": 4.887,
      "theoretical_loss": 5.520987364714657,
      "tokens_seen": 37421056
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004993179538615848,
      "loss": 4.8183,
      "theoretical_loss": 5.519463004045237,
      "tokens_seen": 37486592
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004993079237713139,
      "loss": 4.993,
      "theoretical_loss": 5.517942050721569,
      "tokens_seen": 37552128
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004992978936810431,
      "loss": 4.6795,
      "theoretical_loss": 5.5164244912007465,
      "tokens_seen": 37617664
    },
    {
      "epoch": 0.01,
      "objective/train/advantage_avg": 0.4941908121109009,
      "objective/train/docs_used": 32883,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 4.9640278816223145,
      "objective/train/original_loss": 4.964028835296631,
      "objective/train/theoretical_loss": 5.514910312017166,
      "objective/train/tokens_used": 58143200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24427852034568787,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506608486175537,
      "objective/train/weighted_lm_loss": 5.215147972106934,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0408108234405518,
      "theoretical_loss": 5.514910312017166,
      "tokens_seen": 37683200
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004992878635907723,
      "loss": 5.0264,
      "theoretical_loss": 5.514910312017166,
      "tokens_seen": 37683200
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004992778335005015,
      "loss": 4.9097,
      "theoretical_loss": 5.513399499781949,
      "tokens_seen": 37748736
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004992678034102307,
      "loss": 5.0727,
      "theoretical_loss": 5.511892041182376,
      "tokens_seen": 37814272
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004992577733199598,
      "loss": 5.1875,
      "theoretical_loss": 5.5103879229813195,
      "tokens_seen": 37879808
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.000499247743229689,
      "loss": 5.0382,
      "theoretical_loss": 5.508887132016688,
      "tokens_seen": 37945344
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004992377131394183,
      "loss": 5.1858,
      "theoretical_loss": 5.507389655200869,
      "tokens_seen": 38010880
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004992276830491475,
      "loss": 5.0991,
      "theoretical_loss": 5.505895479520182,
      "tokens_seen": 38076416
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004992176529588767,
      "loss": 5.009,
      "theoretical_loss": 5.504404592034333,
      "tokens_seen": 38141952
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004992076228686059,
      "loss": 4.9756,
      "theoretical_loss": 5.5029169798758755,
      "tokens_seen": 38207488
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.000499197592778335,
      "loss": 4.8651,
      "theoretical_loss": 5.5014326302496785,
      "tokens_seen": 38273024
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004991875626880642,
      "loss": 4.7149,
      "theoretical_loss": 5.499951530432394,
      "tokens_seen": 38338560
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004991775325977934,
      "loss": 4.9537,
      "theoretical_loss": 5.498473667771936,
      "tokens_seen": 38404096
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004991675025075226,
      "loss": 4.8177,
      "theoretical_loss": 5.496999029686955,
      "tokens_seen": 38469632
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004991574724172518,
      "loss": 5.1228,
      "theoretical_loss": 5.4955276036663285,
      "tokens_seen": 38535168
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.000499147442326981,
      "loss": 5.1861,
      "theoretical_loss": 5.494059377268647,
      "tokens_seen": 38600704
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004991374122367101,
      "loss": 4.763,
      "theoretical_loss": 5.492594338121711,
      "tokens_seen": 38666240
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004991273821464393,
      "loss": 5.0462,
      "theoretical_loss": 5.4911324739220255,
      "tokens_seen": 38731776
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004991173520561685,
      "loss": 4.8001,
      "theoretical_loss": 5.4896737724343065,
      "tokens_seen": 38797312
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004991073219658977,
      "loss": 4.9527,
      "theoretical_loss": 5.488218221490987,
      "tokens_seen": 38862848
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.000499097291875627,
      "loss": 5.0325,
      "theoretical_loss": 5.486765808991725,
      "tokens_seen": 38928384
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.000499087261785356,
      "loss": 5.0607,
      "theoretical_loss": 5.48531652290293,
      "tokens_seen": 38993920
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004990772316950853,
      "loss": 4.869,
      "theoretical_loss": 5.483870351257271,
      "tokens_seen": 39059456
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004990672016048144,
      "loss": 5.116,
      "theoretical_loss": 5.482427282153209,
      "tokens_seen": 39124992
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004990571715145437,
      "loss": 4.8826,
      "theoretical_loss": 5.480987303754524,
      "tokens_seen": 39190528
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004990471414242729,
      "loss": 4.8493,
      "theoretical_loss": 5.479550404289846,
      "tokens_seen": 39256064
    },
    {
      "epoch": 0.01,
      "objective/train/advantage_avg": 0.4922616183757782,
      "objective/train/docs_used": 34166,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 4.600875377655029,
      "objective/train/original_loss": 4.600874900817871,
      "objective/train/theoretical_loss": 5.478116572052198,
      "objective/train/tokens_used": 59781600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24241067469120026,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050458312034607,
      "objective/train/weighted_lm_loss": 4.832952976226807,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0338950157165527,
      "theoretical_loss": 5.478116572052198,
      "tokens_seen": 39321600
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.000499037111334002,
      "loss": 4.9662,
      "theoretical_loss": 5.478116572052198,
      "tokens_seen": 39321600
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004990270812437312,
      "loss": 5.0489,
      "theoretical_loss": 5.476685795398528,
      "tokens_seen": 39387136
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004990170511534604,
      "loss": 4.7121,
      "theoretical_loss": 5.475258062749262,
      "tokens_seen": 39452672
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004990070210631896,
      "loss": 4.6496,
      "theoretical_loss": 5.4738333625878495,
      "tokens_seen": 39518208
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004989969909729188,
      "loss": 4.8225,
      "theoretical_loss": 5.472411683460317,
      "tokens_seen": 39583744
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.000498986960882648,
      "loss": 4.7348,
      "theoretical_loss": 5.470993013974829,
      "tokens_seen": 39649280
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004989769307923771,
      "loss": 5.2324,
      "theoretical_loss": 5.469577342801242,
      "tokens_seen": 39714816
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004989669007021063,
      "loss": 5.0489,
      "theoretical_loss": 5.468164658670672,
      "tokens_seen": 39780352
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004989568706118355,
      "loss": 4.9712,
      "theoretical_loss": 5.466754950375066,
      "tokens_seen": 39845888
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004989468405215647,
      "loss": 5.1365,
      "theoretical_loss": 5.465348206766773,
      "tokens_seen": 39911424
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004989368104312939,
      "loss": 4.8438,
      "theoretical_loss": 5.463944416758115,
      "tokens_seen": 39976960
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.000498926780341023,
      "loss": 4.8405,
      "theoretical_loss": 5.4625435693209745,
      "tokens_seen": 40042496
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004989167502507523,
      "loss": 4.992,
      "theoretical_loss": 5.461145653486371,
      "tokens_seen": 40108032
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004989067201604814,
      "loss": 5.1223,
      "theoretical_loss": 5.459750658344051,
      "tokens_seen": 40173568
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004988966900702107,
      "loss": 5.0841,
      "theoretical_loss": 5.45835857304208,
      "tokens_seen": 40239104
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004988866599799398,
      "loss": 4.8155,
      "theoretical_loss": 5.456969386786434,
      "tokens_seen": 40304640
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004988766298896691,
      "loss": 4.8269,
      "theoretical_loss": 5.455583088840594,
      "tokens_seen": 40370176
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004988665997993982,
      "loss": 4.8451,
      "theoretical_loss": 5.454199668525155,
      "tokens_seen": 40435712
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004988565697091274,
      "loss": 4.9097,
      "theoretical_loss": 5.452819115217421,
      "tokens_seen": 40501248
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004988465396188566,
      "loss": 4.9266,
      "theoretical_loss": 5.4514414183510205,
      "tokens_seen": 40566784
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004988365095285858,
      "loss": 4.3644,
      "theoretical_loss": 5.4500665674155115,
      "tokens_seen": 40632320
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.000498826479438315,
      "loss": 5.0489,
      "theoretical_loss": 5.448694551956001,
      "tokens_seen": 40697856
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004988164493480441,
      "loss": 4.7673,
      "theoretical_loss": 5.447325361572755,
      "tokens_seen": 40763392
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004988064192577733,
      "loss": 4.6884,
      "theoretical_loss": 5.445958985920832,
      "tokens_seen": 40828928
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004987963891675025,
      "loss": 4.9417,
      "theoretical_loss": 5.444595414709691,
      "tokens_seen": 40894464
    },
    {
      "epoch": 0.01,
      "objective/train/advantage_avg": 0.4927012622356415,
      "objective/train/docs_used": 34866,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 4.573291778564453,
      "objective/train/original_loss": 4.573292255401611,
      "objective/train/theoretical_loss": 5.443234637702833,
      "objective/train/tokens_used": 61420000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24287398159503937,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505046844482422,
      "objective/train/weighted_lm_loss": 4.803847312927246,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0396549701690674,
      "theoretical_loss": 5.443234637702833,
      "tokens_seen": 40960000
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004987863590772317,
      "loss": 4.7859,
      "theoretical_loss": 5.443234637702833,
      "tokens_seen": 40960000
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004987763289869609,
      "loss": 4.9553,
      "theoretical_loss": 5.441876644717422,
      "tokens_seen": 41025536
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.00049876629889669,
      "loss": 4.899,
      "theoretical_loss": 5.440521425623924,
      "tokens_seen": 41091072
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004987562688064192,
      "loss": 4.8412,
      "theoretical_loss": 5.439168970345737,
      "tokens_seen": 41156608
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004987462387161484,
      "loss": 5.0822,
      "theoretical_loss": 5.4378192688588385,
      "tokens_seen": 41222144
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004987362086258777,
      "loss": 4.8155,
      "theoretical_loss": 5.436472311191425,
      "tokens_seen": 41287680
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004987261785356068,
      "loss": 5.0032,
      "theoretical_loss": 5.435128087423554,
      "tokens_seen": 41353216
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004987161484453361,
      "loss": 4.9404,
      "theoretical_loss": 5.433786587686798,
      "tokens_seen": 41418752
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004987061183550651,
      "loss": 4.8672,
      "theoretical_loss": 5.432447802163894,
      "tokens_seen": 41484288
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004986960882647944,
      "loss": 4.9075,
      "theoretical_loss": 5.431111721088394,
      "tokens_seen": 41549824
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004986860581745236,
      "loss": 4.9918,
      "theoretical_loss": 5.429778334744331,
      "tokens_seen": 41615360
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004986760280842528,
      "loss": 4.8588,
      "theoretical_loss": 5.428447633465867,
      "tokens_seen": 41680896
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.000498665997993982,
      "loss": 4.969,
      "theoretical_loss": 5.427119607636971,
      "tokens_seen": 41746432
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004986559679037111,
      "loss": 4.8339,
      "theoretical_loss": 5.42579424769107,
      "tokens_seen": 41811968
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004986459378134403,
      "loss": 5.0642,
      "theoretical_loss": 5.424471544110728,
      "tokens_seen": 41877504
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004986359077231695,
      "loss": 4.7846,
      "theoretical_loss": 5.423151487427312,
      "tokens_seen": 41943040
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004986258776328987,
      "loss": 4.6307,
      "theoretical_loss": 5.421834068220666,
      "tokens_seen": 42008576
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004986158475426279,
      "loss": 4.5768,
      "theoretical_loss": 5.420519277118791,
      "tokens_seen": 42074112
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.000498605817452357,
      "loss": 4.825,
      "theoretical_loss": 5.419207104797523,
      "tokens_seen": 42139648
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004985957873620862,
      "loss": 5.0369,
      "theoretical_loss": 5.41789754198021,
      "tokens_seen": 42205184
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004985857572718154,
      "loss": 4.8771,
      "theoretical_loss": 5.416590579437406,
      "tokens_seen": 42270720
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004985757271815446,
      "loss": 4.8526,
      "theoretical_loss": 5.415286207986547,
      "tokens_seen": 42336256
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004985656970912738,
      "loss": 4.6646,
      "theoretical_loss": 5.413984418491651,
      "tokens_seen": 42401792
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004985556670010031,
      "loss": 4.7217,
      "theoretical_loss": 5.412685201863004,
      "tokens_seen": 42467328
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004985456369107321,
      "loss": 4.7501,
      "theoretical_loss": 5.411388549056857,
      "tokens_seen": 42532864
    },
    {
      "epoch": 0.01,
      "objective/train/advantage_avg": 0.4947051405906677,
      "objective/train/docs_used": 35868,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 4.4424614906311035,
      "objective/train/original_loss": 4.4424614906311035,
      "objective/train/theoretical_loss": 5.410094451075121,
      "objective/train/tokens_used": 63058400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24480774998664856,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507148504257202,
      "objective/train/weighted_lm_loss": 4.667764186859131,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 5.410094451075121,
      "tokens_seen": 42598400
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004985356068204614,
      "loss": 4.7571,
      "theoretical_loss": 5.410094451075121,
      "tokens_seen": 42598400
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004985255767301905,
      "loss": 4.8982,
      "theoretical_loss": 5.408802898965068,
      "tokens_seen": 42663936
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004985155466399198,
      "loss": 4.9409,
      "theoretical_loss": 5.4075138838190355,
      "tokens_seen": 42729472
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.000498505516549649,
      "loss": 4.8996,
      "theoretical_loss": 5.406227396774128,
      "tokens_seen": 42795008
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004984954864593782,
      "loss": 4.7546,
      "theoretical_loss": 5.4049434290119205,
      "tokens_seen": 42860544
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004984854563691073,
      "loss": 4.8412,
      "theoretical_loss": 5.4036619717581775,
      "tokens_seen": 42926080
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004984754262788365,
      "loss": 4.8446,
      "theoretical_loss": 5.402383016282556,
      "tokens_seen": 42991616
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004984653961885657,
      "loss": 4.3612,
      "theoretical_loss": 5.401106553898321,
      "tokens_seen": 43057152
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004984553660982949,
      "loss": 4.9882,
      "theoretical_loss": 5.3998325759620664,
      "tokens_seen": 43122688
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004984453360080241,
      "loss": 4.7743,
      "theoretical_loss": 5.3985610738734255,
      "tokens_seen": 43188224
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004984353059177532,
      "loss": 4.378,
      "theoretical_loss": 5.397292039074801,
      "tokens_seen": 43253760
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004984252758274825,
      "loss": 4.763,
      "theoretical_loss": 5.396025463051079,
      "tokens_seen": 43319296
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004984152457372116,
      "loss": 4.8128,
      "theoretical_loss": 5.394761337329362,
      "tokens_seen": 43384832
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004984052156469409,
      "loss": 5.0514,
      "theoretical_loss": 5.3934996534786865,
      "tokens_seen": 43450368
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.00049839518555667,
      "loss": 4.8722,
      "theoretical_loss": 5.392240403109765,
      "tokens_seen": 43515904
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004983851554663993,
      "loss": 4.9351,
      "theoretical_loss": 5.390983577874708,
      "tokens_seen": 43581440
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004983751253761284,
      "loss": 4.8217,
      "theoretical_loss": 5.3897291694667615,
      "tokens_seen": 43646976
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004983650952858576,
      "loss": 4.8447,
      "theoretical_loss": 5.388477169620044,
      "tokens_seen": 43712512
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004983550651955868,
      "loss": 4.9096,
      "theoretical_loss": 5.38722757010928,
      "tokens_seen": 43778048
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.000498345035105316,
      "loss": 4.7463,
      "theoretical_loss": 5.385980362749544,
      "tokens_seen": 43843584
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004983350050150452,
      "loss": 4.6107,
      "theoretical_loss": 5.384735539396003,
      "tokens_seen": 43909120
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004983249749247743,
      "loss": 4.5777,
      "theoretical_loss": 5.383493091943657,
      "tokens_seen": 43974656
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004983149448345035,
      "loss": 4.5833,
      "theoretical_loss": 5.382253012327088,
      "tokens_seen": 44040192
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004983049147442327,
      "loss": 4.9201,
      "theoretical_loss": 5.3810152925202095,
      "tokens_seen": 44105728
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004982948846539619,
      "loss": 4.5119,
      "theoretical_loss": 5.379779924536011,
      "tokens_seen": 44171264
    },
    {
      "epoch": 0.01,
      "objective/train/advantage_avg": 0.49375373125076294,
      "objective/train/docs_used": 36544,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 4.484834671020508,
      "objective/train/original_loss": 4.48483419418335,
      "objective/train/theoretical_loss": 5.3785469004263184,
      "objective/train/tokens_used": 64696800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24387480318546295,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506151914596558,
      "objective/train/weighted_lm_loss": 4.711275577545166,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0304546356201172,
      "theoretical_loss": 5.3785469004263184,
      "tokens_seen": 44236800
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004982848545636911,
      "loss": 4.8583,
      "theoretical_loss": 5.3785469004263184,
      "tokens_seen": 44236800
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004982748244734202,
      "loss": 4.525,
      "theoretical_loss": 5.377316212281538,
      "tokens_seen": 44302336
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004982647943831494,
      "loss": 4.8104,
      "theoretical_loss": 5.3760878522304205,
      "tokens_seen": 44367872
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004982547642928786,
      "loss": 4.7647,
      "theoretical_loss": 5.374861812439816,
      "tokens_seen": 44433408
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004982447342026079,
      "loss": 4.4843,
      "theoretical_loss": 5.373638085114435,
      "tokens_seen": 44498944
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.000498234704112337,
      "loss": 4.7091,
      "theoretical_loss": 5.372416662496606,
      "tokens_seen": 44564480
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004982246740220663,
      "loss": 4.9236,
      "theoretical_loss": 5.371197536866044,
      "tokens_seen": 44630016
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004982146439317953,
      "loss": 4.8081,
      "theoretical_loss": 5.36998070053961,
      "tokens_seen": 44695552
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004982046138415246,
      "loss": 5.1032,
      "theoretical_loss": 5.368766145871087,
      "tokens_seen": 44761088
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004981945837512538,
      "loss": 4.7481,
      "theoretical_loss": 5.367553865250939,
      "tokens_seen": 44826624
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.000498184553660983,
      "loss": 4.8747,
      "theoretical_loss": 5.3663438511060875,
      "tokens_seen": 44892160
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004981745235707122,
      "loss": 4.9846,
      "theoretical_loss": 5.3651360958996825,
      "tokens_seen": 44957696
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004981644934804413,
      "loss": 4.9424,
      "theoretical_loss": 5.3639305921308775,
      "tokens_seen": 45023232
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004981544633901705,
      "loss": 4.7992,
      "theoretical_loss": 5.362727332334602,
      "tokens_seen": 45088768
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004981444332998997,
      "loss": 4.7888,
      "theoretical_loss": 5.361526309081345,
      "tokens_seen": 45154304
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004981344032096289,
      "loss": 4.5437,
      "theoretical_loss": 5.3603275149769285,
      "tokens_seen": 45219840
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004981243731193581,
      "loss": 4.971,
      "theoretical_loss": 5.3591309426622935,
      "tokens_seen": 45285376
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004981143430290873,
      "loss": 4.5755,
      "theoretical_loss": 5.357936584813277,
      "tokens_seen": 45350912
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004981043129388164,
      "loss": 4.6638,
      "theoretical_loss": 5.3567444341404045,
      "tokens_seen": 45416448
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004980942828485456,
      "loss": 4.8723,
      "theoretical_loss": 5.355554483388666,
      "tokens_seen": 45481984
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004980842527582748,
      "loss": 4.8567,
      "theoretical_loss": 5.354366725337309,
      "tokens_seen": 45547520
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.000498074222668004,
      "loss": 4.578,
      "theoretical_loss": 5.353181152799634,
      "tokens_seen": 45613056
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004980641925777333,
      "loss": 4.6074,
      "theoretical_loss": 5.351997758622769,
      "tokens_seen": 45678592
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004980541624874623,
      "loss": 4.9217,
      "theoretical_loss": 5.35081653568748,
      "tokens_seen": 45744128
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004980441323971916,
      "loss": 4.6212,
      "theoretical_loss": 5.3496374769079535,
      "tokens_seen": 45809664
    },
    {
      "epoch": 0.01,
      "objective/train/advantage_avg": 0.4935201406478882,
      "objective/train/docs_used": 37915,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.783914566040039,
      "objective/train/original_loss": 3.783914566040039,
      "objective/train/theoretical_loss": 5.348460575231594,
      "objective/train/tokens_used": 66335200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24365109205245972,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505905151367188,
      "objective/train/weighted_lm_loss": 3.9749910831451416,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0363596677780151,
      "theoretical_loss": 5.348460575231594,
      "tokens_seen": 45875200
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004980341023069207,
      "loss": 4.6933,
      "theoretical_loss": 5.348460575231594,
      "tokens_seen": 45875200
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.00049802407221665,
      "loss": 4.535,
      "theoretical_loss": 5.347285823638824,
      "tokens_seen": 45940736
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004980140421263792,
      "loss": 4.776,
      "theoretical_loss": 5.3461132151428785,
      "tokens_seen": 46006272
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004980040120361084,
      "loss": 4.8102,
      "theoretical_loss": 5.344942742789613,
      "tokens_seen": 46071808
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004979939819458375,
      "loss": 4.8361,
      "theoretical_loss": 5.343774399657292,
      "tokens_seen": 46137344
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004979839518555667,
      "loss": 4.8563,
      "theoretical_loss": 5.342608178856405,
      "tokens_seen": 46202880
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004979739217652959,
      "loss": 4.7956,
      "theoretical_loss": 5.341444073529464,
      "tokens_seen": 46268416
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004979638916750251,
      "loss": 4.9068,
      "theoretical_loss": 5.340282076850814,
      "tokens_seen": 46333952
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004979538615847543,
      "loss": 4.6787,
      "theoretical_loss": 5.339122182026433,
      "tokens_seen": 46399488
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004979438314944834,
      "loss": 4.638,
      "theoretical_loss": 5.337964382293748,
      "tokens_seen": 46465024
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004979338014042126,
      "loss": 4.5943,
      "theoretical_loss": 5.336808670921445,
      "tokens_seen": 46530560
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004979237713139418,
      "loss": 4.6283,
      "theoretical_loss": 5.335655041209275,
      "tokens_seen": 46596096
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.000497913741223671,
      "loss": 4.7468,
      "theoretical_loss": 5.3345034864878755,
      "tokens_seen": 46661632
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004979037111334002,
      "loss": 4.6526,
      "theoretical_loss": 5.333354000118574,
      "tokens_seen": 46727168
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004978936810431293,
      "loss": 4.9602,
      "theoretical_loss": 5.332206575493219,
      "tokens_seen": 46792704
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004978836509528586,
      "loss": 4.5988,
      "theoretical_loss": 5.33106120603398,
      "tokens_seen": 46858240
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004978736208625877,
      "loss": 4.8558,
      "theoretical_loss": 5.329917885193181,
      "tokens_seen": 46923776
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.000497863590772317,
      "loss": 4.5915,
      "theoretical_loss": 5.328776606453111,
      "tokens_seen": 46989312
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004978535606820461,
      "loss": 4.7125,
      "theoretical_loss": 5.327637363325849,
      "tokens_seen": 47054848
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004978435305917754,
      "loss": 4.6375,
      "theoretical_loss": 5.326500149353086,
      "tokens_seen": 47120384
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004978335005015045,
      "loss": 4.5484,
      "theoretical_loss": 5.325364958105951,
      "tokens_seen": 47185920
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004978234704112337,
      "loss": 4.7712,
      "theoretical_loss": 5.324231783184828,
      "tokens_seen": 47251456
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004978134403209629,
      "loss": 4.7839,
      "theoretical_loss": 5.323100618219199,
      "tokens_seen": 47316992
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004978034102306921,
      "loss": 4.8194,
      "theoretical_loss": 5.32197145686745,
      "tokens_seen": 47382528
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004977933801404213,
      "loss": 4.8657,
      "theoretical_loss": 5.3208442928167194,
      "tokens_seen": 47448064
    },
    {
      "epoch": 0.01,
      "objective/train/advantage_avg": 0.4946458339691162,
      "objective/train/docs_used": 38651,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 4.209277629852295,
      "objective/train/original_loss": 4.209278106689453,
      "objective/train/theoretical_loss": 5.319719119782715,
      "objective/train/tokens_used": 67973600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24475158751010895,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507086515426636,
      "objective/train/weighted_lm_loss": 4.422505855560303,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 5.319719119782715,
      "tokens_seen": 47513600
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004977833500501504,
      "loss": 4.6473,
      "theoretical_loss": 5.319719119782715,
      "tokens_seen": 47513600
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004977733199598796,
      "loss": 4.5452,
      "theoretical_loss": 5.318595931509554,
      "tokens_seen": 47579136
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004977632898696088,
      "loss": 4.694,
      "theoretical_loss": 5.317474721769591,
      "tokens_seen": 47644672
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.000497753259779338,
      "loss": 4.6493,
      "theoretical_loss": 5.3163554843632514,
      "tokens_seen": 47710208
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004977432296890672,
      "loss": 4.6584,
      "theoretical_loss": 5.31523821311887,
      "tokens_seen": 47775744
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004977331995987965,
      "loss": 4.807,
      "theoretical_loss": 5.314122901892526,
      "tokens_seen": 47841280
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004977231695085255,
      "loss": 4.8187,
      "theoretical_loss": 5.31300954456788,
      "tokens_seen": 47906816
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004977131394182548,
      "loss": 4.6498,
      "theoretical_loss": 5.31189813505601,
      "tokens_seen": 47972352
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.000497703109327984,
      "loss": 4.7185,
      "theoretical_loss": 5.310788667295256,
      "tokens_seen": 48037888
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004976930792377132,
      "loss": 4.7191,
      "theoretical_loss": 5.3096811352510604,
      "tokens_seen": 48103424
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004976830491474424,
      "loss": 4.843,
      "theoretical_loss": 5.308575532915805,
      "tokens_seen": 48168960
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004976730190571715,
      "loss": 4.6198,
      "theoretical_loss": 5.307471854308661,
      "tokens_seen": 48234496
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004976629889669007,
      "loss": 4.6078,
      "theoretical_loss": 5.306370093475426,
      "tokens_seen": 48300032
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004976529588766299,
      "loss": 4.9114,
      "theoretical_loss": 5.305270244488379,
      "tokens_seen": 48365568
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004976429287863591,
      "loss": 4.7976,
      "theoretical_loss": 5.304172301446117,
      "tokens_seen": 48431104
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004976328986960883,
      "loss": 4.5802,
      "theoretical_loss": 5.303076258473412,
      "tokens_seen": 48496640
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004976228686058175,
      "loss": 4.5806,
      "theoretical_loss": 5.30198210972105,
      "tokens_seen": 48562176
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004976128385155466,
      "loss": 4.5835,
      "theoretical_loss": 5.300889849365689,
      "tokens_seen": 48627712
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004976028084252758,
      "loss": 4.5418,
      "theoretical_loss": 5.299799471609704,
      "tokens_seen": 48693248
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.000497592778335005,
      "loss": 4.6508,
      "theoretical_loss": 5.298710970681048,
      "tokens_seen": 48758784
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004975827482447342,
      "loss": 4.523,
      "theoretical_loss": 5.2976243408330905,
      "tokens_seen": 48824320
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004975727181544635,
      "loss": 4.3496,
      "theoretical_loss": 5.296539576344487,
      "tokens_seen": 48889856
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004975626880641925,
      "loss": 4.671,
      "theoretical_loss": 5.295456671519022,
      "tokens_seen": 48955392
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004975526579739218,
      "loss": 4.6712,
      "theoretical_loss": 5.294375620685475,
      "tokens_seen": 49020928
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004975426278836509,
      "loss": 4.6292,
      "theoretical_loss": 5.293296418197466,
      "tokens_seen": 49086464
    },
    {
      "epoch": 0.01,
      "objective/train/advantage_avg": 0.4936019480228424,
      "objective/train/docs_used": 39958,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 4.204228401184082,
      "objective/train/original_loss": 4.20422887802124,
      "objective/train/theoretical_loss": 5.292219058433327,
      "objective/train/tokens_used": 69612000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24368762969970703,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505988597869873,
      "objective/train/weighted_lm_loss": 4.417058944702148,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.044721245765686,
      "theoretical_loss": 5.292219058433327,
      "tokens_seen": 49152000
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004975325977933802,
      "loss": 4.4328,
      "theoretical_loss": 5.292219058433327,
      "tokens_seen": 49152000
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004975225677031094,
      "loss": 4.5289,
      "theoretical_loss": 5.29114353579595,
      "tokens_seen": 49217536
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004975125376128386,
      "loss": 4.4027,
      "theoretical_loss": 5.290069844712654,
      "tokens_seen": 49283072
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004975025075225677,
      "loss": 4.566,
      "theoretical_loss": 5.2889979796350435,
      "tokens_seen": 49348608
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004974924774322969,
      "loss": 4.2385,
      "theoretical_loss": 5.28792793503887,
      "tokens_seen": 49414144
    },
    {
      "epoch": 0.01,
      "learning_rate": 0.0004974824473420261,
      "loss": 4.5296,
      "theoretical_loss": 5.286859705423897,
      "tokens_seen": 49479680
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004974724172517553,
      "loss": 4.7595,
      "theoretical_loss": 5.285793285313759,
      "tokens_seen": 49545216
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004974623871614845,
      "loss": 4.2492,
      "theoretical_loss": 5.284728669255836,
      "tokens_seen": 49610752
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004974523570712136,
      "loss": 4.8316,
      "theoretical_loss": 5.283665851821111,
      "tokens_seen": 49676288
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004974423269809428,
      "loss": 4.5118,
      "theoretical_loss": 5.282604827604038,
      "tokens_seen": 49741824
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.000497432296890672,
      "loss": 4.3841,
      "theoretical_loss": 5.2815455912224145,
      "tokens_seen": 49807360
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004974222668004012,
      "loss": 4.6406,
      "theoretical_loss": 5.280488137317242,
      "tokens_seen": 49872896
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004974122367101304,
      "loss": 4.7142,
      "theoretical_loss": 5.2794324605526075,
      "tokens_seen": 49938432
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004974022066198595,
      "loss": 4.7943,
      "theoretical_loss": 5.278378555615541,
      "tokens_seen": 50003968
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004973921765295888,
      "loss": 4.4601,
      "theoretical_loss": 5.277326417215896,
      "tokens_seen": 50069504
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004973821464393179,
      "loss": 4.518,
      "theoretical_loss": 5.276276040086216,
      "tokens_seen": 50135040
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004973721163490472,
      "loss": 4.5064,
      "theoretical_loss": 5.275227418981615,
      "tokens_seen": 50200576
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004973620862587763,
      "loss": 4.5981,
      "theoretical_loss": 5.27418054867964,
      "tokens_seen": 50266112
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004973520561685056,
      "loss": 4.7312,
      "theoretical_loss": 5.273135423980159,
      "tokens_seen": 50331648
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004973420260782347,
      "loss": 4.709,
      "theoretical_loss": 5.272092039705226,
      "tokens_seen": 50397184
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004973319959879639,
      "loss": 4.5308,
      "theoretical_loss": 5.271050390698967,
      "tokens_seen": 50462720
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004973219658976931,
      "loss": 4.6363,
      "theoretical_loss": 5.270010471827446,
      "tokens_seen": 50528256
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004973119358074223,
      "loss": 4.5769,
      "theoretical_loss": 5.268972277978554,
      "tokens_seen": 50593792
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004973019057171515,
      "loss": 4.2669,
      "theoretical_loss": 5.267935804061883,
      "tokens_seen": 50659328
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004972918756268806,
      "loss": 4.8887,
      "theoretical_loss": 5.266901045008607,
      "tokens_seen": 50724864
    },
    {
      "epoch": 0.02,
      "objective/train/advantage_avg": 0.4941946864128113,
      "objective/train/docs_used": 40584,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 4.018439292907715,
      "objective/train/original_loss": 4.018439769744873,
      "objective/train/theoretical_loss": 5.265867995771362,
      "objective/train/tokens_used": 71250400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24433079361915588,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506614446640015,
      "objective/train/weighted_lm_loss": 4.221940994262695,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0427242517471313,
      "theoretical_loss": 5.265867995771362,
      "tokens_seen": 50790400
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004972818455366098,
      "loss": 4.5725,
      "theoretical_loss": 5.265867995771362,
      "tokens_seen": 50790400
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.000497271815446339,
      "loss": 4.3168,
      "theoretical_loss": 5.26483665132413,
      "tokens_seen": 50855936
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004972617853560682,
      "loss": 4.6475,
      "theoretical_loss": 5.263807006662114,
      "tokens_seen": 50921472
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004972517552657974,
      "loss": 4.6026,
      "theoretical_loss": 5.2627790568016355,
      "tokens_seen": 50987008
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004972417251755266,
      "loss": 4.5413,
      "theoretical_loss": 5.2617527967800015,
      "tokens_seen": 51052544
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004972316950852557,
      "loss": 4.5608,
      "theoretical_loss": 5.260728221655401,
      "tokens_seen": 51118080
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004972216649949849,
      "loss": 4.6027,
      "theoretical_loss": 5.259705326506787,
      "tokens_seen": 51183616
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004972116349047142,
      "loss": 4.6334,
      "theoretical_loss": 5.258684106433764,
      "tokens_seen": 51249152
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004972016048144433,
      "loss": 4.6029,
      "theoretical_loss": 5.257664556556469,
      "tokens_seen": 51314688
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004971915747241726,
      "loss": 4.6729,
      "theoretical_loss": 5.256646672015468,
      "tokens_seen": 51380224
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004971815446339017,
      "loss": 4.268,
      "theoretical_loss": 5.255630447971641,
      "tokens_seen": 51445760
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004971715145436309,
      "loss": 4.7965,
      "theoretical_loss": 5.25461587960607,
      "tokens_seen": 51511296
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004971614844533601,
      "loss": 4.5971,
      "theoretical_loss": 5.253602962119929,
      "tokens_seen": 51576832
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004971514543630893,
      "loss": 4.7953,
      "theoretical_loss": 5.252591690734379,
      "tokens_seen": 51642368
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004971414242728185,
      "loss": 4.2577,
      "theoretical_loss": 5.251582060690451,
      "tokens_seen": 51707904
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004971313941825477,
      "loss": 4.5668,
      "theoretical_loss": 5.250574067248951,
      "tokens_seen": 51773440
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004971213640922768,
      "loss": 4.8581,
      "theoretical_loss": 5.249567705690339,
      "tokens_seen": 51838976
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.000497111334002006,
      "loss": 4.0796,
      "theoretical_loss": 5.248562971314636,
      "tokens_seen": 51904512
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004971013039117352,
      "loss": 4.6666,
      "theoretical_loss": 5.247559859441305,
      "tokens_seen": 51970048
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004970912738214644,
      "loss": 4.8388,
      "theoretical_loss": 5.2465583654091565,
      "tokens_seen": 52035584
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004970812437311936,
      "loss": 4.5071,
      "theoretical_loss": 5.245558484576243,
      "tokens_seen": 52101120
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004970712136409227,
      "loss": 4.7291,
      "theoretical_loss": 5.244560212319749,
      "tokens_seen": 52166656
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004970611835506519,
      "loss": 4.3623,
      "theoretical_loss": 5.243563544035896,
      "tokens_seen": 52232192
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004970511534603811,
      "loss": 4.6702,
      "theoretical_loss": 5.242568475139834,
      "tokens_seen": 52297728
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004970411233701103,
      "loss": 4.6831,
      "theoretical_loss": 5.241575001065544,
      "tokens_seen": 52363264
    },
    {
      "epoch": 0.02,
      "objective/train/advantage_avg": 0.4946448504924774,
      "objective/train/docs_used": 41941,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 4.303009986877441,
      "objective/train/original_loss": 4.303009986877441,
      "objective/train/theoretical_loss": 5.240583117265738,
      "objective/train/tokens_used": 72888800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24472913146018982,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050708532333374,
      "objective/train/weighted_lm_loss": 4.52082633972168,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.044721245765686,
      "theoretical_loss": 5.240583117265738,
      "tokens_seen": 52428800
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004970310932798396,
      "loss": 4.6315,
      "theoretical_loss": 5.240583117265738,
      "tokens_seen": 52428800
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004970210631895686,
      "loss": 4.6174,
      "theoretical_loss": 5.239592819211753,
      "tokens_seen": 52494336
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004970110330992979,
      "loss": 4.6364,
      "theoretical_loss": 5.238604102393458,
      "tokens_seen": 52559872
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004970010030090271,
      "loss": 4.4603,
      "theoretical_loss": 5.237616962319154,
      "tokens_seen": 52625408
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004969909729187563,
      "loss": 4.8385,
      "theoretical_loss": 5.236631394515473,
      "tokens_seen": 52690944
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004969809428284855,
      "loss": 4.6938,
      "theoretical_loss": 5.235647394527281,
      "tokens_seen": 52756480
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004969709127382147,
      "loss": 4.4751,
      "theoretical_loss": 5.234664957917584,
      "tokens_seen": 52822016
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004969608826479438,
      "loss": 4.1713,
      "theoretical_loss": 5.2336840802674285,
      "tokens_seen": 52887552
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.000496950852557673,
      "loss": 4.4198,
      "theoretical_loss": 5.232704757175809,
      "tokens_seen": 52953088
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004969408224674022,
      "loss": 4.6276,
      "theoretical_loss": 5.23172698425957,
      "tokens_seen": 53018624
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004969307923771314,
      "loss": 4.5722,
      "theoretical_loss": 5.230750757153314,
      "tokens_seen": 53084160
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004969207622868606,
      "loss": 4.6475,
      "theoretical_loss": 5.2297760715093045,
      "tokens_seen": 53149696
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004969107321965897,
      "loss": 4.5503,
      "theoretical_loss": 5.228802922997378,
      "tokens_seen": 53215232
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.000496900702106319,
      "loss": 4.5429,
      "theoretical_loss": 5.2278313073048475,
      "tokens_seen": 53280768
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004968906720160481,
      "loss": 4.5053,
      "theoretical_loss": 5.22686122013641,
      "tokens_seen": 53346304
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004968806419257774,
      "loss": 4.4878,
      "theoretical_loss": 5.22589265721406,
      "tokens_seen": 53411840
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004968706118355065,
      "loss": 4.5113,
      "theoretical_loss": 5.224925614276991,
      "tokens_seen": 53477376
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004968605817452358,
      "loss": 4.4335,
      "theoretical_loss": 5.223960087081515,
      "tokens_seen": 53542912
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004968505516549649,
      "loss": 4.4968,
      "theoretical_loss": 5.222996071400964,
      "tokens_seen": 53608448
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004968405215646941,
      "loss": 4.5401,
      "theoretical_loss": 5.2220335630256045,
      "tokens_seen": 53673984
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004968304914744233,
      "loss": 4.5543,
      "theoretical_loss": 5.221072557762554,
      "tokens_seen": 53739520
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004968204613841525,
      "loss": 4.4579,
      "theoretical_loss": 5.220113051435683,
      "tokens_seen": 53805056
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004968104312938817,
      "loss": 4.6697,
      "theoretical_loss": 5.2191550398855355,
      "tokens_seen": 53870592
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004968004012036108,
      "loss": 4.3856,
      "theoretical_loss": 5.21819851896924,
      "tokens_seen": 53936128
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.00049679037111334,
      "loss": 4.7186,
      "theoretical_loss": 5.2172434845604245,
      "tokens_seen": 54001664
    },
    {
      "epoch": 0.02,
      "objective/train/advantage_avg": 0.49444031715393066,
      "objective/train/docs_used": 42708,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 4.476953029632568,
      "objective/train/original_loss": 4.476953029632568,
      "objective/train/theoretical_loss": 5.2162899325491265,
      "objective/train/tokens_used": 74527200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24455030262470245,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506871938705444,
      "objective/train/weighted_lm_loss": 4.703628063201904,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0396549701690674,
      "theoretical_loss": 5.2162899325491265,
      "tokens_seen": 54067200
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004967803410230692,
      "loss": 4.4755,
      "theoretical_loss": 5.2162899325491265,
      "tokens_seen": 54067200
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004967703109327984,
      "loss": 4.6205,
      "theoretical_loss": 5.215337858841714,
      "tokens_seen": 54132736
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004967602808425276,
      "loss": 4.2445,
      "theoretical_loss": 5.214387259360796,
      "tokens_seen": 54198272
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004967502507522568,
      "loss": 4.3588,
      "theoretical_loss": 5.213438130045141,
      "tokens_seen": 54263808
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004967402206619859,
      "loss": 4.3361,
      "theoretical_loss": 5.212490466849594,
      "tokens_seen": 54329344
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004967301905717151,
      "loss": 4.2061,
      "theoretical_loss": 5.211544265744993,
      "tokens_seen": 54394880
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004967201604814444,
      "loss": 4.5445,
      "theoretical_loss": 5.210599522718084,
      "tokens_seen": 54460416
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004967101303911735,
      "loss": 4.4089,
      "theoretical_loss": 5.209656233771442,
      "tokens_seen": 54525952
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004967001003009028,
      "loss": 4.5317,
      "theoretical_loss": 5.2087143949233905,
      "tokens_seen": 54591488
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.000496690070210632,
      "loss": 4.3417,
      "theoretical_loss": 5.207774002207918,
      "tokens_seen": 54657024
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004966800401203611,
      "loss": 4.4835,
      "theoretical_loss": 5.206835051674599,
      "tokens_seen": 54722560
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004966700100300903,
      "loss": 4.2474,
      "theoretical_loss": 5.205897539388513,
      "tokens_seen": 54788096
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004966599799398195,
      "loss": 4.246,
      "theoretical_loss": 5.204961461430168,
      "tokens_seen": 54853632
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004966499498495487,
      "loss": 4.6727,
      "theoretical_loss": 5.2040268138954175,
      "tokens_seen": 54919168
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004966399197592779,
      "loss": 4.2236,
      "theoretical_loss": 5.203093592895387,
      "tokens_seen": 54984704
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.000496629889669007,
      "loss": 4.3918,
      "theoretical_loss": 5.2021617945563925,
      "tokens_seen": 55050240
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004966198595787362,
      "loss": 4.4632,
      "theoretical_loss": 5.201231415019864,
      "tokens_seen": 55115776
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004966098294884654,
      "loss": 4.4291,
      "theoretical_loss": 5.200302450442269,
      "tokens_seen": 55181312
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004965997993981946,
      "loss": 4.6119,
      "theoretical_loss": 5.199374896995037,
      "tokens_seen": 55246848
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004965897693079238,
      "loss": 4.2445,
      "theoretical_loss": 5.198448750864481,
      "tokens_seen": 55312384
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004965797392176529,
      "loss": 4.5173,
      "theoretical_loss": 5.197524008251724,
      "tokens_seen": 55377920
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004965697091273821,
      "loss": 4.5509,
      "theoretical_loss": 5.196600665372626,
      "tokens_seen": 55443456
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004965596790371113,
      "loss": 4.4482,
      "theoretical_loss": 5.195678718457707,
      "tokens_seen": 55508992
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004965496489468405,
      "loss": 4.37,
      "theoretical_loss": 5.194758163752068,
      "tokens_seen": 55574528
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004965396188565698,
      "loss": 4.3796,
      "theoretical_loss": 5.19383899751533,
      "tokens_seen": 55640064
    },
    {
      "epoch": 0.02,
      "objective/train/advantage_avg": 0.49307334423065186,
      "objective/train/docs_used": 44116,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 4.753168106079102,
      "objective/train/original_loss": 4.753168106079102,
      "objective/train/theoretical_loss": 5.192921216021549,
      "objective/train/tokens_used": 76165600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24322925508022308,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505436658859253,
      "objective/train/weighted_lm_loss": 4.993282318115234,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 5.192921216021549,
      "tokens_seen": 55705600
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004965295887662988,
      "loss": 4.5604,
      "theoretical_loss": 5.192921216021549,
      "tokens_seen": 55705600
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004965195586760281,
      "loss": 4.1184,
      "theoretical_loss": 5.19200481555915,
      "tokens_seen": 55771136
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004965095285857573,
      "loss": 3.9914,
      "theoretical_loss": 5.19108979243085,
      "tokens_seen": 55836672
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004964994984954865,
      "loss": 4.4661,
      "theoretical_loss": 5.190176142953593,
      "tokens_seen": 55902208
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004964894684052157,
      "loss": 4.5023,
      "theoretical_loss": 5.189263863458471,
      "tokens_seen": 55967744
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004964794383149449,
      "loss": 4.206,
      "theoretical_loss": 5.188352950290662,
      "tokens_seen": 56033280
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.000496469408224674,
      "loss": 4.1254,
      "theoretical_loss": 5.1874433998093465,
      "tokens_seen": 56098816
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004964593781344032,
      "loss": 3.8751,
      "theoretical_loss": 5.186535208387652,
      "tokens_seen": 56164352
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004964493480441324,
      "loss": 4.6588,
      "theoretical_loss": 5.185628372412573,
      "tokens_seen": 56229888
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004964393179538616,
      "loss": 4.5179,
      "theoretical_loss": 5.1847228882849095,
      "tokens_seen": 56295424
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004964292878635908,
      "loss": 4.7043,
      "theoretical_loss": 5.1838187524191905,
      "tokens_seen": 56360960
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.00049641925777332,
      "loss": 4.6698,
      "theoretical_loss": 5.18291596124361,
      "tokens_seen": 56426496
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004964092276830491,
      "loss": 4.0778,
      "theoretical_loss": 5.182014511199963,
      "tokens_seen": 56492032
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004963991975927783,
      "loss": 4.4092,
      "theoretical_loss": 5.181114398743569,
      "tokens_seen": 56557568
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004963891675025075,
      "loss": 4.2628,
      "theoretical_loss": 5.180215620343211,
      "tokens_seen": 56623104
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004963791374122367,
      "loss": 4.5339,
      "theoretical_loss": 5.17931817248107,
      "tokens_seen": 56688640
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004963691073219659,
      "loss": 4.4627,
      "theoretical_loss": 5.178422051652653,
      "tokens_seen": 56754176
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004963590772316951,
      "loss": 4.4031,
      "theoretical_loss": 5.1775272543667334,
      "tokens_seen": 56819712
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004963490471414242,
      "loss": 4.3977,
      "theoretical_loss": 5.17663377714528,
      "tokens_seen": 56885248
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004963390170511535,
      "loss": 4.4641,
      "theoretical_loss": 5.175741616523398,
      "tokens_seen": 56950784
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004963289869608827,
      "loss": 4.2967,
      "theoretical_loss": 5.174850769049258,
      "tokens_seen": 57016320
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004963189568706119,
      "loss": 4.311,
      "theoretical_loss": 5.173961231284037,
      "tokens_seen": 57081856
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.000496308926780341,
      "loss": 4.3563,
      "theoretical_loss": 5.1730729998018505,
      "tokens_seen": 57147392
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004962988966900702,
      "loss": 4.2635,
      "theoretical_loss": 5.172186071189692,
      "tokens_seen": 57212928
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004962888665997994,
      "loss": 4.4541,
      "theoretical_loss": 5.171300442047371,
      "tokens_seen": 57278464
    },
    {
      "epoch": 0.02,
      "objective/train/advantage_avg": 0.4938618242740631,
      "objective/train/docs_used": 44698,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.721799850463867,
      "objective/train/original_loss": 3.721799612045288,
      "objective/train/theoretical_loss": 5.1704161089874425,
      "objective/train/tokens_used": 77804000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2439718395471573,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506263971328735,
      "objective/train/weighted_lm_loss": 3.9098925590515137,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0449823141098022,
      "theoretical_loss": 5.1704161089874425,
      "tokens_seen": 57344000
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004962788365095286,
      "loss": 4.3096,
      "theoretical_loss": 5.1704161089874425,
      "tokens_seen": 57344000
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004962688064192578,
      "loss": 4.2482,
      "theoretical_loss": 5.169533068635155,
      "tokens_seen": 57409536
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.000496258776328987,
      "loss": 4.6035,
      "theoretical_loss": 5.168651317628383,
      "tokens_seen": 57475072
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004962487462387161,
      "loss": 4.6162,
      "theoretical_loss": 5.167770852617565,
      "tokens_seen": 57540608
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004962387161484453,
      "loss": 4.3426,
      "theoretical_loss": 5.1668916702656436,
      "tokens_seen": 57606144
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004962286860581746,
      "loss": 4.3866,
      "theoretical_loss": 5.166013767248007,
      "tokens_seen": 57671680
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004962186559679037,
      "loss": 4.4506,
      "theoretical_loss": 5.165137140252424,
      "tokens_seen": 57737216
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.000496208625877633,
      "loss": 4.3336,
      "theoretical_loss": 5.1642617859789866,
      "tokens_seen": 57802752
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004961985957873621,
      "loss": 4.8221,
      "theoretical_loss": 5.1633877011400475,
      "tokens_seen": 57868288
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004961885656970913,
      "loss": 4.3566,
      "theoretical_loss": 5.162514882460167,
      "tokens_seen": 57933824
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004961785356068205,
      "loss": 4.6184,
      "theoretical_loss": 5.161643326676048,
      "tokens_seen": 57999360
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004961685055165497,
      "loss": 4.3034,
      "theoretical_loss": 5.160773030536475,
      "tokens_seen": 58064896
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004961584754262789,
      "loss": 4.333,
      "theoretical_loss": 5.15990399080227,
      "tokens_seen": 58130432
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004961484453360081,
      "loss": 4.2409,
      "theoretical_loss": 5.159036204246215,
      "tokens_seen": 58195968
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004961384152457372,
      "loss": 4.4203,
      "theoretical_loss": 5.158169667653005,
      "tokens_seen": 58261504
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004961283851554664,
      "loss": 4.3096,
      "theoretical_loss": 5.1573043778191945,
      "tokens_seen": 58327040
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004961183550651956,
      "loss": 4.1037,
      "theoretical_loss": 5.1564403315531315,
      "tokens_seen": 58392576
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004961083249749248,
      "loss": 4.5979,
      "theoretical_loss": 5.155577525674904,
      "tokens_seen": 58458112
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.000496098294884654,
      "loss": 4.5338,
      "theoretical_loss": 5.15471595701629,
      "tokens_seen": 58523648
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004960882647943831,
      "loss": 4.6439,
      "theoretical_loss": 5.153855622420689,
      "tokens_seen": 58589184
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004960782347041123,
      "loss": 4.4228,
      "theoretical_loss": 5.152996518743078,
      "tokens_seen": 58654720
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004960682046138415,
      "loss": 4.4668,
      "theoretical_loss": 5.152138642849951,
      "tokens_seen": 58720256
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004960581745235707,
      "loss": 4.3785,
      "theoretical_loss": 5.151281991619264,
      "tokens_seen": 58785792
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004960481444333,
      "loss": 4.454,
      "theoretical_loss": 5.150426561940381,
      "tokens_seen": 58851328
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.000496038114343029,
      "loss": 4.3284,
      "theoretical_loss": 5.1495723507140205,
      "tokens_seen": 58916864
    },
    {
      "epoch": 0.02,
      "objective/train/advantage_avg": 0.49369367957115173,
      "objective/train/docs_used": 46033,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.437378168106079,
      "objective/train/original_loss": 3.437378168106079,
      "objective/train/theoretical_loss": 5.148719354852201,
      "objective/train/tokens_used": 79442400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24382761120796204,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506088733673096,
      "objective/train/weighted_lm_loss": 3.610949993133545,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0448545217514038,
      "theoretical_loss": 5.148719354852201,
      "tokens_seen": 58982400
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004960280842527583,
      "loss": 3.9517,
      "theoretical_loss": 5.148719354852201,
      "tokens_seen": 58982400
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004960180541624875,
      "loss": 4.3888,
      "theoretical_loss": 5.147867571278184,
      "tokens_seen": 59047936
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004960080240722167,
      "loss": 4.5351,
      "theoretical_loss": 5.147016996926427,
      "tokens_seen": 59113472
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004959979939819459,
      "loss": 4.3233,
      "theoretical_loss": 5.146167628742525,
      "tokens_seen": 59179008
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004959879638916751,
      "loss": 4.379,
      "theoretical_loss": 5.145319463683162,
      "tokens_seen": 59244544
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004959779338014042,
      "loss": 4.5175,
      "theoretical_loss": 5.144472498716053,
      "tokens_seen": 59310080
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004959679037111334,
      "loss": 4.2339,
      "theoretical_loss": 5.143626730819898,
      "tokens_seen": 59375616
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004959578736208626,
      "loss": 4.1974,
      "theoretical_loss": 5.142782156984327,
      "tokens_seen": 59441152
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004959478435305918,
      "loss": 4.1546,
      "theoretical_loss": 5.141938774209849,
      "tokens_seen": 59506688
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.000495937813440321,
      "loss": 4.4953,
      "theoretical_loss": 5.1410965795077965,
      "tokens_seen": 59572224
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004959277833500501,
      "loss": 4.5032,
      "theoretical_loss": 5.140255569900284,
      "tokens_seen": 59637760
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004959177532597793,
      "loss": 4.5296,
      "theoretical_loss": 5.13941574242015,
      "tokens_seen": 59703296
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004959077231695085,
      "loss": 4.4584,
      "theoretical_loss": 5.138577094110906,
      "tokens_seen": 59768832
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004958976930792377,
      "loss": 4.2329,
      "theoretical_loss": 5.137739622026694,
      "tokens_seen": 59834368
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004958876629889669,
      "loss": 4.1395,
      "theoretical_loss": 5.136903323232229,
      "tokens_seen": 59899904
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.000495877632898696,
      "loss": 4.4248,
      "theoretical_loss": 5.13606819480275,
      "tokens_seen": 59965440
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004958676028084253,
      "loss": 4.5579,
      "theoretical_loss": 5.135234233823978,
      "tokens_seen": 60030976
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004958575727181544,
      "loss": 4.2691,
      "theoretical_loss": 5.134401437392057,
      "tokens_seen": 60096512
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004958475426278837,
      "loss": 4.5295,
      "theoretical_loss": 5.133569802613515,
      "tokens_seen": 60162048
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004958375125376129,
      "loss": 4.4688,
      "theoretical_loss": 5.132739326605209,
      "tokens_seen": 60227584
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004958274824473421,
      "loss": 4.2877,
      "theoretical_loss": 5.131910006494278,
      "tokens_seen": 60293120
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004958174523570712,
      "loss": 4.5165,
      "theoretical_loss": 5.131081839418099,
      "tokens_seen": 60358656
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004958074222668004,
      "loss": 4.6061,
      "theoretical_loss": 5.130254822524234,
      "tokens_seen": 60424192
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004957973921765296,
      "loss": 3.986,
      "theoretical_loss": 5.129428952970388,
      "tokens_seen": 60489728
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004957873620862588,
      "loss": 4.4992,
      "theoretical_loss": 5.128604227924356,
      "tokens_seen": 60555264
    },
    {
      "epoch": 0.02,
      "objective/train/advantage_avg": 0.49483078718185425,
      "objective/train/docs_used": 46630,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.756608247756958,
      "objective/train/original_loss": 3.7566089630126953,
      "objective/train/theoretical_loss": 5.127780644563984,
      "objective/train/tokens_used": 81080800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24492616951465607,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.05072820186615,
      "objective/train/weighted_lm_loss": 3.9469802379608154,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0338950157165527,
      "theoretical_loss": 5.127780644563984,
      "tokens_seen": 60620800
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.000495777331995988,
      "loss": 4.0839,
      "theoretical_loss": 5.127780644563984,
      "tokens_seen": 60620800
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004957673019057172,
      "loss": 4.1284,
      "theoretical_loss": 5.126958200077116,
      "tokens_seen": 60686336
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004957572718154463,
      "loss": 3.9825,
      "theoretical_loss": 5.126136891661549,
      "tokens_seen": 60751872
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004957472417251755,
      "loss": 4.3157,
      "theoretical_loss": 5.12531671652499,
      "tokens_seen": 60817408
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004957372116349047,
      "loss": 4.4569,
      "theoretical_loss": 5.124497671885009,
      "tokens_seen": 60882944
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004957271815446339,
      "loss": 4.083,
      "theoretical_loss": 5.1236797549689905,
      "tokens_seen": 60948480
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004957171514543631,
      "loss": 4.5398,
      "theoretical_loss": 5.122862963014094,
      "tokens_seen": 61014016
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004957071213640923,
      "loss": 4.1554,
      "theoretical_loss": 5.122047293267204,
      "tokens_seen": 61079552
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004956970912738214,
      "loss": 4.3156,
      "theoretical_loss": 5.121232742984887,
      "tokens_seen": 61145088
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004956870611835507,
      "loss": 4.5212,
      "theoretical_loss": 5.120419309433351,
      "tokens_seen": 61210624
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004956770310932798,
      "loss": 4.2125,
      "theoretical_loss": 5.119606989888393,
      "tokens_seen": 61276160
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004956670010030091,
      "loss": 4.3805,
      "theoretical_loss": 5.118795781635365,
      "tokens_seen": 61341696
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004956569709127383,
      "loss": 4.4767,
      "theoretical_loss": 5.11798568196912,
      "tokens_seen": 61407232
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004956469408224674,
      "loss": 4.4162,
      "theoretical_loss": 5.117176688193979,
      "tokens_seen": 61472768
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004956369107321966,
      "loss": 4.2819,
      "theoretical_loss": 5.1163687976236805,
      "tokens_seen": 61538304
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004956268806419258,
      "loss": 4.4022,
      "theoretical_loss": 5.115562007581341,
      "tokens_seen": 61603840
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.000495616850551655,
      "loss": 4.1331,
      "theoretical_loss": 5.114756315399407,
      "tokens_seen": 61669376
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004956068204613842,
      "loss": 4.3285,
      "theoretical_loss": 5.113951718419624,
      "tokens_seen": 61734912
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004955967903711133,
      "loss": 4.4688,
      "theoretical_loss": 5.113148213992981,
      "tokens_seen": 61800448
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004955867602808425,
      "loss": 4.1892,
      "theoretical_loss": 5.112345799479678,
      "tokens_seen": 61865984
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004955767301905717,
      "loss": 4.227,
      "theoretical_loss": 5.111544472249082,
      "tokens_seen": 61931520
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004955667001003009,
      "loss": 4.1764,
      "theoretical_loss": 5.110744229679679,
      "tokens_seen": 61997056
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004955566700100301,
      "loss": 4.416,
      "theoretical_loss": 5.1099450691590445,
      "tokens_seen": 62062592
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004955466399197592,
      "loss": 4.4475,
      "theoretical_loss": 5.109146988083792,
      "tokens_seen": 62128128
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004955366098294884,
      "loss": 4.5097,
      "theoretical_loss": 5.108349983859538,
      "tokens_seen": 62193664
    },
    {
      "epoch": 0.02,
      "objective/train/advantage_avg": 0.49333709478378296,
      "objective/train/docs_used": 47896,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 4.239388465881348,
      "objective/train/original_loss": 4.239388465881348,
      "objective/train/theoretical_loss": 5.107554053900861,
      "objective/train/tokens_used": 82719200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24344682693481445,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505712032318115,
      "objective/train/weighted_lm_loss": 4.453501224517822,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0363596677780151,
      "theoretical_loss": 5.107554053900861,
      "tokens_seen": 62259200
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004955265797392177,
      "loss": 4.3925,
      "theoretical_loss": 5.107554053900861,
      "tokens_seen": 62259200
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004955165496489468,
      "loss": 4.3169,
      "theoretical_loss": 5.106759195631259,
      "tokens_seen": 62324736
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004955065195586761,
      "loss": 4.358,
      "theoretical_loss": 5.105965406483108,
      "tokens_seen": 62390272
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004954964894684052,
      "loss": 4.2898,
      "theoretical_loss": 5.105172683897628,
      "tokens_seen": 62455808
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004954864593781344,
      "loss": 4.2903,
      "theoretical_loss": 5.104381025324839,
      "tokens_seen": 62521344
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004954764292878636,
      "loss": 4.0272,
      "theoretical_loss": 5.103590428223521,
      "tokens_seen": 62586880
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004954663991975928,
      "loss": 4.2074,
      "theoretical_loss": 5.102800890061178,
      "tokens_seen": 62652416
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.000495456369107322,
      "loss": 4.1031,
      "theoretical_loss": 5.102012408313994,
      "tokens_seen": 62717952
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004954463390170512,
      "loss": 4.1956,
      "theoretical_loss": 5.101224980466801,
      "tokens_seen": 62783488
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004954363089267803,
      "loss": 4.2279,
      "theoretical_loss": 5.100438604013036,
      "tokens_seen": 62849024
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004954262788365095,
      "loss": 3.9214,
      "theoretical_loss": 5.0996532764547,
      "tokens_seen": 62914560
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004954162487462387,
      "loss": 4.0007,
      "theoretical_loss": 5.098868995302327,
      "tokens_seen": 62980096
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004954062186559679,
      "loss": 4.0319,
      "theoretical_loss": 5.098085758074939,
      "tokens_seen": 63045632
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004953961885656971,
      "loss": 4.1372,
      "theoretical_loss": 5.097303562300015,
      "tokens_seen": 63111168
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004953861584754263,
      "loss": 4.4267,
      "theoretical_loss": 5.096522405513447,
      "tokens_seen": 63176704
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004953761283851555,
      "loss": 4.2734,
      "theoretical_loss": 5.095742285259507,
      "tokens_seen": 63242240
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004953660982948846,
      "loss": 4.2464,
      "theoretical_loss": 5.094963199090807,
      "tokens_seen": 63307776
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004953560682046139,
      "loss": 4.244,
      "theoretical_loss": 5.094185144568264,
      "tokens_seen": 63373312
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004953460381143431,
      "loss": 4.2373,
      "theoretical_loss": 5.093408119261066,
      "tokens_seen": 63438848
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004953360080240723,
      "loss": 4.1552,
      "theoretical_loss": 5.092632120746626,
      "tokens_seen": 63504384
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004953259779338014,
      "loss": 4.0642,
      "theoretical_loss": 5.091857146610557,
      "tokens_seen": 63569920
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004953159478435306,
      "loss": 4.3112,
      "theoretical_loss": 5.091083194446627,
      "tokens_seen": 63635456
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004953059177532598,
      "loss": 4.3166,
      "theoretical_loss": 5.0903102618567315,
      "tokens_seen": 63700992
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.000495295887662989,
      "loss": 4.2729,
      "theoretical_loss": 5.089538346450848,
      "tokens_seen": 63766528
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004952858575727182,
      "loss": 4.1442,
      "theoretical_loss": 5.088767445847009,
      "tokens_seen": 63832064
    },
    {
      "epoch": 0.02,
      "objective/train/advantage_avg": 0.4939124286174774,
      "objective/train/docs_used": 48551,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.9862353801727295,
      "objective/train/original_loss": 3.9862349033355713,
      "objective/train/theoretical_loss": 5.087997557671262,
      "objective/train/tokens_used": 84357600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24402378499507904,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506317615509033,
      "objective/train/weighted_lm_loss": 4.18788480758667,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 5.087997557671262,
      "tokens_seen": 63897600
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004952758274824474,
      "loss": 4.3147,
      "theoretical_loss": 5.087997557671262,
      "tokens_seen": 63897600
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004952657973921765,
      "loss": 4.2262,
      "theoretical_loss": 5.087228679557634,
      "tokens_seen": 63963136
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004952557673019057,
      "loss": 4.1329,
      "theoretical_loss": 5.0864608091481,
      "tokens_seen": 64028672
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004952457372116349,
      "loss": 4.1592,
      "theoretical_loss": 5.085693944092547,
      "tokens_seen": 64094208
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004952357071213641,
      "loss": 4.2147,
      "theoretical_loss": 5.084928082048736,
      "tokens_seen": 64159744
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004952256770310933,
      "loss": 4.2102,
      "theoretical_loss": 5.084163220682272,
      "tokens_seen": 64225280
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004952156469408225,
      "loss": 4.0707,
      "theoretical_loss": 5.083399357666567,
      "tokens_seen": 64290816
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004952056168505516,
      "loss": 4.3884,
      "theoretical_loss": 5.082636490682806,
      "tokens_seen": 64356352
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004951955867602809,
      "loss": 4.2642,
      "theoretical_loss": 5.081874617419919,
      "tokens_seen": 64421888
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.00049518555667001,
      "loss": 4.1196,
      "theoretical_loss": 5.081113735574535,
      "tokens_seen": 64487424
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004951755265797393,
      "loss": 4.2007,
      "theoretical_loss": 5.080353842850963,
      "tokens_seen": 64552960
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004951654964894685,
      "loss": 4.0549,
      "theoretical_loss": 5.079594936961146,
      "tokens_seen": 64618496
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004951554663991976,
      "loss": 3.7922,
      "theoretical_loss": 5.078837015624638,
      "tokens_seen": 64684032
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004951454363089268,
      "loss": 4.1245,
      "theoretical_loss": 5.078080076568562,
      "tokens_seen": 64749568
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.000495135406218656,
      "loss": 3.9915,
      "theoretical_loss": 5.077324117527585,
      "tokens_seen": 64815104
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004951253761283852,
      "loss": 4.063,
      "theoretical_loss": 5.076569136243883,
      "tokens_seen": 64880640
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004951153460381144,
      "loss": 4.1592,
      "theoretical_loss": 5.075815130467101,
      "tokens_seen": 64946176
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004951053159478435,
      "loss": 4.109,
      "theoretical_loss": 5.075062097954335,
      "tokens_seen": 65011712
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004950952858575727,
      "loss": 4.1862,
      "theoretical_loss": 5.0743100364700915,
      "tokens_seen": 65077248
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004950852557673019,
      "loss": 4.1673,
      "theoretical_loss": 5.07355894378625,
      "tokens_seen": 65142784
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004950752256770311,
      "loss": 4.0981,
      "theoretical_loss": 5.072808817682045,
      "tokens_seen": 65208320
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004950651955867603,
      "loss": 3.8126,
      "theoretical_loss": 5.072059655944021,
      "tokens_seen": 65273856
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004950551654964894,
      "loss": 4.2484,
      "theoretical_loss": 5.071311456366011,
      "tokens_seen": 65339392
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004950451354062186,
      "loss": 4.062,
      "theoretical_loss": 5.070564216749103,
      "tokens_seen": 65404928
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004950351053159479,
      "loss": 4.3331,
      "theoretical_loss": 5.069817934901602,
      "tokens_seen": 65470464
    },
    {
      "epoch": 0.02,
      "objective/train/advantage_avg": 0.4943424165248871,
      "objective/train/docs_used": 49128,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 4.048988342285156,
      "objective/train/original_loss": 4.048988342285156,
      "objective/train/theoretical_loss": 5.069072608639006,
      "objective/train/tokens_used": 85996000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2444460093975067,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506768226623535,
      "objective/train/weighted_lm_loss": 4.253940582275391,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0408108234405518,
      "theoretical_loss": 5.069072608639006,
      "tokens_seen": 65536000
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.000495025075225677,
      "loss": 4.0629,
      "theoretical_loss": 5.069072608639006,
      "tokens_seen": 65536000
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004950150451354063,
      "loss": 4.0976,
      "theoretical_loss": 5.0683282357839765,
      "tokens_seen": 65601536
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004950050150451354,
      "loss": 4.3642,
      "theoretical_loss": 5.067584814166302,
      "tokens_seen": 65667072
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004949949849548646,
      "loss": 3.8511,
      "theoretical_loss": 5.066842341622871,
      "tokens_seen": 65732608
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004949849548645938,
      "loss": 3.9856,
      "theoretical_loss": 5.066100815997643,
      "tokens_seen": 65798144
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.000494974924774323,
      "loss": 4.1402,
      "theoretical_loss": 5.065360235141615,
      "tokens_seen": 65863680
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004949648946840522,
      "loss": 4.0803,
      "theoretical_loss": 5.0646205969127935,
      "tokens_seen": 65929216
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004949548645937814,
      "loss": 4.0995,
      "theoretical_loss": 5.063881899176165,
      "tokens_seen": 65994752
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004949448345035105,
      "loss": 3.9438,
      "theoretical_loss": 5.063144139803664,
      "tokens_seen": 66060288
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004949348044132397,
      "loss": 3.9722,
      "theoretical_loss": 5.062407316674149,
      "tokens_seen": 66125824
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004949247743229689,
      "loss": 3.9218,
      "theoretical_loss": 5.061671427673364,
      "tokens_seen": 66191360
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004949147442326981,
      "loss": 4.0186,
      "theoretical_loss": 5.060936470693919,
      "tokens_seen": 66256896
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004949047141424273,
      "loss": 3.8859,
      "theoretical_loss": 5.060202443635257,
      "tokens_seen": 66322432
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004948946840521565,
      "loss": 4.0886,
      "theoretical_loss": 5.05946934440362,
      "tokens_seen": 66387968
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004948846539618856,
      "loss": 4.3486,
      "theoretical_loss": 5.058737170912032,
      "tokens_seen": 66453504
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004948746238716148,
      "loss": 3.9679,
      "theoretical_loss": 5.058005921080258,
      "tokens_seen": 66519040
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.000494864593781344,
      "loss": 3.8967,
      "theoretical_loss": 5.057275592834783,
      "tokens_seen": 66584576
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004948545636910733,
      "loss": 3.9073,
      "theoretical_loss": 5.0565461841087815,
      "tokens_seen": 66650112
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004948445336008024,
      "loss": 3.6229,
      "theoretical_loss": 5.055817692842091,
      "tokens_seen": 66715648
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004948345035105316,
      "loss": 4.2866,
      "theoretical_loss": 5.05509011698118,
      "tokens_seen": 66781184
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004948244734202607,
      "loss": 4.0394,
      "theoretical_loss": 5.054363454479127,
      "tokens_seen": 66846720
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.00049481444332999,
      "loss": 3.9762,
      "theoretical_loss": 5.053637703295585,
      "tokens_seen": 66912256
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004948044132397192,
      "loss": 4.0162,
      "theoretical_loss": 5.0529128613967575,
      "tokens_seen": 66977792
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004947943831494484,
      "loss": 4.046,
      "theoretical_loss": 5.052188926755374,
      "tokens_seen": 67043328
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004947843530591776,
      "loss": 4.3418,
      "theoretical_loss": 5.051465897350656,
      "tokens_seen": 67108864
    },
    {
      "epoch": 0.02,
      "objective/train/advantage_avg": 0.4955059587955475,
      "objective/train/docs_used": 50658,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.570134162902832,
      "objective/train/original_loss": 3.570133686065674,
      "objective/train/theoretical_loss": 5.0507437711683,
      "objective/train/tokens_used": 87634400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2456275373697281,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507991313934326,
      "objective/train/weighted_lm_loss": 3.751565933227539,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0338950157165527,
      "theoretical_loss": 5.0507437711683,
      "tokens_seen": 67174400
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004947743229689067,
      "loss": 3.9164,
      "theoretical_loss": 5.0507437711683,
      "tokens_seen": 67174400
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004947642928786359,
      "loss": 3.9343,
      "theoretical_loss": 5.050022546200437,
      "tokens_seen": 67239936
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004947542627883651,
      "loss": 4.0572,
      "theoretical_loss": 5.049302220445618,
      "tokens_seen": 67305472
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004947442326980943,
      "loss": 4.107,
      "theoretical_loss": 5.048582791908779,
      "tokens_seen": 67371008
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004947342026078235,
      "loss": 3.9679,
      "theoretical_loss": 5.04786425860122,
      "tokens_seen": 67436544
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004947241725175527,
      "loss": 4.1725,
      "theoretical_loss": 5.047146618540575,
      "tokens_seen": 67502080
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004947141424272818,
      "loss": 4.2684,
      "theoretical_loss": 5.046429869750787,
      "tokens_seen": 67567616
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004947041123370111,
      "loss": 4.2291,
      "theoretical_loss": 5.045714010262081,
      "tokens_seen": 67633152
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004946940822467402,
      "loss": 4.1668,
      "theoretical_loss": 5.044999038110941,
      "tokens_seen": 67698688
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004946840521564695,
      "loss": 4.079,
      "theoretical_loss": 5.044284951340078,
      "tokens_seen": 67764224
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004946740220661987,
      "loss": 3.952,
      "theoretical_loss": 5.043571747998412,
      "tokens_seen": 67829760
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004946639919759278,
      "loss": 4.1665,
      "theoretical_loss": 5.042859426141039,
      "tokens_seen": 67895296
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.000494653961885657,
      "loss": 3.8341,
      "theoretical_loss": 5.042147983829212,
      "tokens_seen": 67960832
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004946439317953862,
      "loss": 3.959,
      "theoretical_loss": 5.041437419130309,
      "tokens_seen": 68026368
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004946339017051154,
      "loss": 4.0019,
      "theoretical_loss": 5.040727730117813,
      "tokens_seen": 68091904
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004946238716148446,
      "loss": 4.1647,
      "theoretical_loss": 5.040018914871285,
      "tokens_seen": 68157440
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004946138415245737,
      "loss": 3.9022,
      "theoretical_loss": 5.039310971476339,
      "tokens_seen": 68222976
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004946038114343029,
      "loss": 4.3287,
      "theoretical_loss": 5.038603898024617,
      "tokens_seen": 68288512
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004945937813440321,
      "loss": 4.0737,
      "theoretical_loss": 5.037897692613765,
      "tokens_seen": 68354048
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004945837512537613,
      "loss": 4.2778,
      "theoretical_loss": 5.037192353347406,
      "tokens_seen": 68419584
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004945737211634905,
      "loss": 3.8717,
      "theoretical_loss": 5.0364878783351195,
      "tokens_seen": 68485120
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004945636910732196,
      "loss": 3.9252,
      "theoretical_loss": 5.035784265692413,
      "tokens_seen": 68550656
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004945536609829488,
      "loss": 3.9759,
      "theoretical_loss": 5.035081513540704,
      "tokens_seen": 68616192
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004945436308926781,
      "loss": 4.0888,
      "theoretical_loss": 5.034379620007284,
      "tokens_seen": 68681728
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004945336008024072,
      "loss": 4.0504,
      "theoretical_loss": 5.03367858322531,
      "tokens_seen": 68747264
    },
    {
      "epoch": 0.02,
      "objective/train/advantage_avg": 0.4945206642150879,
      "objective/train/docs_used": 51428,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.251833200454712,
      "objective/train/original_loss": 3.251833438873291,
      "objective/train/theoretical_loss": 5.032978401333766,
      "objective/train/tokens_used": 89272800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24466046690940857,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050695776939392,
      "objective/train/weighted_lm_loss": 3.416572093963623,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0396549701690674,
      "theoretical_loss": 5.032978401333766,
      "tokens_seen": 68812800
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004945235707121365,
      "loss": 3.8943,
      "theoretical_loss": 5.032978401333766,
      "tokens_seen": 68812800
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004945135406218656,
      "loss": 4.114,
      "theoretical_loss": 5.0322790724774515,
      "tokens_seen": 68878336
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004945035105315948,
      "loss": 4.0012,
      "theoretical_loss": 5.031580594806948,
      "tokens_seen": 68943872
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.000494493480441324,
      "loss": 3.9671,
      "theoretical_loss": 5.030882966478602,
      "tokens_seen": 69009408
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004944834503510532,
      "loss": 4.0944,
      "theoretical_loss": 5.0301861856544985,
      "tokens_seen": 69074944
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004944734202607824,
      "loss": 4.0814,
      "theoretical_loss": 5.029490250502439,
      "tokens_seen": 69140480
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004944633901705116,
      "loss": 4.0359,
      "theoretical_loss": 5.028795159195919,
      "tokens_seen": 69206016
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004944533600802407,
      "loss": 4.0729,
      "theoretical_loss": 5.0281009099141,
      "tokens_seen": 69271552
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004944433299899699,
      "loss": 4.2058,
      "theoretical_loss": 5.027407500841795,
      "tokens_seen": 69337088
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004944332998996991,
      "loss": 4.0286,
      "theoretical_loss": 5.026714930169438,
      "tokens_seen": 69402624
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004944232698094283,
      "loss": 4.1885,
      "theoretical_loss": 5.026023196093064,
      "tokens_seen": 69468160
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004944132397191575,
      "loss": 4.0135,
      "theoretical_loss": 5.02533229681429,
      "tokens_seen": 69533696
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004944032096288867,
      "loss": 3.981,
      "theoretical_loss": 5.0246422305402865,
      "tokens_seen": 69599232
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004943931795386158,
      "loss": 4.1082,
      "theoretical_loss": 5.023952995483757,
      "tokens_seen": 69664768
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.000494383149448345,
      "loss": 3.9366,
      "theoretical_loss": 5.0232645898629205,
      "tokens_seen": 69730304
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004943731193580742,
      "loss": 4.0215,
      "theoretical_loss": 5.022577011901481,
      "tokens_seen": 69795840
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004943630892678035,
      "loss": 3.9569,
      "theoretical_loss": 5.021890259828613,
      "tokens_seen": 69861376
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004943530591775326,
      "loss": 3.9607,
      "theoretical_loss": 5.021204331878934,
      "tokens_seen": 69926912
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004943430290872618,
      "loss": 4.1765,
      "theoretical_loss": 5.020519226292487,
      "tokens_seen": 69992448
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004943329989969909,
      "loss": 3.9463,
      "theoretical_loss": 5.019834941314716,
      "tokens_seen": 70057984
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004943229689067202,
      "loss": 3.7498,
      "theoretical_loss": 5.0191514751964466,
      "tokens_seen": 70123520
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004943129388164494,
      "loss": 3.7498,
      "theoretical_loss": 5.01846882619386,
      "tokens_seen": 70189056
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004943029087261786,
      "loss": 3.827,
      "theoretical_loss": 5.01778699256848,
      "tokens_seen": 70254592
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004942928786359078,
      "loss": 4.1351,
      "theoretical_loss": 5.017105972587139,
      "tokens_seen": 70320128
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004942828485456369,
      "loss": 4.0903,
      "theoretical_loss": 5.016425764521972,
      "tokens_seen": 70385664
    },
    {
      "epoch": 0.02,
      "objective/train/advantage_avg": 0.49469470977783203,
      "objective/train/docs_used": 52769,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.424621105194092,
      "objective/train/original_loss": 3.42462158203125,
      "objective/train/theoretical_loss": 5.015746366650382,
      "objective/train/tokens_used": 90911200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24478507041931152,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507137775421143,
      "objective/train/weighted_lm_loss": 3.5982704162597656,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0396549701690674,
      "theoretical_loss": 5.015746366650382,
      "tokens_seen": 70451200
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004942728184553661,
      "loss": 3.851,
      "theoretical_loss": 5.015746366650382,
      "tokens_seen": 70451200
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004942627883650953,
      "loss": 4.18,
      "theoretical_loss": 5.0150677772550285,
      "tokens_seen": 70516736
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004942527582748245,
      "loss": 3.8604,
      "theoretical_loss": 5.014389994623801,
      "tokens_seen": 70582272
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004942427281845537,
      "loss": 4.0104,
      "theoretical_loss": 5.013713017049801,
      "tokens_seen": 70647808
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004942326980942828,
      "loss": 4.0433,
      "theoretical_loss": 5.013036842831319,
      "tokens_seen": 70713344
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.000494222668004012,
      "loss": 4.0087,
      "theoretical_loss": 5.012361470271816,
      "tokens_seen": 70778880
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004942126379137412,
      "loss": 4.0436,
      "theoretical_loss": 5.0116868976799065,
      "tokens_seen": 70844416
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004942026078234704,
      "loss": 4.146,
      "theoretical_loss": 5.0110131233693265,
      "tokens_seen": 70909952
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004941925777331996,
      "loss": 3.8508,
      "theoretical_loss": 5.010340145658926,
      "tokens_seen": 70975488
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004941825476429289,
      "loss": 3.865,
      "theoretical_loss": 5.009667962872642,
      "tokens_seen": 71041024
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004941725175526579,
      "loss": 3.818,
      "theoretical_loss": 5.008996573339481,
      "tokens_seen": 71106560
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004941624874623872,
      "loss": 4.1922,
      "theoretical_loss": 5.008325975393495,
      "tokens_seen": 71172096
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004941524573721163,
      "loss": 4.0203,
      "theoretical_loss": 5.00765616737377,
      "tokens_seen": 71237632
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004941424272818456,
      "loss": 4.0068,
      "theoretical_loss": 5.006987147624395,
      "tokens_seen": 71303168
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004941323971915748,
      "loss": 4.0126,
      "theoretical_loss": 5.006318914494451,
      "tokens_seen": 71368704
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004941223671013039,
      "loss": 4.3379,
      "theoretical_loss": 5.00565146633799,
      "tokens_seen": 71434240
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004941123370110331,
      "loss": 4.013,
      "theoretical_loss": 5.004984801514011,
      "tokens_seen": 71499776
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004941023069207623,
      "loss": 3.7059,
      "theoretical_loss": 5.004318918386446,
      "tokens_seen": 71565312
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004940922768304915,
      "loss": 3.9921,
      "theoretical_loss": 5.003653815324136,
      "tokens_seen": 71630848
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004940822467402207,
      "loss": 3.974,
      "theoretical_loss": 5.002989490700816,
      "tokens_seen": 71696384
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004940722166499498,
      "loss": 4.2446,
      "theoretical_loss": 5.002325942895094,
      "tokens_seen": 71761920
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.000494062186559679,
      "loss": 3.8794,
      "theoretical_loss": 5.001663170290431,
      "tokens_seen": 71827456
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004940521564694082,
      "loss": 3.8235,
      "theoretical_loss": 5.0010011712751234,
      "tokens_seen": 71892992
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004940421263791374,
      "loss": 4.1968,
      "theoretical_loss": 5.000339944242283,
      "tokens_seen": 71958528
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004940320962888666,
      "loss": 4.1597,
      "theoretical_loss": 4.9996794875898205,
      "tokens_seen": 72024064
    },
    {
      "epoch": 0.02,
      "objective/train/advantage_avg": 0.49523860216140747,
      "objective/train/docs_used": 53229,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.6649441719055176,
      "objective/train/original_loss": 3.664944648742676,
      "objective/train/theoretical_loss": 4.999019799720424,
      "objective/train/tokens_used": 92549600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24532879889011383,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507709980010986,
      "objective/train/weighted_lm_loss": 3.850813627243042,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 4.999019799720424,
      "tokens_seen": 72089600
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004940220661985958,
      "loss": 4.1341,
      "theoretical_loss": 4.999019799720424,
      "tokens_seen": 72089600
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004940120361083249,
      "loss": 3.8158,
      "theoretical_loss": 4.998360879041542,
      "tokens_seen": 72155136
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004940020060180542,
      "loss": 4.1715,
      "theoretical_loss": 4.997702723965366,
      "tokens_seen": 72220672
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004939919759277834,
      "loss": 4.0753,
      "theoretical_loss": 4.997045332908808,
      "tokens_seen": 72286208
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004939819458375126,
      "loss": 4.3294,
      "theoretical_loss": 4.996388704293487,
      "tokens_seen": 72351744
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004939719157472418,
      "loss": 4.0927,
      "theoretical_loss": 4.995732836545709,
      "tokens_seen": 72417280
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.000493961885656971,
      "loss": 3.6558,
      "theoretical_loss": 4.995077728096446,
      "tokens_seen": 72482816
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004939518555667001,
      "loss": 3.711,
      "theoretical_loss": 4.994423377381324,
      "tokens_seen": 72548352
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004939418254764293,
      "loss": 4.1526,
      "theoretical_loss": 4.993769782840598,
      "tokens_seen": 72613888
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004939317953861585,
      "loss": 3.9014,
      "theoretical_loss": 4.993116942919141,
      "tokens_seen": 72679424
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004939217652958877,
      "loss": 4.2458,
      "theoretical_loss": 4.9924648560664195,
      "tokens_seen": 72744960
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004939117352056169,
      "loss": 3.938,
      "theoretical_loss": 4.99181352073648,
      "tokens_seen": 72810496
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.000493901705115346,
      "loss": 4.1992,
      "theoretical_loss": 4.991162935387935,
      "tokens_seen": 72876032
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004938916750250752,
      "loss": 3.9707,
      "theoretical_loss": 4.990513098483933,
      "tokens_seen": 72941568
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004938816449348044,
      "loss": 4.0937,
      "theoretical_loss": 4.989864008492155,
      "tokens_seen": 73007104
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004938716148445337,
      "loss": 4.015,
      "theoretical_loss": 4.98921566388479,
      "tokens_seen": 73072640
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004938615847542628,
      "loss": 3.9119,
      "theoretical_loss": 4.9885680631385165,
      "tokens_seen": 73138176
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.000493851554663992,
      "loss": 3.7522,
      "theoretical_loss": 4.987921204734489,
      "tokens_seen": 73203712
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004938415245737211,
      "loss": 3.9585,
      "theoretical_loss": 4.987275087158322,
      "tokens_seen": 73269248
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004938314944834504,
      "loss": 3.8053,
      "theoretical_loss": 4.986629708900064,
      "tokens_seen": 73334784
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004938214643931796,
      "loss": 3.9592,
      "theoretical_loss": 4.985985068454193,
      "tokens_seen": 73400320
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004938114343029088,
      "loss": 4.1512,
      "theoretical_loss": 4.985341164319592,
      "tokens_seen": 73465856
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.000493801404212638,
      "loss": 4.1783,
      "theoretical_loss": 4.984697994999531,
      "tokens_seen": 73531392
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004937913741223671,
      "loss": 3.8504,
      "theoretical_loss": 4.984055559001657,
      "tokens_seen": 73596928
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004937813440320963,
      "loss": 4.0787,
      "theoretical_loss": 4.983413854837971,
      "tokens_seen": 73662464
    },
    {
      "epoch": 0.02,
      "objective/train/advantage_avg": 0.4944208264350891,
      "objective/train/docs_used": 53865,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.7699058055877686,
      "objective/train/original_loss": 3.7699058055877686,
      "objective/train/theoretical_loss": 4.982772881024813,
      "objective/train/tokens_used": 94188000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24454523622989655,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506852865219116,
      "objective/train/weighted_lm_loss": 3.960937023162842,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0338950157165527,
      "theoretical_loss": 4.982772881024813,
      "tokens_seen": 73728000
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004937713139418255,
      "loss": 3.9805,
      "theoretical_loss": 4.982772881024813,
      "tokens_seen": 73728000
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004937612838515547,
      "loss": 4.142,
      "theoretical_loss": 4.982132636082847,
      "tokens_seen": 73793536
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004937512537612839,
      "loss": 4.026,
      "theoretical_loss": 4.981493118537046,
      "tokens_seen": 73859072
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.000493741223671013,
      "loss": 3.9153,
      "theoretical_loss": 4.9808543269166705,
      "tokens_seen": 73924608
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004937311935807422,
      "loss": 3.9272,
      "theoretical_loss": 4.980216259755256,
      "tokens_seen": 73990144
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004937211634904714,
      "loss": 4.0449,
      "theoretical_loss": 4.979578915590595,
      "tokens_seen": 74055680
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004937111334002006,
      "loss": 3.822,
      "theoretical_loss": 4.978942292964726,
      "tokens_seen": 74121216
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004937011033099298,
      "loss": 3.9778,
      "theoretical_loss": 4.978306390423909,
      "tokens_seen": 74186752
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004936910732196591,
      "loss": 3.6457,
      "theoretical_loss": 4.977671206518615,
      "tokens_seen": 74252288
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004936810431293881,
      "loss": 3.9133,
      "theoretical_loss": 4.977036739803509,
      "tokens_seen": 74317824
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004936710130391174,
      "loss": 3.9339,
      "theoretical_loss": 4.976402988837435,
      "tokens_seen": 74383360
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004936609829488465,
      "loss": 3.9859,
      "theoretical_loss": 4.9757699521834,
      "tokens_seen": 74448896
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004936509528585758,
      "loss": 3.8749,
      "theoretical_loss": 4.975137628408555,
      "tokens_seen": 74514432
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.000493640922768305,
      "loss": 4.0567,
      "theoretical_loss": 4.9745060160841845,
      "tokens_seen": 74579968
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004936308926780341,
      "loss": 3.7959,
      "theoretical_loss": 4.973875113785686,
      "tokens_seen": 74645504
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004936208625877633,
      "loss": 3.7341,
      "theoretical_loss": 4.9732449200925615,
      "tokens_seen": 74711040
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004936108324974925,
      "loss": 3.9589,
      "theoretical_loss": 4.972615433588394,
      "tokens_seen": 74776576
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004936008024072217,
      "loss": 3.7351,
      "theoretical_loss": 4.971986652860837,
      "tokens_seen": 74842112
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004935907723169509,
      "loss": 3.8683,
      "theoretical_loss": 4.971358576501599,
      "tokens_seen": 74907648
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.00049358074222668,
      "loss": 3.8785,
      "theoretical_loss": 4.970731203106427,
      "tokens_seen": 74973184
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004935707121364092,
      "loss": 3.8532,
      "theoretical_loss": 4.970104531275091,
      "tokens_seen": 75038720
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004935606820461384,
      "loss": 3.9076,
      "theoretical_loss": 4.96947855961137,
      "tokens_seen": 75104256
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004935506519558676,
      "loss": 3.86,
      "theoretical_loss": 4.968853286723041,
      "tokens_seen": 75169792
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004935406218655968,
      "loss": 3.5368,
      "theoretical_loss": 4.968228711221853,
      "tokens_seen": 75235328
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.000493530591775326,
      "loss": 3.5886,
      "theoretical_loss": 4.967604831723525,
      "tokens_seen": 75300864
    },
    {
      "epoch": 0.02,
      "objective/train/advantage_avg": 0.4944094717502594,
      "objective/train/docs_used": 54881,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.303663730621338,
      "objective/train/original_loss": 3.303663492202759,
      "objective/train/theoretical_loss": 4.966981646847723,
      "objective/train/tokens_used": 95826400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2445027083158493,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050683856010437,
      "objective/train/weighted_lm_loss": 3.470911741256714,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0253151655197144,
      "theoretical_loss": 4.966981646847723,
      "tokens_seen": 75366400
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004935205616850551,
      "loss": 3.7441,
      "theoretical_loss": 4.966981646847723,
      "tokens_seen": 75366400
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004935105315947844,
      "loss": 4.0297,
      "theoretical_loss": 4.966359155218047,
      "tokens_seen": 75431936
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004935005015045135,
      "loss": 3.8285,
      "theoretical_loss": 4.96573735546202,
      "tokens_seen": 75497472
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004934904714142428,
      "loss": 3.7561,
      "theoretical_loss": 4.965116246211068,
      "tokens_seen": 75563008
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004934804413239719,
      "loss": 3.9857,
      "theoretical_loss": 4.964495826100508,
      "tokens_seen": 75628544
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004934704112337011,
      "loss": 3.8798,
      "theoretical_loss": 4.963876093769536,
      "tokens_seen": 75694080
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004934603811434303,
      "loss": 3.7239,
      "theoretical_loss": 4.963257047861209,
      "tokens_seen": 75759616
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004934503510531595,
      "loss": 3.9851,
      "theoretical_loss": 4.962638687022432,
      "tokens_seen": 75825152
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004934403209628887,
      "loss": 3.6969,
      "theoretical_loss": 4.962021009903943,
      "tokens_seen": 75890688
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004934302908726179,
      "loss": 3.5063,
      "theoretical_loss": 4.9614040151603,
      "tokens_seen": 75956224
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.000493420260782347,
      "loss": 3.9226,
      "theoretical_loss": 4.960787701449868,
      "tokens_seen": 76021760
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004934102306920762,
      "loss": 4.2231,
      "theoretical_loss": 4.960172067434801,
      "tokens_seen": 76087296
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004934002006018054,
      "loss": 4.0389,
      "theoretical_loss": 4.959557111781034,
      "tokens_seen": 76152832
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004933901705115346,
      "loss": 4.0787,
      "theoretical_loss": 4.958942833158262,
      "tokens_seen": 76218368
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004933801404212638,
      "loss": 4.0188,
      "theoretical_loss": 4.958329230239932,
      "tokens_seen": 76283904
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.000493370110330993,
      "loss": 3.7401,
      "theoretical_loss": 4.9577163017032255,
      "tokens_seen": 76349440
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004933600802407221,
      "loss": 3.9433,
      "theoretical_loss": 4.957104046229048,
      "tokens_seen": 76414976
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004933500501504513,
      "loss": 3.5619,
      "theoretical_loss": 4.9564924625020135,
      "tokens_seen": 76480512
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004933400200601805,
      "loss": 4.1905,
      "theoretical_loss": 4.955881549210428,
      "tokens_seen": 76546048
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004933299899699098,
      "loss": 3.5923,
      "theoretical_loss": 4.955271305046283,
      "tokens_seen": 76611584
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004933199598796389,
      "loss": 3.8552,
      "theoretical_loss": 4.954661728705236,
      "tokens_seen": 76677120
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004933099297893682,
      "loss": 3.835,
      "theoretical_loss": 4.9540528188866,
      "tokens_seen": 76742656
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004932998996990972,
      "loss": 3.9461,
      "theoretical_loss": 4.953444574293328,
      "tokens_seen": 76808192
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004932898696088265,
      "loss": 3.9267,
      "theoretical_loss": 4.952836993632001,
      "tokens_seen": 76873728
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004932798395185557,
      "loss": 3.675,
      "theoretical_loss": 4.952230075612816,
      "tokens_seen": 76939264
    },
    {
      "epoch": 0.02,
      "objective/train/advantage_avg": 0.49505066871643066,
      "objective/train/docs_used": 55424,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.8856399059295654,
      "objective/train/original_loss": 3.885639190673828,
      "objective/train/theoretical_loss": 4.951623818949572,
      "objective/train/tokens_used": 97464800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24512356519699097,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507512092590332,
      "objective/train/weighted_lm_loss": 4.082724094390869,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0425469875335693,
      "theoretical_loss": 4.951623818949572,
      "tokens_seen": 77004800
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004932698094282849,
      "loss": 3.9717,
      "theoretical_loss": 4.951623818949572,
      "tokens_seen": 77004800
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004932597793380141,
      "loss": 3.9184,
      "theoretical_loss": 4.951018222359654,
      "tokens_seen": 77070336
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004932497492477432,
      "loss": 3.9951,
      "theoretical_loss": 4.950413284564025,
      "tokens_seen": 77135872
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004932397191574724,
      "loss": 4.0531,
      "theoretical_loss": 4.949809004287209,
      "tokens_seen": 77201408
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004932296890672016,
      "loss": 3.6452,
      "theoretical_loss": 4.949205380257279,
      "tokens_seen": 77266944
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004932196589769308,
      "loss": 3.7128,
      "theoretical_loss": 4.948602411205847,
      "tokens_seen": 77332480
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.00049320962888666,
      "loss": 4.0703,
      "theoretical_loss": 4.948000095868046,
      "tokens_seen": 77398016
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004931995987963893,
      "loss": 4.0413,
      "theoretical_loss": 4.9473984329825225,
      "tokens_seen": 77463552
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004931895687061183,
      "loss": 3.7341,
      "theoretical_loss": 4.946797421291418,
      "tokens_seen": 77529088
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004931795386158476,
      "loss": 4.035,
      "theoretical_loss": 4.946197059540362,
      "tokens_seen": 77594624
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004931695085255767,
      "loss": 3.8448,
      "theoretical_loss": 4.945597346478458,
      "tokens_seen": 77660160
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.000493159478435306,
      "loss": 3.7663,
      "theoretical_loss": 4.944998280858268,
      "tokens_seen": 77725696
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004931494483450352,
      "loss": 3.9068,
      "theoretical_loss": 4.944399861435803,
      "tokens_seen": 77791232
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004931394182547643,
      "loss": 3.8891,
      "theoretical_loss": 4.943802086970509,
      "tokens_seen": 77856768
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004931293881644935,
      "loss": 4.1114,
      "theoretical_loss": 4.943204956225257,
      "tokens_seen": 77922304
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004931193580742227,
      "loss": 3.6977,
      "theoretical_loss": 4.942608467966326,
      "tokens_seen": 77987840
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004931093279839519,
      "loss": 4.0549,
      "theoretical_loss": 4.942012620963396,
      "tokens_seen": 78053376
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004930992978936811,
      "loss": 3.8734,
      "theoretical_loss": 4.941417413989534,
      "tokens_seen": 78118912
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004930892678034102,
      "loss": 3.6231,
      "theoretical_loss": 4.940822845821179,
      "tokens_seen": 78184448
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004930792377131394,
      "loss": 3.9297,
      "theoretical_loss": 4.940228915238132,
      "tokens_seen": 78249984
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004930692076228686,
      "loss": 3.9805,
      "theoretical_loss": 4.939635621023548,
      "tokens_seen": 78315520
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004930591775325978,
      "loss": 3.7019,
      "theoretical_loss": 4.939042961963916,
      "tokens_seen": 78381056
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.000493049147442327,
      "loss": 3.642,
      "theoretical_loss": 4.938450936849053,
      "tokens_seen": 78446592
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004930391173520562,
      "loss": 4.0539,
      "theoretical_loss": 4.937859544472092,
      "tokens_seen": 78512128
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004930290872617853,
      "loss": 3.6814,
      "theoretical_loss": 4.937268783629463,
      "tokens_seen": 78577664
    },
    {
      "epoch": 0.02,
      "objective/train/advantage_avg": 0.4944605827331543,
      "objective/train/docs_used": 56825,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.531144142150879,
      "objective/train/original_loss": 3.5311439037323,
      "objective/train/theoretical_loss": 4.936678653120895,
      "objective/train/tokens_used": 99103200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24460074305534363,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506893396377563,
      "objective/train/weighted_lm_loss": 3.709916114807129,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0408108234405518,
      "theoretical_loss": 4.936678653120895,
      "tokens_seen": 78643200
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004930190571715146,
      "loss": 3.9277,
      "theoretical_loss": 4.936678653120895,
      "tokens_seen": 78643200
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004930090270812437,
      "loss": 4.1213,
      "theoretical_loss": 4.936089151749389,
      "tokens_seen": 78708736
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.000492998996990973,
      "loss": 4.0393,
      "theoretical_loss": 4.935500278321216,
      "tokens_seen": 78774272
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004929889669007021,
      "loss": 3.8018,
      "theoretical_loss": 4.934912031645903,
      "tokens_seen": 78839808
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004929789368104313,
      "loss": 3.6553,
      "theoretical_loss": 4.934324410536221,
      "tokens_seen": 78905344
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004929689067201605,
      "loss": 3.7482,
      "theoretical_loss": 4.933737413808172,
      "tokens_seen": 78970880
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004929588766298897,
      "loss": 3.938,
      "theoretical_loss": 4.933151040280983,
      "tokens_seen": 79036416
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004929488465396189,
      "loss": 3.9175,
      "theoretical_loss": 4.932565288777084,
      "tokens_seen": 79101952
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004929388164493481,
      "loss": 3.9345,
      "theoretical_loss": 4.93198015812211,
      "tokens_seen": 79167488
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004929287863590773,
      "loss": 3.9198,
      "theoretical_loss": 4.931395647144881,
      "tokens_seen": 79233024
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004929187562688064,
      "loss": 3.6705,
      "theoretical_loss": 4.930811754677391,
      "tokens_seen": 79298560
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004929087261785356,
      "loss": 4.3002,
      "theoretical_loss": 4.930228479554799,
      "tokens_seen": 79364096
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004928986960882648,
      "loss": 3.7063,
      "theoretical_loss": 4.929645820615419,
      "tokens_seen": 79429632
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.000492888665997994,
      "loss": 3.7041,
      "theoretical_loss": 4.929063776700706,
      "tokens_seen": 79495168
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004928786359077232,
      "loss": 3.7324,
      "theoretical_loss": 4.928482346655246,
      "tokens_seen": 79560704
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004928686058174523,
      "loss": 3.6676,
      "theoretical_loss": 4.927901529326744,
      "tokens_seen": 79626240
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004928585757271815,
      "loss": 3.9444,
      "theoretical_loss": 4.927321323566017,
      "tokens_seen": 79691776
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004928485456369107,
      "loss": 3.9064,
      "theoretical_loss": 4.926741728226978,
      "tokens_seen": 79757312
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.00049283851554664,
      "loss": 4.0572,
      "theoretical_loss": 4.926162742166628,
      "tokens_seen": 79822848
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004928284854563691,
      "loss": 3.7362,
      "theoretical_loss": 4.925584364245044,
      "tokens_seen": 79888384
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004928184553660984,
      "loss": 3.8292,
      "theoretical_loss": 4.925006593325367,
      "tokens_seen": 79953920
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004928084252758274,
      "loss": 3.9818,
      "theoretical_loss": 4.924429428273798,
      "tokens_seen": 80019456
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004927983951855567,
      "loss": 3.8849,
      "theoretical_loss": 4.923852867959576,
      "tokens_seen": 80084992
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004927883650952859,
      "loss": 4.0675,
      "theoretical_loss": 4.923276911254978,
      "tokens_seen": 80150528
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004927783350050151,
      "loss": 3.9406,
      "theoretical_loss": 4.9227015570353005,
      "tokens_seen": 80216064
    },
    {
      "epoch": 0.02,
      "objective/train/advantage_avg": 0.49447476863861084,
      "objective/train/docs_used": 57296,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 4.007246971130371,
      "objective/train/original_loss": 4.007246494293213,
      "objective/train/theoretical_loss": 4.9221268041788555,
      "objective/train/tokens_used": 100741600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24457311630249023,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050690770149231,
      "objective/train/weighted_lm_loss": 4.210178375244141,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0363596677780151,
      "theoretical_loss": 4.9221268041788555,
      "tokens_seen": 80281600
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004927683049147443,
      "loss": 4.1506,
      "theoretical_loss": 4.9221268041788555,
      "tokens_seen": 80281600
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004927582748244734,
      "loss": 3.5674,
      "theoretical_loss": 4.921552651566954,
      "tokens_seen": 80347136
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004927482447342026,
      "loss": 3.958,
      "theoretical_loss": 4.920979098083901,
      "tokens_seen": 80412672
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004927382146439318,
      "loss": 3.9651,
      "theoretical_loss": 4.920406142616981,
      "tokens_seen": 80478208
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.000492728184553661,
      "loss": 3.6684,
      "theoretical_loss": 4.919833784056449,
      "tokens_seen": 80543744
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004927181544633902,
      "loss": 3.6663,
      "theoretical_loss": 4.919262021295519,
      "tokens_seen": 80609280
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004927081243731193,
      "loss": 3.7485,
      "theoretical_loss": 4.918690853230357,
      "tokens_seen": 80674816
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004926980942828485,
      "loss": 3.743,
      "theoretical_loss": 4.918120278760069,
      "tokens_seen": 80740352
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004926880641925777,
      "loss": 3.7716,
      "theoretical_loss": 4.917550296786688,
      "tokens_seen": 80805888
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004926780341023069,
      "loss": 3.9656,
      "theoretical_loss": 4.9169809062151675,
      "tokens_seen": 80871424
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004926680040120361,
      "loss": 3.7532,
      "theoretical_loss": 4.916412105953371,
      "tokens_seen": 80936960
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004926579739217654,
      "loss": 4.0714,
      "theoretical_loss": 4.915843894912058,
      "tokens_seen": 81002496
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004926479438314944,
      "loss": 3.7998,
      "theoretical_loss": 4.915276272004883,
      "tokens_seen": 81068032
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004926379137412237,
      "loss": 3.7885,
      "theoretical_loss": 4.914709236148374,
      "tokens_seen": 81133568
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004926278836509528,
      "loss": 3.5449,
      "theoretical_loss": 4.9141427862619285,
      "tokens_seen": 81199104
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004926178535606821,
      "loss": 3.7449,
      "theoretical_loss": 4.913576921267806,
      "tokens_seen": 81264640
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004926078234704113,
      "loss": 3.7853,
      "theoretical_loss": 4.913011640091115,
      "tokens_seen": 81330176
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004925977933801404,
      "loss": 3.7895,
      "theoretical_loss": 4.9124469416598,
      "tokens_seen": 81395712
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004925877632898696,
      "loss": 3.822,
      "theoretical_loss": 4.91188282490464,
      "tokens_seen": 81461248
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004925777331995988,
      "loss": 3.5687,
      "theoretical_loss": 4.9113192887592305,
      "tokens_seen": 81526784
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.000492567703109328,
      "loss": 3.6957,
      "theoretical_loss": 4.91075633215998,
      "tokens_seen": 81592320
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004925576730190572,
      "loss": 3.7991,
      "theoretical_loss": 4.910193954046093,
      "tokens_seen": 81657856
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004925476429287864,
      "loss": 3.7094,
      "theoretical_loss": 4.9096321533595715,
      "tokens_seen": 81723392
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004925376128385155,
      "loss": 3.8998,
      "theoretical_loss": 4.909070929045194,
      "tokens_seen": 81788928
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004925275827482447,
      "loss": 3.7221,
      "theoretical_loss": 4.908510280050512,
      "tokens_seen": 81854464
    },
    {
      "epoch": 0.02,
      "objective/train/advantage_avg": 0.4938044846057892,
      "objective/train/docs_used": 58605,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 4.067493438720703,
      "objective/train/original_loss": 4.067493915557861,
      "objective/train/theoretical_loss": 4.907950205325841,
      "objective/train/tokens_used": 102380000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24391718208789825,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506203174591064,
      "objective/train/weighted_lm_loss": 4.273360729217529,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 4.907950205325841,
      "tokens_seen": 81920000
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004925175526579739,
      "loss": 4.1255,
      "theoretical_loss": 4.907950205325841,
      "tokens_seen": 81920000
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004925075225677031,
      "loss": 3.6976,
      "theoretical_loss": 4.907390703824248,
      "tokens_seen": 81985536
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004924974924774323,
      "loss": 4.1035,
      "theoretical_loss": 4.906831774501544,
      "tokens_seen": 82051072
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004924874623871615,
      "loss": 3.9791,
      "theoretical_loss": 4.9062734163162744,
      "tokens_seen": 82116608
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004924774322968907,
      "loss": 3.7101,
      "theoretical_loss": 4.905715628229709,
      "tokens_seen": 82182144
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004924674022066199,
      "loss": 3.7834,
      "theoretical_loss": 4.905158409205833,
      "tokens_seen": 82247680
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004924573721163491,
      "loss": 4.0789,
      "theoretical_loss": 4.904601758211339,
      "tokens_seen": 82313216
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004924473420260783,
      "loss": 3.8384,
      "theoretical_loss": 4.904045674215614,
      "tokens_seen": 82378752
    },
    {
      "epoch": 0.02,
      "learning_rate": 0.0004924373119358075,
      "loss": 3.8386,
      "theoretical_loss": 4.9034901561907365,
      "tokens_seen": 82444288
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004924272818455366,
      "loss": 3.6868,
      "theoretical_loss": 4.902935203111462,
      "tokens_seen": 82509824
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004924172517552658,
      "loss": 3.4293,
      "theoretical_loss": 4.902380813955215,
      "tokens_seen": 82575360
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.000492407221664995,
      "loss": 3.7685,
      "theoretical_loss": 4.901826987702082,
      "tokens_seen": 82640896
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004923971915747242,
      "loss": 4.0858,
      "theoretical_loss": 4.9012737233348,
      "tokens_seen": 82706432
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004923871614844534,
      "loss": 3.9125,
      "theoretical_loss": 4.900721019838748,
      "tokens_seen": 82771968
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004923771313941825,
      "loss": 3.5377,
      "theoretical_loss": 4.90016887620194,
      "tokens_seen": 82837504
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004923671013039117,
      "loss": 3.8075,
      "theoretical_loss": 4.899617291415016,
      "tokens_seen": 82903040
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004923570712136409,
      "loss": 3.5951,
      "theoretical_loss": 4.899066264471228,
      "tokens_seen": 82968576
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004923470411233702,
      "loss": 4.0688,
      "theoretical_loss": 4.898515794366439,
      "tokens_seen": 83034112
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004923370110330993,
      "loss": 3.7673,
      "theoretical_loss": 4.897965880099106,
      "tokens_seen": 83099648
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004923269809428286,
      "loss": 4.1155,
      "theoretical_loss": 4.897416520670279,
      "tokens_seen": 83165184
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004923169508525576,
      "loss": 3.7113,
      "theoretical_loss": 4.896867715083587,
      "tokens_seen": 83230720
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004923069207622869,
      "loss": 3.3438,
      "theoretical_loss": 4.896319462345232,
      "tokens_seen": 83296256
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004922968906720161,
      "loss": 3.6881,
      "theoretical_loss": 4.895771761463979,
      "tokens_seen": 83361792
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004922868605817453,
      "loss": 3.6947,
      "theoretical_loss": 4.895224611451146,
      "tokens_seen": 83427328
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004922768304914745,
      "loss": 3.7941,
      "theoretical_loss": 4.8946780113206,
      "tokens_seen": 83492864
    },
    {
      "epoch": 0.03,
      "objective/train/advantage_avg": 0.4932117462158203,
      "objective/train/docs_used": 59310,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.804917573928833,
      "objective/train/original_loss": 3.804917335510254,
      "objective/train/theoretical_loss": 4.894131960088744,
      "objective/train/tokens_used": 104018400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2434072494506836,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505585670471191,
      "objective/train/weighted_lm_loss": 3.9972569942474365,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 4.894131960088744,
      "tokens_seen": 83558400
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004922668004012036,
      "loss": 3.9463,
      "theoretical_loss": 4.894131960088744,
      "tokens_seen": 83558400
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004922567703109328,
      "loss": 3.8654,
      "theoretical_loss": 4.89358645677451,
      "tokens_seen": 83623936
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.000492246740220662,
      "loss": 3.84,
      "theoretical_loss": 4.8930415003993515,
      "tokens_seen": 83689472
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004922367101303912,
      "loss": 3.8044,
      "theoretical_loss": 4.892497089987232,
      "tokens_seen": 83755008
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004922266800401204,
      "loss": 3.8692,
      "theoretical_loss": 4.891953224564623,
      "tokens_seen": 83820544
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004922166499498495,
      "loss": 4.1269,
      "theoretical_loss": 4.891409903160486,
      "tokens_seen": 83886080
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004922066198595787,
      "loss": 3.3756,
      "theoretical_loss": 4.890867124806276,
      "tokens_seen": 83951616
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004921965897693079,
      "loss": 3.9552,
      "theoretical_loss": 4.890324888535918,
      "tokens_seen": 84017152
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004921865596790371,
      "loss": 3.7822,
      "theoretical_loss": 4.889783193385815,
      "tokens_seen": 84082688
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004921765295887663,
      "loss": 3.8355,
      "theoretical_loss": 4.88924203839483,
      "tokens_seen": 84148224
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004921664994984956,
      "loss": 3.9897,
      "theoretical_loss": 4.88870142260428,
      "tokens_seen": 84213760
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004921564694082246,
      "loss": 3.8534,
      "theoretical_loss": 4.888161345057926,
      "tokens_seen": 84279296
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004921464393179539,
      "loss": 3.679,
      "theoretical_loss": 4.88762180480197,
      "tokens_seen": 84344832
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.000492136409227683,
      "loss": 3.806,
      "theoretical_loss": 4.8870828008850395,
      "tokens_seen": 84410368
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004921263791374123,
      "loss": 3.6148,
      "theoretical_loss": 4.886544332358189,
      "tokens_seen": 84475904
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004921163490471415,
      "loss": 3.5951,
      "theoretical_loss": 4.886006398274881,
      "tokens_seen": 84541440
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004921063189568706,
      "loss": 3.6971,
      "theoretical_loss": 4.885468997690987,
      "tokens_seen": 84606976
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004920962888665998,
      "loss": 3.7775,
      "theoretical_loss": 4.884932129664774,
      "tokens_seen": 84672512
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.000492086258776329,
      "loss": 3.6966,
      "theoretical_loss": 4.8843957932569,
      "tokens_seen": 84738048
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004920762286860582,
      "loss": 3.5616,
      "theoretical_loss": 4.883859987530404,
      "tokens_seen": 84803584
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004920661985957874,
      "loss": 3.5689,
      "theoretical_loss": 4.8833247115507,
      "tokens_seen": 84869120
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004920561685055166,
      "loss": 3.9699,
      "theoretical_loss": 4.882789964385566,
      "tokens_seen": 84934656
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004920461384152457,
      "loss": 3.8289,
      "theoretical_loss": 4.88225574510514,
      "tokens_seen": 85000192
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004920361083249749,
      "loss": 3.6677,
      "theoretical_loss": 4.8817220527819085,
      "tokens_seen": 85065728
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004920260782347041,
      "loss": 3.89,
      "theoretical_loss": 4.881188886490703,
      "tokens_seen": 85131264
    },
    {
      "epoch": 0.03,
      "objective/train/advantage_avg": 0.49440962076187134,
      "objective/train/docs_used": 60492,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.410545825958252,
      "objective/train/original_loss": 3.410545825958252,
      "objective/train/theoretical_loss": 4.880656245308686,
      "objective/train/tokens_used": 105656800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24451497197151184,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050683856010437,
      "objective/train/weighted_lm_loss": 3.583144426345825,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0363596677780151,
      "theoretical_loss": 4.880656245308686,
      "tokens_seen": 85196800
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004920160481444333,
      "loss": 3.8163,
      "theoretical_loss": 4.880656245308686,
      "tokens_seen": 85196800
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004920060180541625,
      "loss": 3.8017,
      "theoretical_loss": 4.880124128315353,
      "tokens_seen": 85262336
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004919959879638916,
      "loss": 3.8179,
      "theoretical_loss": 4.879592534592515,
      "tokens_seen": 85327872
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004919859578736209,
      "loss": 3.7257,
      "theoretical_loss": 4.879061463224295,
      "tokens_seen": 85393408
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.00049197592778335,
      "loss": 3.6286,
      "theoretical_loss": 4.87853091329712,
      "tokens_seen": 85458944
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004919658976930793,
      "loss": 3.9954,
      "theoretical_loss": 4.878000883899718,
      "tokens_seen": 85524480
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004919558676028084,
      "loss": 3.9364,
      "theoretical_loss": 4.877471374123102,
      "tokens_seen": 85590016
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004919458375125377,
      "loss": 3.7739,
      "theoretical_loss": 4.876942383060569,
      "tokens_seen": 85655552
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004919358074222668,
      "loss": 3.887,
      "theoretical_loss": 4.876413909807688,
      "tokens_seen": 85721088
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.000491925777331996,
      "loss": 3.6055,
      "theoretical_loss": 4.875885953462296,
      "tokens_seen": 85786624
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004919157472417252,
      "loss": 3.6926,
      "theoretical_loss": 4.8753585131244925,
      "tokens_seen": 85852160
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004919057171514544,
      "loss": 3.648,
      "theoretical_loss": 4.874831587896624,
      "tokens_seen": 85917696
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004918956870611836,
      "loss": 3.7943,
      "theoretical_loss": 4.874305176883285,
      "tokens_seen": 85983232
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004918856569709127,
      "loss": 3.4472,
      "theoretical_loss": 4.873779279191307,
      "tokens_seen": 86048768
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004918756268806419,
      "loss": 4.0346,
      "theoretical_loss": 4.87325389392975,
      "tokens_seen": 86114304
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004918655967903711,
      "loss": 3.7369,
      "theoretical_loss": 4.872729020209899,
      "tokens_seen": 86179840
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004918555667001003,
      "loss": 3.8494,
      "theoretical_loss": 4.872204657145253,
      "tokens_seen": 86245376
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004918455366098295,
      "loss": 3.9131,
      "theoretical_loss": 4.871680803851521,
      "tokens_seen": 86310912
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004918355065195586,
      "loss": 3.7522,
      "theoretical_loss": 4.8711574594466125,
      "tokens_seen": 86376448
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004918254764292878,
      "loss": 3.9465,
      "theoretical_loss": 4.87063462305063,
      "tokens_seen": 86441984
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.000491815446339017,
      "loss": 3.9634,
      "theoretical_loss": 4.870112293785865,
      "tokens_seen": 86507520
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004918054162487463,
      "loss": 3.7144,
      "theoretical_loss": 4.869590470776787,
      "tokens_seen": 86573056
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004917953861584754,
      "loss": 3.6877,
      "theoretical_loss": 4.869069153150041,
      "tokens_seen": 86638592
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004917853560682047,
      "loss": 3.585,
      "theoretical_loss": 4.868548340034435,
      "tokens_seen": 86704128
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004917753259779337,
      "loss": 3.4763,
      "theoretical_loss": 4.868028030560937,
      "tokens_seen": 86769664
    },
    {
      "epoch": 0.03,
      "objective/train/advantage_avg": 0.4928877055644989,
      "objective/train/docs_used": 61005,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.3430697917938232,
      "objective/train/original_loss": 3.343069553375244,
      "objective/train/theoretical_loss": 4.867508223862666,
      "objective/train/tokens_used": 107295200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24301739037036896,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505239963531494,
      "objective/train/weighted_lm_loss": 3.5122222900390625,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0408108234405518,
      "theoretical_loss": 4.867508223862666,
      "tokens_seen": 86835200
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.000491765295887663,
      "loss": 3.7507,
      "theoretical_loss": 4.867508223862666,
      "tokens_seen": 86835200
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004917552657973922,
      "loss": 3.8105,
      "theoretical_loss": 4.866988919074889,
      "tokens_seen": 86900736
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004917452357071214,
      "loss": 3.88,
      "theoretical_loss": 4.866470115335007,
      "tokens_seen": 86966272
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004917352056168506,
      "loss": 3.4849,
      "theoretical_loss": 4.865951811782555,
      "tokens_seen": 87031808
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004917251755265797,
      "loss": 3.6316,
      "theoretical_loss": 4.865434007559189,
      "tokens_seen": 87097344
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004917151454363089,
      "loss": 3.5496,
      "theoretical_loss": 4.864916701808687,
      "tokens_seen": 87162880
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004917051153460381,
      "loss": 3.7789,
      "theoretical_loss": 4.864399893676933,
      "tokens_seen": 87228416
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004916950852557673,
      "loss": 3.7991,
      "theoretical_loss": 4.863883582311917,
      "tokens_seen": 87293952
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004916850551654965,
      "loss": 3.3787,
      "theoretical_loss": 4.863367766863728,
      "tokens_seen": 87359488
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004916750250752258,
      "loss": 3.6662,
      "theoretical_loss": 4.86285244648454,
      "tokens_seen": 87425024
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004916649949849548,
      "loss": 3.9157,
      "theoretical_loss": 4.862337620328617,
      "tokens_seen": 87490560
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004916549648946841,
      "loss": 3.4893,
      "theoretical_loss": 4.861823287552293,
      "tokens_seen": 87556096
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004916449348044132,
      "loss": 4.0004,
      "theoretical_loss": 4.861309447313978,
      "tokens_seen": 87621632
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004916349047141425,
      "loss": 3.7753,
      "theoretical_loss": 4.860796098774145,
      "tokens_seen": 87687168
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004916248746238717,
      "loss": 3.6325,
      "theoretical_loss": 4.8602832410953205,
      "tokens_seen": 87752704
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004916148445336008,
      "loss": 3.7378,
      "theoretical_loss": 4.859770873442082,
      "tokens_seen": 87818240
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.00049160481444333,
      "loss": 4.0049,
      "theoretical_loss": 4.859258994981058,
      "tokens_seen": 87883776
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004915947843530592,
      "loss": 3.7842,
      "theoretical_loss": 4.858747604880904,
      "tokens_seen": 87949312
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004915847542627884,
      "loss": 3.8695,
      "theoretical_loss": 4.858236702312313,
      "tokens_seen": 88014848
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004915747241725176,
      "loss": 3.7079,
      "theoretical_loss": 4.857726286448001,
      "tokens_seen": 88080384
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004915646940822468,
      "loss": 3.7249,
      "theoretical_loss": 4.857216356462702,
      "tokens_seen": 88145920
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004915546639919759,
      "loss": 3.9564,
      "theoretical_loss": 4.856706911533163,
      "tokens_seen": 88211456
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004915446339017051,
      "loss": 3.6887,
      "theoretical_loss": 4.85619795083813,
      "tokens_seen": 88276992
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004915346038114343,
      "loss": 4.0395,
      "theoretical_loss": 4.855689473558357,
      "tokens_seen": 88342528
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004915245737211635,
      "loss": 4.0529,
      "theoretical_loss": 4.855181478876584,
      "tokens_seen": 88408064
    },
    {
      "epoch": 0.03,
      "objective/train/advantage_avg": 0.4952790141105652,
      "objective/train/docs_used": 61597,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.75881290435791,
      "objective/train/original_loss": 3.758812427520752,
      "objective/train/theoretical_loss": 4.854673965977539,
      "objective/train/tokens_used": 108933600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2453484982252121,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050775170326233,
      "objective/train/weighted_lm_loss": 3.9494822025299072,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0451053380966187,
      "theoretical_loss": 4.854673965977539,
      "tokens_seen": 88473600
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004915145436308927,
      "loss": 3.8843,
      "theoretical_loss": 4.854673965977539,
      "tokens_seen": 88473600
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004915045135406218,
      "loss": 3.6474,
      "theoretical_loss": 4.85416693404793,
      "tokens_seen": 88539136
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004914944834503511,
      "loss": 3.7979,
      "theoretical_loss": 4.8536603822764395,
      "tokens_seen": 88604672
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004914844533600802,
      "loss": 3.578,
      "theoretical_loss": 4.853154309853716,
      "tokens_seen": 88670208
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004914744232698095,
      "loss": 3.6503,
      "theoretical_loss": 4.85264871597237,
      "tokens_seen": 88735744
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004914643931795386,
      "loss": 3.61,
      "theoretical_loss": 4.852143599826965,
      "tokens_seen": 88801280
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004914543630892679,
      "loss": 3.743,
      "theoretical_loss": 4.851638960614015,
      "tokens_seen": 88866816
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.000491444332998997,
      "loss": 3.872,
      "theoretical_loss": 4.851134797531978,
      "tokens_seen": 88932352
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004914343029087262,
      "loss": 3.8565,
      "theoretical_loss": 4.850631109781246,
      "tokens_seen": 88997888
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004914242728184554,
      "loss": 3.916,
      "theoretical_loss": 4.850127896564143,
      "tokens_seen": 89063424
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004914142427281846,
      "loss": 3.9232,
      "theoretical_loss": 4.849625157084915,
      "tokens_seen": 89128960
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004914042126379138,
      "loss": 3.7873,
      "theoretical_loss": 4.8491228905497294,
      "tokens_seen": 89194496
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004913941825476429,
      "loss": 3.9636,
      "theoretical_loss": 4.848621096166663,
      "tokens_seen": 89260032
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004913841524573721,
      "loss": 3.6194,
      "theoretical_loss": 4.848119773145701,
      "tokens_seen": 89325568
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004913741223671013,
      "loss": 3.719,
      "theoretical_loss": 4.847618920698728,
      "tokens_seen": 89391104
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004913640922768305,
      "loss": 3.8057,
      "theoretical_loss": 4.847118538039522,
      "tokens_seen": 89456640
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004913540621865597,
      "loss": 3.8469,
      "theoretical_loss": 4.846618624383751,
      "tokens_seen": 89522176
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004913440320962888,
      "loss": 3.7962,
      "theoretical_loss": 4.846119178948966,
      "tokens_seen": 89587712
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.000491334002006018,
      "loss": 3.8938,
      "theoretical_loss": 4.84562020095459,
      "tokens_seen": 89653248
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004913239719157472,
      "loss": 4.0468,
      "theoretical_loss": 4.845121689621922,
      "tokens_seen": 89718784
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004913139418254765,
      "loss": 3.6846,
      "theoretical_loss": 4.844623644174124,
      "tokens_seen": 89784320
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004913039117352056,
      "loss": 3.6867,
      "theoretical_loss": 4.844126063836216,
      "tokens_seen": 89849856
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004912938816449349,
      "loss": 3.9784,
      "theoretical_loss": 4.84362894783507,
      "tokens_seen": 89915392
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004912838515546639,
      "loss": 3.5115,
      "theoretical_loss": 4.8431322953994105,
      "tokens_seen": 89980928
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004912738214643932,
      "loss": 3.6379,
      "theoretical_loss": 4.8426361057598,
      "tokens_seen": 90046464
    },
    {
      "epoch": 0.03,
      "objective/train/advantage_avg": 0.4942990839481354,
      "objective/train/docs_used": 63098,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.5812742710113525,
      "objective/train/original_loss": 3.5812737941741943,
      "objective/train/theoretical_loss": 4.8421403781486365,
      "objective/train/tokens_used": 110572000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24439339339733124,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050672173500061,
      "objective/train/weighted_lm_loss": 3.7626872062683105,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0425469875335693,
      "theoretical_loss": 4.8421403781486365,
      "tokens_seen": 90112000
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004912637913741224,
      "loss": 3.6925,
      "theoretical_loss": 4.8421403781486365,
      "tokens_seen": 90112000
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004912537612838516,
      "loss": 3.4008,
      "theoretical_loss": 4.8416451118001484,
      "tokens_seen": 90177536
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004912437311935808,
      "loss": 3.5174,
      "theoretical_loss": 4.841150305950391,
      "tokens_seen": 90243072
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.00049123370110331,
      "loss": 3.3859,
      "theoretical_loss": 4.840655959837236,
      "tokens_seen": 90308608
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004912236710130391,
      "loss": 3.5937,
      "theoretical_loss": 4.8401620727003705,
      "tokens_seen": 90374144
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004912136409227683,
      "loss": 3.6419,
      "theoretical_loss": 4.839668643781287,
      "tokens_seen": 90439680
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004912036108324975,
      "loss": 3.7207,
      "theoretical_loss": 4.839175672323279,
      "tokens_seen": 90505216
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004911935807422267,
      "loss": 3.706,
      "theoretical_loss": 4.838683157571442,
      "tokens_seen": 90570752
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004911835506519559,
      "loss": 3.6671,
      "theoretical_loss": 4.838191098772656,
      "tokens_seen": 90636288
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.000491173520561685,
      "loss": 3.6839,
      "theoretical_loss": 4.837699495175589,
      "tokens_seen": 90701824
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004911634904714142,
      "loss": 3.7562,
      "theoretical_loss": 4.837208346030691,
      "tokens_seen": 90767360
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004911534603811434,
      "loss": 3.5432,
      "theoretical_loss": 4.836717650590181,
      "tokens_seen": 90832896
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004911434302908726,
      "loss": 4.1462,
      "theoretical_loss": 4.836227408108052,
      "tokens_seen": 90898432
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004911334002006019,
      "loss": 3.6991,
      "theoretical_loss": 4.835737617840058,
      "tokens_seen": 90963968
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004911233701103309,
      "loss": 3.6676,
      "theoretical_loss": 4.835248279043711,
      "tokens_seen": 91029504
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004911133400200602,
      "loss": 3.7131,
      "theoretical_loss": 4.834759390978274,
      "tokens_seen": 91095040
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004911033099297893,
      "loss": 3.8646,
      "theoretical_loss": 4.834270952904761,
      "tokens_seen": 91160576
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004910932798395186,
      "loss": 3.7868,
      "theoretical_loss": 4.833782964085925,
      "tokens_seen": 91226112
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004910832497492478,
      "loss": 3.8456,
      "theoretical_loss": 4.8332954237862555,
      "tokens_seen": 91291648
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.000491073219658977,
      "loss": 3.682,
      "theoretical_loss": 4.832808331271973,
      "tokens_seen": 91357184
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004910631895687061,
      "loss": 3.95,
      "theoretical_loss": 4.832321685811024,
      "tokens_seen": 91422720
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004910531594784353,
      "loss": 3.9957,
      "theoretical_loss": 4.831835486673079,
      "tokens_seen": 91488256
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004910431293881645,
      "loss": 3.5659,
      "theoretical_loss": 4.831349733129516,
      "tokens_seen": 91553792
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004910330992978937,
      "loss": 3.7207,
      "theoretical_loss": 4.83086442445343,
      "tokens_seen": 91619328
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004910230692076229,
      "loss": 3.9654,
      "theoretical_loss": 4.830379559919616,
      "tokens_seen": 91684864
    },
    {
      "epoch": 0.03,
      "objective/train/advantage_avg": 0.49309033155441284,
      "objective/train/docs_used": 63840,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.506807565689087,
      "objective/train/original_loss": 3.506807565689087,
      "objective/train/theoretical_loss": 4.829895138804573,
      "objective/train/tokens_used": 112210400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24322980642318726,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050545334815979,
      "objective/train/weighted_lm_loss": 3.6838057041168213,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 4.829895138804573,
      "tokens_seen": 91750400
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.000491013039117352,
      "loss": 3.6523,
      "theoretical_loss": 4.829895138804573,
      "tokens_seen": 91750400
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004910030090270812,
      "loss": 3.5502,
      "theoretical_loss": 4.829411160386489,
      "tokens_seen": 91815936
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004909929789368104,
      "loss": 3.6347,
      "theoretical_loss": 4.828927623945246,
      "tokens_seen": 91881472
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004909829488465397,
      "loss": 3.5057,
      "theoretical_loss": 4.828444528762407,
      "tokens_seen": 91947008
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004909729187562688,
      "loss": 3.9223,
      "theoretical_loss": 4.827961874121215,
      "tokens_seen": 92012544
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004909628886659981,
      "loss": 3.7191,
      "theoretical_loss": 4.827479659306585,
      "tokens_seen": 92078080
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004909528585757272,
      "loss": 3.8858,
      "theoretical_loss": 4.826997883605104,
      "tokens_seen": 92143616
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004909428284854564,
      "loss": 3.5103,
      "theoretical_loss": 4.826516546305019,
      "tokens_seen": 92209152
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004909327983951856,
      "loss": 3.7002,
      "theoretical_loss": 4.826035646696238,
      "tokens_seen": 92274688
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004909227683049148,
      "loss": 4.026,
      "theoretical_loss": 4.825555184070321,
      "tokens_seen": 92340224
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.000490912738214644,
      "loss": 3.9292,
      "theoretical_loss": 4.825075157720478,
      "tokens_seen": 92405760
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004909027081243731,
      "loss": 3.6154,
      "theoretical_loss": 4.824595566941559,
      "tokens_seen": 92471296
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004908926780341023,
      "loss": 3.639,
      "theoretical_loss": 4.824116411030056,
      "tokens_seen": 92536832
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004908826479438315,
      "loss": 3.7659,
      "theoretical_loss": 4.823637689284094,
      "tokens_seen": 92602368
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004908726178535607,
      "loss": 3.7322,
      "theoretical_loss": 4.823159401003425,
      "tokens_seen": 92667904
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004908625877632899,
      "loss": 3.7145,
      "theoretical_loss": 4.8226815454894245,
      "tokens_seen": 92733440
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.000490852557673019,
      "loss": 3.7054,
      "theoretical_loss": 4.822204122045089,
      "tokens_seen": 92798976
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004908425275827482,
      "loss": 3.7537,
      "theoretical_loss": 4.821727129975027,
      "tokens_seen": 92864512
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004908324974924774,
      "loss": 3.4858,
      "theoretical_loss": 4.821250568585455,
      "tokens_seen": 92930048
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004908224674022067,
      "loss": 3.687,
      "theoretical_loss": 4.8207744371841965,
      "tokens_seen": 92995584
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004908124373119358,
      "loss": 3.6625,
      "theoretical_loss": 4.820298735080673,
      "tokens_seen": 93061120
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004908024072216651,
      "loss": 3.6448,
      "theoretical_loss": 4.819823461585898,
      "tokens_seen": 93126656
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004907923771313941,
      "loss": 3.5431,
      "theoretical_loss": 4.819348616012476,
      "tokens_seen": 93192192
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004907823470411234,
      "loss": 4.044,
      "theoretical_loss": 4.818874197674601,
      "tokens_seen": 93257728
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004907723169508526,
      "loss": 3.5817,
      "theoretical_loss": 4.8184002058880395,
      "tokens_seen": 93323264
    },
    {
      "epoch": 0.03,
      "objective/train/advantage_avg": 0.4954739809036255,
      "objective/train/docs_used": 65224,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.492338180541992,
      "objective/train/original_loss": 3.4923384189605713,
      "objective/train/theoretical_loss": 4.817926639970137,
      "objective/train/tokens_used": 113848800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2455521821975708,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507956743240356,
      "objective/train/weighted_lm_loss": 3.669647216796875,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0363596677780151,
      "theoretical_loss": 4.817926639970137,
      "tokens_seen": 93388800
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004907622868605818,
      "loss": 3.7307,
      "theoretical_loss": 4.817926639970137,
      "tokens_seen": 93388800
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.000490752256770311,
      "loss": 3.6544,
      "theoretical_loss": 4.817453499239813,
      "tokens_seen": 93454336
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004907422266800401,
      "loss": 3.5088,
      "theoretical_loss": 4.816980783017545,
      "tokens_seen": 93519872
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004907321965897693,
      "loss": 3.7166,
      "theoretical_loss": 4.8165084906253774,
      "tokens_seen": 93585408
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004907221664994985,
      "loss": 3.6557,
      "theoretical_loss": 4.81603662138691,
      "tokens_seen": 93650944
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004907121364092277,
      "loss": 3.5854,
      "theoretical_loss": 4.815565174627296,
      "tokens_seen": 93716480
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004907021063189569,
      "loss": 3.8646,
      "theoretical_loss": 4.815094149673232,
      "tokens_seen": 93782016
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.000490692076228686,
      "loss": 3.8162,
      "theoretical_loss": 4.8146235458529585,
      "tokens_seen": 93847552
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004906820461384152,
      "loss": 3.6981,
      "theoretical_loss": 4.814153362496257,
      "tokens_seen": 93913088
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004906720160481444,
      "loss": 3.4162,
      "theoretical_loss": 4.813683598934439,
      "tokens_seen": 93978624
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004906619859578736,
      "loss": 3.6876,
      "theoretical_loss": 4.813214254500345,
      "tokens_seen": 94044160
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004906519558676028,
      "loss": 3.8465,
      "theoretical_loss": 4.812745328528342,
      "tokens_seen": 94109696
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004906419257773321,
      "loss": 3.6749,
      "theoretical_loss": 4.812276820354314,
      "tokens_seen": 94175232
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004906318956870611,
      "loss": 3.6306,
      "theoretical_loss": 4.811808729315663,
      "tokens_seen": 94240768
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004906218655967904,
      "loss": 3.2468,
      "theoretical_loss": 4.811341054751299,
      "tokens_seen": 94306304
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004906118355065195,
      "loss": 3.6852,
      "theoretical_loss": 4.810873796001641,
      "tokens_seen": 94371840
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004906018054162488,
      "loss": 4.0577,
      "theoretical_loss": 4.810406952408607,
      "tokens_seen": 94437376
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.000490591775325978,
      "loss": 3.623,
      "theoretical_loss": 4.809940523315612,
      "tokens_seen": 94502912
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004905817452357072,
      "loss": 3.7406,
      "theoretical_loss": 4.809474508067569,
      "tokens_seen": 94568448
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004905717151454363,
      "loss": 3.409,
      "theoretical_loss": 4.809008906010872,
      "tokens_seen": 94633984
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004905616850551655,
      "loss": 3.7801,
      "theoretical_loss": 4.808543716493404,
      "tokens_seen": 94699520
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004905516549648947,
      "loss": 4.0312,
      "theoretical_loss": 4.808078938864527,
      "tokens_seen": 94765056
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004905416248746239,
      "loss": 3.6556,
      "theoretical_loss": 4.8076145724750745,
      "tokens_seen": 94830592
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004905315947843531,
      "loss": 3.7045,
      "theoretical_loss": 4.807150616677357,
      "tokens_seen": 94896128
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004905215646940822,
      "loss": 3.6924,
      "theoretical_loss": 4.806687070825147,
      "tokens_seen": 94961664
    },
    {
      "epoch": 0.03,
      "objective/train/advantage_avg": 0.4944251775741577,
      "objective/train/docs_used": 65494,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.7475695610046387,
      "objective/train/original_loss": 3.7475695610046387,
      "objective/train/theoretical_loss": 4.80622393427368,
      "objective/train/tokens_used": 115487200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24452589452266693,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506855249404907,
      "objective/train/weighted_lm_loss": 3.937241315841675,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0253151655197144,
      "theoretical_loss": 4.80622393427368,
      "tokens_seen": 95027200
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004905115346038114,
      "loss": 3.636,
      "theoretical_loss": 4.80622393427368,
      "tokens_seen": 95027200
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004905015045135406,
      "loss": 3.7031,
      "theoretical_loss": 4.80576120637965,
      "tokens_seen": 95092736
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004904914744232698,
      "loss": 3.7707,
      "theoretical_loss": 4.8052988865012045,
      "tokens_seen": 95158272
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.000490481444332999,
      "loss": 3.5449,
      "theoretical_loss": 4.80483697399794,
      "tokens_seen": 95223808
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004904714142427281,
      "loss": 3.7316,
      "theoretical_loss": 4.804375468230898,
      "tokens_seen": 95289344
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004904613841524574,
      "loss": 3.5955,
      "theoretical_loss": 4.803914368562561,
      "tokens_seen": 95354880
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004904513540621865,
      "loss": 3.8927,
      "theoretical_loss": 4.803453674356847,
      "tokens_seen": 95420416
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004904413239719158,
      "loss": 3.765,
      "theoretical_loss": 4.802993384979108,
      "tokens_seen": 95485952
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004904312938816449,
      "loss": 3.8282,
      "theoretical_loss": 4.802533499796123,
      "tokens_seen": 95551488
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004904212637913742,
      "loss": 3.6726,
      "theoretical_loss": 4.8020740181760955,
      "tokens_seen": 95617024
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004904112337011033,
      "loss": 3.4998,
      "theoretical_loss": 4.8016149394886485,
      "tokens_seen": 95682560
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004904012036108325,
      "loss": 3.8195,
      "theoretical_loss": 4.80115626310482,
      "tokens_seen": 95748096
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004903911735205617,
      "loss": 3.7201,
      "theoretical_loss": 4.800697988397061,
      "tokens_seen": 95813632
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004903811434302909,
      "loss": 3.8363,
      "theoretical_loss": 4.800240114739227,
      "tokens_seen": 95879168
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004903711133400201,
      "loss": 3.7531,
      "theoretical_loss": 4.799782641506579,
      "tokens_seen": 95944704
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004903610832497492,
      "loss": 3.5596,
      "theoretical_loss": 4.799325568075776,
      "tokens_seen": 96010240
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004903510531594784,
      "loss": 3.7619,
      "theoretical_loss": 4.798868893824874,
      "tokens_seen": 96075776
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004903410230692076,
      "loss": 3.8497,
      "theoretical_loss": 4.798412618133318,
      "tokens_seen": 96141312
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004903309929789368,
      "loss": 3.9879,
      "theoretical_loss": 4.797956740381938,
      "tokens_seen": 96206848
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.000490320962888666,
      "loss": 3.6913,
      "theoretical_loss": 4.797501259952952,
      "tokens_seen": 96272384
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004903109327983952,
      "loss": 3.9713,
      "theoretical_loss": 4.797046176229951,
      "tokens_seen": 96337920
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004903009027081243,
      "loss": 3.6287,
      "theoretical_loss": 4.796591488597907,
      "tokens_seen": 96403456
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004902908726178535,
      "loss": 3.5612,
      "theoretical_loss": 4.796137196443157,
      "tokens_seen": 96468992
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004902808425275828,
      "loss": 3.937,
      "theoretical_loss": 4.795683299153408,
      "tokens_seen": 96534528
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004902708124373119,
      "loss": 3.8828,
      "theoretical_loss": 4.795229796117728,
      "tokens_seen": 96600064
    },
    {
      "epoch": 0.03,
      "objective/train/advantage_avg": 0.49317771196365356,
      "objective/train/docs_used": 66055,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.6881799697875977,
      "objective/train/original_loss": 3.688180446624756,
      "objective/train/theoretical_loss": 4.7947766867265464,
      "objective/train/tokens_used": 117125600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24330544471740723,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505545139312744,
      "objective/train/weighted_lm_loss": 3.8745532035827637,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0417574644088745,
      "theoretical_loss": 4.7947766867265464,
      "tokens_seen": 96665600
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004902607823470412,
      "loss": 3.8242,
      "theoretical_loss": 4.7947766867265464,
      "tokens_seen": 96665600
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004902507522567703,
      "loss": 3.8501,
      "theoretical_loss": 4.794323970371646,
      "tokens_seen": 96731136
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004902407221664995,
      "loss": 3.7862,
      "theoretical_loss": 4.793871646446159,
      "tokens_seen": 96796672
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004902306920762287,
      "loss": 3.5676,
      "theoretical_loss": 4.793419714344568,
      "tokens_seen": 96862208
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004902206619859579,
      "loss": 3.7169,
      "theoretical_loss": 4.792968173462697,
      "tokens_seen": 96927744
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004902106318956871,
      "loss": 3.8776,
      "theoretical_loss": 4.792517023197708,
      "tokens_seen": 96993280
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004902006018054163,
      "loss": 3.8573,
      "theoretical_loss": 4.792066262948102,
      "tokens_seen": 97058816
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004901905717151454,
      "loss": 3.8907,
      "theoretical_loss": 4.7916158921137075,
      "tokens_seen": 97124352
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004901805416248746,
      "loss": 3.6802,
      "theoretical_loss": 4.791165910095685,
      "tokens_seen": 97189888
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004901705115346038,
      "loss": 3.8489,
      "theoretical_loss": 4.790716316296514,
      "tokens_seen": 97255424
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.000490160481444333,
      "loss": 3.8958,
      "theoretical_loss": 4.790267110119999,
      "tokens_seen": 97320960
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004901504513540623,
      "loss": 3.516,
      "theoretical_loss": 4.789818290971258,
      "tokens_seen": 97386496
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004901404212637913,
      "loss": 3.6904,
      "theoretical_loss": 4.7893698582567215,
      "tokens_seen": 97452032
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004901303911735206,
      "loss": 3.9245,
      "theoretical_loss": 4.788921811384128,
      "tokens_seen": 97517568
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004901203610832497,
      "loss": 3.6737,
      "theoretical_loss": 4.788474149762525,
      "tokens_seen": 97583104
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.000490110330992979,
      "loss": 3.7189,
      "theoretical_loss": 4.7880268728022575,
      "tokens_seen": 97648640
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004901003009027082,
      "loss": 3.7271,
      "theoretical_loss": 4.787579979914966,
      "tokens_seen": 97714176
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004900902708124374,
      "loss": 3.7924,
      "theoretical_loss": 4.7871334705135915,
      "tokens_seen": 97779712
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004900802407221665,
      "loss": 3.8066,
      "theoretical_loss": 4.78668734401236,
      "tokens_seen": 97845248
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004900702106318957,
      "loss": 3.8914,
      "theoretical_loss": 4.786241599826784,
      "tokens_seen": 97910784
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004900601805416249,
      "loss": 3.445,
      "theoretical_loss": 4.78579623737366,
      "tokens_seen": 97976320
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004900501504513541,
      "loss": 3.8743,
      "theoretical_loss": 4.785351256071063,
      "tokens_seen": 98041856
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004900401203610833,
      "loss": 3.4863,
      "theoretical_loss": 4.784906655338345,
      "tokens_seen": 98107392
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004900300902708124,
      "loss": 3.7286,
      "theoretical_loss": 4.7844624345961275,
      "tokens_seen": 98172928
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004900200601805416,
      "loss": 3.8187,
      "theoretical_loss": 4.7840185932663,
      "tokens_seen": 98238464
    },
    {
      "epoch": 0.03,
      "objective/train/advantage_avg": 0.4944189488887787,
      "objective/train/docs_used": 66835,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.9753668308258057,
      "objective/train/original_loss": 3.9753668308258057,
      "objective/train/theoretical_loss": 4.783575130772016,
      "objective/train/tokens_used": 118764000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24451906979084015,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506848096847534,
      "objective/train/weighted_lm_loss": 4.176583290100098,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0451053380966187,
      "theoretical_loss": 4.783575130772016,
      "tokens_seen": 98304000
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004900100300902708,
      "loss": 3.8193,
      "theoretical_loss": 4.783575130772016,
      "tokens_seen": 98304000
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.00049,
      "loss": 3.5732,
      "theoretical_loss": 4.783132046537694,
      "tokens_seen": 98369536
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004899899699097292,
      "loss": 3.6769,
      "theoretical_loss": 4.782689339989005,
      "tokens_seen": 98435072
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004899799398194583,
      "loss": 3.5904,
      "theoretical_loss": 4.782247010552876,
      "tokens_seen": 98500608
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004899699097291876,
      "loss": 3.6234,
      "theoretical_loss": 4.781805057657483,
      "tokens_seen": 98566144
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004899598796389167,
      "loss": 3.932,
      "theoretical_loss": 4.7813634807322485,
      "tokens_seen": 98631680
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.000489949849548646,
      "loss": 3.8231,
      "theoretical_loss": 4.780922279207839,
      "tokens_seen": 98697216
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004899398194583751,
      "loss": 4.0201,
      "theoretical_loss": 4.780481452516161,
      "tokens_seen": 98762752
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004899297893681044,
      "loss": 3.8406,
      "theoretical_loss": 4.780041000090356,
      "tokens_seen": 98828288
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004899197592778335,
      "loss": 3.7729,
      "theoretical_loss": 4.779600921364795,
      "tokens_seen": 98893824
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004899097291875627,
      "loss": 3.5471,
      "theoretical_loss": 4.779161215775083,
      "tokens_seen": 98959360
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004898996990972919,
      "loss": 3.7998,
      "theoretical_loss": 4.778721882758049,
      "tokens_seen": 99024896
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004898896690070211,
      "loss": 3.5508,
      "theoretical_loss": 4.7782829217517415,
      "tokens_seen": 99090432
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004898796389167503,
      "loss": 3.8053,
      "theoretical_loss": 4.77784433219543,
      "tokens_seen": 99155968
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004898696088264794,
      "loss": 3.7494,
      "theoretical_loss": 4.777406113529597,
      "tokens_seen": 99221504
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004898595787362086,
      "loss": 3.6822,
      "theoretical_loss": 4.77696826519594,
      "tokens_seen": 99287040
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004898495486459378,
      "loss": 3.8446,
      "theoretical_loss": 4.776530786637362,
      "tokens_seen": 99352576
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.000489839518555667,
      "loss": 4.0797,
      "theoretical_loss": 4.776093677297971,
      "tokens_seen": 99418112
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004898294884653962,
      "loss": 3.6439,
      "theoretical_loss": 4.775656936623077,
      "tokens_seen": 99483648
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004898194583751254,
      "loss": 3.6336,
      "theoretical_loss": 4.77522056405919,
      "tokens_seen": 99549184
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004898094282848545,
      "loss": 3.6495,
      "theoretical_loss": 4.774784559054009,
      "tokens_seen": 99614720
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004897993981945837,
      "loss": 3.7024,
      "theoretical_loss": 4.774348921056429,
      "tokens_seen": 99680256
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.000489789368104313,
      "loss": 3.7196,
      "theoretical_loss": 4.773913649516535,
      "tokens_seen": 99745792
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004897793380140421,
      "loss": 3.7771,
      "theoretical_loss": 4.77347874388559,
      "tokens_seen": 99811328
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004897693079237714,
      "loss": 3.7941,
      "theoretical_loss": 4.773044203616044,
      "tokens_seen": 99876864
    },
    {
      "epoch": 0.03,
      "objective/train/advantage_avg": 0.49295082688331604,
      "objective/train/docs_used": 67525,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.3860185146331787,
      "objective/train/original_loss": 3.3860180377960205,
      "objective/train/theoretical_loss": 4.7726100281615205,
      "objective/train/tokens_used": 120402400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24308420717716217,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505306720733643,
      "objective/train/weighted_lm_loss": 3.557058811187744,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0338950157165527,
      "theoretical_loss": 4.7726100281615205,
      "tokens_seen": 99942400
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004897592778335005,
      "loss": 3.6832,
      "theoretical_loss": 4.7726100281615205,
      "tokens_seen": 99942400
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004897492477432297,
      "loss": 3.5841,
      "theoretical_loss": 4.772176216976822,
      "tokens_seen": 100007936
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004897392176529589,
      "loss": 3.6594,
      "theoretical_loss": 4.771742769517919,
      "tokens_seen": 100073472
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004897291875626881,
      "loss": 3.7327,
      "theoretical_loss": 4.771309685241952,
      "tokens_seen": 100139008
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004897191574724173,
      "loss": 3.6835,
      "theoretical_loss": 4.7708769636072255,
      "tokens_seen": 100204544
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004897091273821465,
      "loss": 3.6973,
      "theoretical_loss": 4.770444604073205,
      "tokens_seen": 100270080
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004896990972918756,
      "loss": 3.7637,
      "theoretical_loss": 4.770012606100515,
      "tokens_seen": 100335616
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004896890672016048,
      "loss": 3.64,
      "theoretical_loss": 4.769580969150937,
      "tokens_seen": 100401152
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.000489679037111334,
      "loss": 3.7661,
      "theoretical_loss": 4.769149692687401,
      "tokens_seen": 100466688
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004896690070210632,
      "loss": 3.8522,
      "theoretical_loss": 4.768718776173987,
      "tokens_seen": 100532224
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004896589769307924,
      "loss": 3.7927,
      "theoretical_loss": 4.76828821907592,
      "tokens_seen": 100597760
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004896489468405215,
      "loss": 3.7956,
      "theoretical_loss": 4.76785802085957,
      "tokens_seen": 100663296
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004896389167502507,
      "loss": 3.6983,
      "theoretical_loss": 4.767428180992441,
      "tokens_seen": 100728832
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004896288866599799,
      "loss": 3.6721,
      "theoretical_loss": 4.766998698943177,
      "tokens_seen": 100794368
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004896188565697091,
      "loss": 3.545,
      "theoretical_loss": 4.766569574181554,
      "tokens_seen": 100859904
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004896088264794384,
      "loss": 3.5315,
      "theoretical_loss": 4.766140806178475,
      "tokens_seen": 100925440
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004895987963891674,
      "loss": 3.6705,
      "theoretical_loss": 4.765712394405973,
      "tokens_seen": 100990976
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004895887662988967,
      "loss": 3.5902,
      "theoretical_loss": 4.765284338337203,
      "tokens_seen": 101056512
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004895787362086259,
      "loss": 3.8205,
      "theoretical_loss": 4.7648566374464405,
      "tokens_seen": 101122048
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004895687061183551,
      "loss": 3.6431,
      "theoretical_loss": 4.764429291209074,
      "tokens_seen": 101187584
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004895586760280843,
      "loss": 3.8463,
      "theoretical_loss": 4.764002299101612,
      "tokens_seen": 101253120
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004895486459378135,
      "loss": 3.4808,
      "theoretical_loss": 4.763575660601672,
      "tokens_seen": 101318656
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004895386158475426,
      "loss": 3.5684,
      "theoretical_loss": 4.763149375187977,
      "tokens_seen": 101384192
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004895285857572718,
      "loss": 3.628,
      "theoretical_loss": 4.7627234423403575,
      "tokens_seen": 101449728
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.000489518555667001,
      "loss": 3.9457,
      "theoretical_loss": 4.762297861539745,
      "tokens_seen": 101515264
    },
    {
      "epoch": 0.03,
      "objective/train/advantage_avg": 0.4940517246723175,
      "objective/train/docs_used": 68617,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.5712461471557617,
      "objective/train/original_loss": 3.571246385574341,
      "objective/train/theoretical_loss": 4.761872632268167,
      "objective/train/tokens_used": 122040800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24416404962539673,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506463050842285,
      "objective/train/weighted_lm_loss": 3.7523512840270996,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0253151655197144,
      "theoretical_loss": 4.761872632268167,
      "tokens_seen": 101580800
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004895085255767302,
      "loss": 3.7506,
      "theoretical_loss": 4.761872632268167,
      "tokens_seen": 101580800
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004894984954864594,
      "loss": 3.6104,
      "theoretical_loss": 4.761447754008753,
      "tokens_seen": 101646336
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004894884653961885,
      "loss": 3.7312,
      "theoretical_loss": 4.761023226245718,
      "tokens_seen": 101711872
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004894784353059178,
      "loss": 3.9694,
      "theoretical_loss": 4.76059904846437,
      "tokens_seen": 101777408
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004894684052156469,
      "loss": 4.0172,
      "theoretical_loss": 4.760175220151105,
      "tokens_seen": 101842944
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004894583751253762,
      "loss": 3.5839,
      "theoretical_loss": 4.7597517407934,
      "tokens_seen": 101908480
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004894483450351053,
      "loss": 3.4024,
      "theoretical_loss": 4.759328609879812,
      "tokens_seen": 101974016
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004894383149448346,
      "loss": 3.7214,
      "theoretical_loss": 4.75890582689998,
      "tokens_seen": 102039552
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004894282848545637,
      "loss": 3.7587,
      "theoretical_loss": 4.758483391344614,
      "tokens_seen": 102105088
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004894182547642929,
      "loss": 3.8431,
      "theoretical_loss": 4.758061302705496,
      "tokens_seen": 102170624
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004894082246740221,
      "loss": 3.7244,
      "theoretical_loss": 4.757639560475478,
      "tokens_seen": 102236160
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004893981945837513,
      "loss": 3.9578,
      "theoretical_loss": 4.757218164148476,
      "tokens_seen": 102301696
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004893881644934805,
      "loss": 3.5938,
      "theoretical_loss": 4.756797113219473,
      "tokens_seen": 102367232
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004893781344032096,
      "loss": 3.6824,
      "theoretical_loss": 4.756376407184508,
      "tokens_seen": 102432768
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004893681043129388,
      "loss": 3.642,
      "theoretical_loss": 4.755956045540676,
      "tokens_seen": 102498304
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.000489358074222668,
      "loss": 3.5022,
      "theoretical_loss": 4.755536027786134,
      "tokens_seen": 102563840
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004893480441323972,
      "loss": 3.746,
      "theoretical_loss": 4.75511635342008,
      "tokens_seen": 102629376
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004893380140421264,
      "loss": 3.6302,
      "theoretical_loss": 4.7546970219427696,
      "tokens_seen": 102694912
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004893279839518556,
      "loss": 3.5982,
      "theoretical_loss": 4.754278032855497,
      "tokens_seen": 102760448
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004893179538615847,
      "loss": 3.6727,
      "theoretical_loss": 4.753859385660604,
      "tokens_seen": 102825984
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004893079237713139,
      "loss": 3.8085,
      "theoretical_loss": 4.75344107986147,
      "tokens_seen": 102891520
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004892978936810432,
      "loss": 3.7106,
      "theoretical_loss": 4.753023114962512,
      "tokens_seen": 102957056
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004892878635907723,
      "loss": 3.5007,
      "theoretical_loss": 4.752605490469182,
      "tokens_seen": 103022592
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004892778335005016,
      "loss": 3.7071,
      "theoretical_loss": 4.752188205887961,
      "tokens_seen": 103088128
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004892678034102307,
      "loss": 3.8305,
      "theoretical_loss": 4.751771260726363,
      "tokens_seen": 103153664
    },
    {
      "epoch": 0.03,
      "objective/train/advantage_avg": 0.4943382143974304,
      "objective/train/docs_used": 69265,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.574159622192383,
      "objective/train/original_loss": 3.574159860610962,
      "objective/train/theoretical_loss": 4.751354654492923,
      "objective/train/tokens_used": 123679200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24443237483501434,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506763458251953,
      "objective/train/weighted_lm_loss": 3.755199432373047,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.038975715637207,
      "theoretical_loss": 4.751354654492923,
      "tokens_seen": 103219200
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004892577733199599,
      "loss": 3.7622,
      "theoretical_loss": 4.751354654492923,
      "tokens_seen": 103219200
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004892477432296891,
      "loss": 3.7381,
      "theoretical_loss": 4.750938386697204,
      "tokens_seen": 103284736
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004892377131394183,
      "loss": 3.7409,
      "theoretical_loss": 4.750522456849783,
      "tokens_seen": 103350272
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004892276830491475,
      "loss": 3.9612,
      "theoretical_loss": 4.7501068644622615,
      "tokens_seen": 103415808
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004892176529588767,
      "loss": 3.6361,
      "theoretical_loss": 4.74969160904725,
      "tokens_seen": 103481344
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004892076228686058,
      "loss": 3.7962,
      "theoretical_loss": 4.749276690118377,
      "tokens_seen": 103546880
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.000489197592778335,
      "loss": 3.7075,
      "theoretical_loss": 4.7488621071902735,
      "tokens_seen": 103612416
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004891875626880642,
      "loss": 3.4739,
      "theoretical_loss": 4.748447859778581,
      "tokens_seen": 103677952
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004891775325977934,
      "loss": 3.8039,
      "theoretical_loss": 4.7480339473999456,
      "tokens_seen": 103743488
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004891675025075226,
      "loss": 3.598,
      "theoretical_loss": 4.747620369572012,
      "tokens_seen": 103809024
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004891574724172517,
      "loss": 3.6406,
      "theoretical_loss": 4.747207125813423,
      "tokens_seen": 103874560
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004891474423269809,
      "loss": 3.8046,
      "theoretical_loss": 4.746794215643821,
      "tokens_seen": 103940096
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004891374122367101,
      "loss": 3.8226,
      "theoretical_loss": 4.746381638583839,
      "tokens_seen": 104005632
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004891273821464393,
      "loss": 3.8495,
      "theoretical_loss": 4.745969394155098,
      "tokens_seen": 104071168
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004891173520561686,
      "loss": 3.3661,
      "theoretical_loss": 4.74555748188021,
      "tokens_seen": 104136704
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004891073219658976,
      "loss": 3.9604,
      "theoretical_loss": 4.745145901282772,
      "tokens_seen": 104202240
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004890972918756269,
      "loss": 3.6761,
      "theoretical_loss": 4.744734651887361,
      "tokens_seen": 104267776
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004890872617853561,
      "loss": 3.7675,
      "theoretical_loss": 4.744323733219536,
      "tokens_seen": 104333312
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004890772316950853,
      "loss": 3.7009,
      "theoretical_loss": 4.743913144805833,
      "tokens_seen": 104398848
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004890672016048145,
      "loss": 3.7883,
      "theoretical_loss": 4.74350288617376,
      "tokens_seen": 104464384
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004890571715145437,
      "loss": 3.6222,
      "theoretical_loss": 4.7430929568517985,
      "tokens_seen": 104529920
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004890471414242728,
      "loss": 3.7689,
      "theoretical_loss": 4.742683356369401,
      "tokens_seen": 104595456
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.000489037111334002,
      "loss": 3.429,
      "theoretical_loss": 4.742274084256983,
      "tokens_seen": 104660992
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004890270812437312,
      "loss": 3.7662,
      "theoretical_loss": 4.7418651400459275,
      "tokens_seen": 104726528
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004890170511534604,
      "loss": 3.7011,
      "theoretical_loss": 4.741456523268576,
      "tokens_seen": 104792064
    },
    {
      "epoch": 0.03,
      "objective/train/advantage_avg": 0.4938974976539612,
      "objective/train/docs_used": 70556,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.4242420196533203,
      "objective/train/original_loss": 3.424241542816162,
      "objective/train/theoretical_loss": 4.741048233458233,
      "objective/train/tokens_used": 125317600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2440257966518402,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506302118301392,
      "objective/train/weighted_lm_loss": 3.597487211227417,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0437887907028198,
      "theoretical_loss": 4.741048233458233,
      "tokens_seen": 104857600
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004890070210631896,
      "loss": 3.6281,
      "theoretical_loss": 4.741048233458233,
      "tokens_seen": 104857600
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004889969909729187,
      "loss": 3.8429,
      "theoretical_loss": 4.7406402701491555,
      "tokens_seen": 104923136
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004889869608826479,
      "loss": 3.5431,
      "theoretical_loss": 4.740232632876554,
      "tokens_seen": 104988672
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004889769307923771,
      "loss": 3.7991,
      "theoretical_loss": 4.739825321176593,
      "tokens_seen": 105054208
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004889669007021063,
      "loss": 3.5498,
      "theoretical_loss": 4.739418334586386,
      "tokens_seen": 105119744
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004889568706118355,
      "loss": 3.5111,
      "theoretical_loss": 4.739011672643988,
      "tokens_seen": 105185280
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004889468405215647,
      "loss": 3.5251,
      "theoretical_loss": 4.738605334888403,
      "tokens_seen": 105250816
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004889368104312939,
      "loss": 3.6046,
      "theoretical_loss": 4.738199320859575,
      "tokens_seen": 105316352
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.000488926780341023,
      "loss": 3.6211,
      "theoretical_loss": 4.737793630098384,
      "tokens_seen": 105381888
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004889167502507523,
      "loss": 3.5136,
      "theoretical_loss": 4.737388262146647,
      "tokens_seen": 105447424
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004889067201604815,
      "loss": 3.8695,
      "theoretical_loss": 4.736983216547117,
      "tokens_seen": 105512960
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004888966900702107,
      "loss": 3.6502,
      "theoretical_loss": 4.736578492843476,
      "tokens_seen": 105578496
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004888866599799398,
      "loss": 3.6241,
      "theoretical_loss": 4.736174090580336,
      "tokens_seen": 105644032
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.000488876629889669,
      "loss": 3.6789,
      "theoretical_loss": 4.735770009303232,
      "tokens_seen": 105709568
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004888665997993982,
      "loss": 3.4279,
      "theoretical_loss": 4.735366248558629,
      "tokens_seen": 105775104
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004888565697091274,
      "loss": 3.6663,
      "theoretical_loss": 4.734962807893908,
      "tokens_seen": 105840640
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004888465396188566,
      "loss": 3.7672,
      "theoretical_loss": 4.734559686857368,
      "tokens_seen": 105906176
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004888365095285858,
      "loss": 3.7052,
      "theoretical_loss": 4.734156884998232,
      "tokens_seen": 105971712
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004888264794383149,
      "loss": 3.7029,
      "theoretical_loss": 4.733754401866626,
      "tokens_seen": 106037248
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004888164493480441,
      "loss": 3.5156,
      "theoretical_loss": 4.7333522370136,
      "tokens_seen": 106102784
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004888064192577733,
      "loss": 3.5609,
      "theoretical_loss": 4.732950389991101,
      "tokens_seen": 106168320
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004887963891675025,
      "loss": 3.6732,
      "theoretical_loss": 4.732548860351992,
      "tokens_seen": 106233856
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004887863590772317,
      "loss": 3.3571,
      "theoretical_loss": 4.732147647650036,
      "tokens_seen": 106299392
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.000488776328986961,
      "loss": 3.7172,
      "theoretical_loss": 4.731746751439898,
      "tokens_seen": 106364928
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.00048876629889669,
      "loss": 3.6971,
      "theoretical_loss": 4.731346171277146,
      "tokens_seen": 106430464
    },
    {
      "epoch": 0.03,
      "objective/train/advantage_avg": 0.49213385581970215,
      "objective/train/docs_used": 70950,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 4.185733795166016,
      "objective/train/original_loss": 4.185732841491699,
      "objective/train/theoretical_loss": 4.73094590671824,
      "objective/train/tokens_used": 126956000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24228474497795105,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0504448413848877,
      "objective/train/weighted_lm_loss": 4.396697998046875,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0363596677780151,
      "theoretical_loss": 4.73094590671824,
      "tokens_seen": 106496000
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004887562688064193,
      "loss": 3.617,
      "theoretical_loss": 4.73094590671824,
      "tokens_seen": 106496000
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004887462387161484,
      "loss": 3.6161,
      "theoretical_loss": 4.730545957320541,
      "tokens_seen": 106561536
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004887362086258777,
      "loss": 3.7525,
      "theoretical_loss": 4.730146322642297,
      "tokens_seen": 106627072
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004887261785356069,
      "loss": 3.6547,
      "theoretical_loss": 4.72974700224265,
      "tokens_seen": 106692608
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.000488716148445336,
      "loss": 3.5106,
      "theoretical_loss": 4.729347995681627,
      "tokens_seen": 106758144
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004887061183550652,
      "loss": 3.785,
      "theoretical_loss": 4.728949302520143,
      "tokens_seen": 106823680
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004886960882647944,
      "loss": 3.7584,
      "theoretical_loss": 4.728550922319995,
      "tokens_seen": 106889216
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004886860581745236,
      "loss": 3.5609,
      "theoretical_loss": 4.728152854643862,
      "tokens_seen": 106954752
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004886760280842528,
      "loss": 3.6038,
      "theoretical_loss": 4.727755099055299,
      "tokens_seen": 107020288
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004886659979939819,
      "loss": 3.6785,
      "theoretical_loss": 4.727357655118739,
      "tokens_seen": 107085824
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004886559679037111,
      "loss": 3.8328,
      "theoretical_loss": 4.726960522399492,
      "tokens_seen": 107151360
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004886459378134403,
      "loss": 3.9864,
      "theoretical_loss": 4.726563700463732,
      "tokens_seen": 107216896
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004886359077231695,
      "loss": 3.4627,
      "theoretical_loss": 4.726167188878511,
      "tokens_seen": 107282432
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004886258776328988,
      "loss": 3.6768,
      "theoretical_loss": 4.725770987211742,
      "tokens_seen": 107347968
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004886158475426278,
      "loss": 3.6744,
      "theoretical_loss": 4.725375095032207,
      "tokens_seen": 107413504
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004886058174523571,
      "loss": 3.5945,
      "theoretical_loss": 4.724979511909547,
      "tokens_seen": 107479040
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004885957873620863,
      "loss": 3.7178,
      "theoretical_loss": 4.724584237414265,
      "tokens_seen": 107544576
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004885857572718155,
      "loss": 3.4882,
      "theoretical_loss": 4.724189271117724,
      "tokens_seen": 107610112
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004885757271815447,
      "loss": 3.5483,
      "theoretical_loss": 4.72379461259214,
      "tokens_seen": 107675648
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004885656970912739,
      "loss": 3.9741,
      "theoretical_loss": 4.723400261410584,
      "tokens_seen": 107741184
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.000488555667001003,
      "loss": 3.8063,
      "theoretical_loss": 4.723006217146979,
      "tokens_seen": 107806720
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004885456369107322,
      "loss": 3.5527,
      "theoretical_loss": 4.722612479376095,
      "tokens_seen": 107872256
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004885356068204614,
      "loss": 3.611,
      "theoretical_loss": 4.722219047673552,
      "tokens_seen": 107937792
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004885255767301906,
      "loss": 4.0956,
      "theoretical_loss": 4.721825921615813,
      "tokens_seen": 108003328
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004885155466399198,
      "loss": 3.5833,
      "theoretical_loss": 4.7214331007801835,
      "tokens_seen": 108068864
    },
    {
      "epoch": 0.03,
      "objective/train/advantage_avg": 0.4960087537765503,
      "objective/train/docs_used": 72030,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.5023000240325928,
      "objective/train/original_loss": 3.502300500869751,
      "objective/train/theoretical_loss": 4.721040584744811,
      "objective/train/tokens_used": 128594400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24605385959148407,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0508517026901245,
      "objective/train/weighted_lm_loss": 3.68025541305542,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.044721245765686,
      "theoretical_loss": 4.721040584744811,
      "tokens_seen": 108134400
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.000488505516549649,
      "loss": 3.8525,
      "theoretical_loss": 4.721040584744811,
      "tokens_seen": 108134400
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004884954864593781,
      "loss": 3.6385,
      "theoretical_loss": 4.720648373088679,
      "tokens_seen": 108199936
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004884854563691073,
      "loss": 3.8062,
      "theoretical_loss": 4.720256465391609,
      "tokens_seen": 108265472
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004884754262788365,
      "loss": 3.7052,
      "theoretical_loss": 4.719864861234255,
      "tokens_seen": 108331008
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004884653961885657,
      "loss": 3.6878,
      "theoretical_loss": 4.7194735601981055,
      "tokens_seen": 108396544
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004884553660982949,
      "loss": 3.7101,
      "theoretical_loss": 4.719082561865474,
      "tokens_seen": 108462080
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004884453360080241,
      "loss": 3.6299,
      "theoretical_loss": 4.7186918658195065,
      "tokens_seen": 108527616
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004884353059177532,
      "loss": 3.8379,
      "theoretical_loss": 4.718301471644171,
      "tokens_seen": 108593152
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004884252758274825,
      "loss": 3.5608,
      "theoretical_loss": 4.71791137892426,
      "tokens_seen": 108658688
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004884152457372117,
      "loss": 3.6002,
      "theoretical_loss": 4.717521587245386,
      "tokens_seen": 108724224
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004884052156469409,
      "loss": 3.4416,
      "theoretical_loss": 4.717132096193982,
      "tokens_seen": 108789760
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.00048839518555667,
      "loss": 3.6384,
      "theoretical_loss": 4.716742905357296,
      "tokens_seen": 108855296
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004883851554663992,
      "loss": 3.5322,
      "theoretical_loss": 4.716354014323393,
      "tokens_seen": 108920832
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004883751253761284,
      "loss": 3.8615,
      "theoretical_loss": 4.7159654226811485,
      "tokens_seen": 108986368
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004883650952858576,
      "loss": 3.5328,
      "theoretical_loss": 4.71557713002025,
      "tokens_seen": 109051904
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004883550651955868,
      "loss": 3.9552,
      "theoretical_loss": 4.715189135931194,
      "tokens_seen": 109117440
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.000488345035105316,
      "loss": 3.6791,
      "theoretical_loss": 4.714801440005282,
      "tokens_seen": 109182976
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004883350050150451,
      "loss": 3.6521,
      "theoretical_loss": 4.714414041834619,
      "tokens_seen": 109248512
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004883249749247743,
      "loss": 3.6571,
      "theoretical_loss": 4.714026941012115,
      "tokens_seen": 109314048
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004883149448345035,
      "loss": 3.4738,
      "theoretical_loss": 4.713640137131478,
      "tokens_seen": 109379584
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004883049147442327,
      "loss": 3.8426,
      "theoretical_loss": 4.713253629787214,
      "tokens_seen": 109445120
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004882948846539619,
      "loss": 3.8313,
      "theoretical_loss": 4.712867418574628,
      "tokens_seen": 109510656
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004882848545636911,
      "loss": 3.3578,
      "theoretical_loss": 4.7124815030898155,
      "tokens_seen": 109576192
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004882748244734203,
      "loss": 3.5288,
      "theoretical_loss": 4.712095882929667,
      "tokens_seen": 109641728
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.00048826479438314946,
      "loss": 3.3902,
      "theoretical_loss": 4.711710557691861,
      "tokens_seen": 109707264
    },
    {
      "epoch": 0.03,
      "objective/train/advantage_avg": 0.49420469999313354,
      "objective/train/docs_used": 72698,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.0535080432891846,
      "objective/train/original_loss": 3.0535085201263428,
      "objective/train/theoretical_loss": 4.711325526974863,
      "objective/train/tokens_used": 130232800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2442803531885147,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506621599197388,
      "objective/train/weighted_lm_loss": 3.2080395221710205,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0468999147415161,
      "theoretical_loss": 4.711325526974863,
      "tokens_seen": 109772800
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.00048825476429287864,
      "loss": 3.5616,
      "theoretical_loss": 4.711325526974863,
      "tokens_seen": 109772800
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004882447342026078,
      "loss": 3.691,
      "theoretical_loss": 4.710940790377929,
      "tokens_seen": 109838336
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.000488234704112337,
      "loss": 3.6722,
      "theoretical_loss": 4.710556347501095,
      "tokens_seen": 109903872
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.00048822467402206624,
      "loss": 3.7433,
      "theoretical_loss": 4.7101721979451785,
      "tokens_seen": 109969408
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.00048821464393179536,
      "loss": 3.6341,
      "theoretical_loss": 4.709788341311781,
      "tokens_seen": 110034944
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004882046138415246,
      "loss": 3.6182,
      "theoretical_loss": 4.709404777203279,
      "tokens_seen": 110100480
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004881945837512537,
      "loss": 3.6323,
      "theoretical_loss": 4.7090215052228235,
      "tokens_seen": 110166016
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.00048818455366098296,
      "loss": 3.4054,
      "theoretical_loss": 4.708638524974344,
      "tokens_seen": 110231552
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.00048817452357071214,
      "loss": 3.6518,
      "theoretical_loss": 4.708255836062538,
      "tokens_seen": 110297088
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004881644934804413,
      "loss": 3.9933,
      "theoretical_loss": 4.707873438092877,
      "tokens_seen": 110362624
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004881544633901705,
      "loss": 3.8643,
      "theoretical_loss": 4.707491330671596,
      "tokens_seen": 110428160
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.00048814443329989974,
      "loss": 3.5789,
      "theoretical_loss": 4.707109513405701,
      "tokens_seen": 110493696
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.00048813440320962887,
      "loss": 3.7069,
      "theoretical_loss": 4.706727985902957,
      "tokens_seen": 110559232
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004881243731193581,
      "loss": 3.5565,
      "theoretical_loss": 4.706346747771898,
      "tokens_seen": 110624768
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.00048811434302908723,
      "loss": 3.6986,
      "theoretical_loss": 4.705965798621811,
      "tokens_seen": 110690304
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.00048810431293881646,
      "loss": 3.6354,
      "theoretical_loss": 4.705585138062747,
      "tokens_seen": 110755840
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.00048809428284854564,
      "loss": 3.7188,
      "theoretical_loss": 4.705204765705512,
      "tokens_seen": 110821376
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004880842527582748,
      "loss": 3.6107,
      "theoretical_loss": 4.704824681161664,
      "tokens_seen": 110886912
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.000488074222668004,
      "loss": 3.8268,
      "theoretical_loss": 4.704444884043518,
      "tokens_seen": 110952448
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004880641925777332,
      "loss": 3.1485,
      "theoretical_loss": 4.704065373964138,
      "tokens_seen": 111017984
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.00048805416248746237,
      "loss": 3.8531,
      "theoretical_loss": 4.703686150537336,
      "tokens_seen": 111083520
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004880441323971916,
      "loss": 3.5555,
      "theoretical_loss": 4.703307213377671,
      "tokens_seen": 111149056
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.00048803410230692073,
      "loss": 3.815,
      "theoretical_loss": 4.70292856210045,
      "tokens_seen": 111214592
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.00048802407221664997,
      "loss": 3.5918,
      "theoretical_loss": 4.702550196321718,
      "tokens_seen": 111280128
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.00048801404212637915,
      "loss": 3.6144,
      "theoretical_loss": 4.702172115658267,
      "tokens_seen": 111345664
    },
    {
      "epoch": 0.03,
      "objective/train/advantage_avg": 0.49525266885757446,
      "objective/train/docs_used": 74080,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.377687692642212,
      "objective/train/original_loss": 3.37768816947937,
      "objective/train/theoretical_loss": 4.701794319727625,
      "objective/train/tokens_used": 131871200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24532251060009003,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507723093032837,
      "objective/train/weighted_lm_loss": 3.5490312576293945,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0417574644088745,
      "theoretical_loss": 4.701794319727625,
      "tokens_seen": 111411200
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.00048800401203610833,
      "loss": 3.4534,
      "theoretical_loss": 4.701794319727625,
      "tokens_seen": 111411200
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004879939819458375,
      "loss": 3.5448,
      "theoretical_loss": 4.70141680814806,
      "tokens_seen": 111476736
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004879839518555667,
      "loss": 3.7637,
      "theoretical_loss": 4.701039580538573,
      "tokens_seen": 111542272
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004879739217652959,
      "loss": 3.7641,
      "theoretical_loss": 4.700662636518901,
      "tokens_seen": 111607808
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004879638916750251,
      "loss": 3.3503,
      "theoretical_loss": 4.700285975709513,
      "tokens_seen": 111673344
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004879538615847543,
      "loss": 3.6977,
      "theoretical_loss": 4.699909597731608,
      "tokens_seen": 111738880
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.00048794383149448347,
      "loss": 3.2646,
      "theoretical_loss": 4.699533502207114,
      "tokens_seen": 111804416
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.00048793380140421265,
      "loss": 3.3262,
      "theoretical_loss": 4.699157688758683,
      "tokens_seen": 111869952
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.00048792377131394183,
      "loss": 3.7805,
      "theoretical_loss": 4.698782157009695,
      "tokens_seen": 111935488
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.00048791374122367107,
      "loss": 3.8039,
      "theoretical_loss": 4.6984069065842515,
      "tokens_seen": 112001024
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004879037111334002,
      "loss": 3.7499,
      "theoretical_loss": 4.698031937107175,
      "tokens_seen": 112066560
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.00048789368104312943,
      "loss": 3.4382,
      "theoretical_loss": 4.69765724820401,
      "tokens_seen": 112132096
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.00048788365095285856,
      "loss": 3.5366,
      "theoretical_loss": 4.697282839501012,
      "tokens_seen": 112197632
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004878736208625878,
      "loss": 3.6161,
      "theoretical_loss": 4.696908710625161,
      "tokens_seen": 112263168
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.00048786359077231697,
      "loss": 3.4427,
      "theoretical_loss": 4.696534861204146,
      "tokens_seen": 112328704
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.00048785356068204615,
      "loss": 3.9622,
      "theoretical_loss": 4.696161290866366,
      "tokens_seen": 112394240
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.00048784353059177533,
      "loss": 3.165,
      "theoretical_loss": 4.695787999240936,
      "tokens_seen": 112459776
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.00048783350050150457,
      "loss": 3.4417,
      "theoretical_loss": 4.695414985957676,
      "tokens_seen": 112525312
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004878234704112337,
      "loss": 3.7395,
      "theoretical_loss": 4.695042250647113,
      "tokens_seen": 112590848
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.00048781344032096293,
      "loss": 3.4153,
      "theoretical_loss": 4.694669792940482,
      "tokens_seen": 112656384
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.00048780341023069206,
      "loss": 3.5013,
      "theoretical_loss": 4.694297612469717,
      "tokens_seen": 112721920
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004877933801404213,
      "loss": 3.5864,
      "theoretical_loss": 4.693925708867458,
      "tokens_seen": 112787456
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004877833500501505,
      "loss": 3.6215,
      "theoretical_loss": 4.693554081767042,
      "tokens_seen": 112852992
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.00048777331995987966,
      "loss": 3.6385,
      "theoretical_loss": 4.693182730802504,
      "tokens_seen": 112918528
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.00048776328986960884,
      "loss": 3.5923,
      "theoretical_loss": 4.692811655608578,
      "tokens_seen": 112984064
    },
    {
      "epoch": 0.03,
      "objective/train/advantage_avg": 0.4950334429740906,
      "objective/train/docs_used": 74817,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.8187148571014404,
      "objective/train/original_loss": 3.8187146186828613,
      "objective/train/theoretical_loss": 4.69244085582069,
      "objective/train/tokens_used": 133509600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2450971156358719,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507493019104004,
      "objective/train/weighted_lm_loss": 4.012685775756836,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0465034246444702,
      "theoretical_loss": 4.69244085582069,
      "tokens_seen": 113049600
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.000487753259779338,
      "loss": 3.6291,
      "theoretical_loss": 4.69244085582069,
      "tokens_seen": 113049600
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004877432296890672,
      "loss": 3.792,
      "theoretical_loss": 4.692070331074961,
      "tokens_seen": 113115136
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.00048773319959879644,
      "loss": 3.3907,
      "theoretical_loss": 4.691700081008202,
      "tokens_seen": 113180672
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.00048772316950852556,
      "loss": 3.6928,
      "theoretical_loss": 4.691330105257913,
      "tokens_seen": 113246208
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004877131394182548,
      "loss": 3.598,
      "theoretical_loss": 4.690960403462283,
      "tokens_seen": 113311744
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004877031093279839,
      "loss": 3.5734,
      "theoretical_loss": 4.6905909752601875,
      "tokens_seen": 113377280
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.00048769307923771316,
      "loss": 3.8307,
      "theoretical_loss": 4.690221820291184,
      "tokens_seen": 113442816
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.00048768304914744234,
      "loss": 3.4673,
      "theoretical_loss": 4.689852938195516,
      "tokens_seen": 113508352
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004876730190571715,
      "loss": 3.7283,
      "theoretical_loss": 4.689484328614107,
      "tokens_seen": 113573888
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004876629889669007,
      "loss": 3.6555,
      "theoretical_loss": 4.689115991188558,
      "tokens_seen": 113639424
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.00048765295887662994,
      "loss": 3.7319,
      "theoretical_loss": 4.688747925561149,
      "tokens_seen": 113704960
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.00048764292878635907,
      "loss": 3.629,
      "theoretical_loss": 4.688380131374835,
      "tokens_seen": 113770496
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004876328986960883,
      "loss": 3.6979,
      "theoretical_loss": 4.688012608273249,
      "tokens_seen": 113836032
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.00048762286860581743,
      "loss": 3.6673,
      "theoretical_loss": 4.687645355900694,
      "tokens_seen": 113901568
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.00048761283851554666,
      "loss": 3.4618,
      "theoretical_loss": 4.687278373902141,
      "tokens_seen": 113967104
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.00048760280842527584,
      "loss": 3.7184,
      "theoretical_loss": 4.6869116619232365,
      "tokens_seen": 114032640
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.000487592778335005,
      "loss": 3.6455,
      "theoretical_loss": 4.686545219610291,
      "tokens_seen": 114098176
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004875827482447342,
      "loss": 3.6242,
      "theoretical_loss": 4.68617904661028,
      "tokens_seen": 114163712
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004875727181544634,
      "loss": 3.5774,
      "theoretical_loss": 4.685813142570847,
      "tokens_seen": 114229248
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.00048756268806419257,
      "loss": 3.5338,
      "theoretical_loss": 4.685447507140298,
      "tokens_seen": 114294784
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004875526579739218,
      "loss": 3.641,
      "theoretical_loss": 4.685082139967594,
      "tokens_seen": 114360320
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.00048754262788365093,
      "loss": 3.7457,
      "theoretical_loss": 4.684717040702367,
      "tokens_seen": 114425856
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.00048753259779338017,
      "loss": 3.8127,
      "theoretical_loss": 4.684352208994895,
      "tokens_seen": 114491392
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.00048752256770310935,
      "loss": 3.3848,
      "theoretical_loss": 4.683987644496119,
      "tokens_seen": 114556928
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.00048751253761283853,
      "loss": 3.6187,
      "theoretical_loss": 4.683623346857635,
      "tokens_seen": 114622464
    },
    {
      "epoch": 0.03,
      "objective/train/advantage_avg": 0.4958915412425995,
      "objective/train/docs_used": 76169,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.4723758697509766,
      "objective/train/original_loss": 3.4723758697509766,
      "objective/train/theoretical_loss": 4.683259315731689,
      "objective/train/tokens_used": 135148000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24596235156059265,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0508394241333008,
      "objective/train/weighted_lm_loss": 3.648906707763672,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0417574644088745,
      "theoretical_loss": 4.683259315731689,
      "tokens_seen": 114688000
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004875025075225677,
      "loss": 3.6413,
      "theoretical_loss": 4.683259315731689,
      "tokens_seen": 114688000
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004874924774322969,
      "loss": 3.6234,
      "theoretical_loss": 4.682895550771182,
      "tokens_seen": 114753536
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.00048748244734202607,
      "loss": 3.6984,
      "theoretical_loss": 4.682532051629661,
      "tokens_seen": 114819072
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004874724172517553,
      "loss": 3.4983,
      "theoretical_loss": 4.682168817961324,
      "tokens_seen": 114884608
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.00048746238716148443,
      "loss": 3.4883,
      "theoretical_loss": 4.681805849421017,
      "tokens_seen": 114950144
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.00048745235707121367,
      "loss": 3.5033,
      "theoretical_loss": 4.681443145664227,
      "tokens_seen": 115015680
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004874423269809428,
      "loss": 3.2676,
      "theoretical_loss": 4.681080706347089,
      "tokens_seen": 115081216
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.00048743229689067203,
      "loss": 3.5586,
      "theoretical_loss": 4.680718531126377,
      "tokens_seen": 115146752
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004874222668004012,
      "loss": 3.6984,
      "theoretical_loss": 4.6803566196595074,
      "tokens_seen": 115212288
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004874122367101304,
      "loss": 3.6818,
      "theoretical_loss": 4.6799949716045335,
      "tokens_seen": 115277824
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.0004874022066198596,
      "loss": 3.3176,
      "theoretical_loss": 4.679633586620149,
      "tokens_seen": 115343360
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.00048739217652958876,
      "loss": 3.528,
      "theoretical_loss": 4.679272464365681,
      "tokens_seen": 115408896
    },
    {
      "epoch": 0.03,
      "learning_rate": 0.00048738214643931794,
      "loss": 3.8141,
      "theoretical_loss": 4.678911604501092,
      "tokens_seen": 115474432
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048737211634904717,
      "loss": 3.4884,
      "theoretical_loss": 4.678551006686977,
      "tokens_seen": 115539968
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004873620862587763,
      "loss": 3.5383,
      "theoretical_loss": 4.678190670584562,
      "tokens_seen": 115605504
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048735205616850553,
      "loss": 3.6703,
      "theoretical_loss": 4.677830595855703,
      "tokens_seen": 115671040
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004873420260782347,
      "loss": 3.8146,
      "theoretical_loss": 4.6774707821628825,
      "tokens_seen": 115736576
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004873319959879639,
      "loss": 3.5697,
      "theoretical_loss": 4.677111229169213,
      "tokens_seen": 115802112
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004873219658976931,
      "loss": 3.6391,
      "theoretical_loss": 4.676751936538429,
      "tokens_seen": 115867648
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048731193580742226,
      "loss": 3.5482,
      "theoretical_loss": 4.676392903934891,
      "tokens_seen": 115933184
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048730190571715144,
      "loss": 3.5471,
      "theoretical_loss": 4.676034131023581,
      "tokens_seen": 115998720
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004872918756268807,
      "loss": 3.5437,
      "theoretical_loss": 4.675675617470099,
      "tokens_seen": 116064256
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004872818455366098,
      "loss": 3.4257,
      "theoretical_loss": 4.6753173629406675,
      "tokens_seen": 116129792
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048727181544633904,
      "loss": 3.5085,
      "theoretical_loss": 4.674959367102126,
      "tokens_seen": 116195328
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048726178535606816,
      "loss": 3.4123,
      "theoretical_loss": 4.674601629621927,
      "tokens_seen": 116260864
    },
    {
      "epoch": 0.04,
      "objective/train/advantage_avg": 0.4944210350513458,
      "objective/train/docs_used": 76931,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.294889450073242,
      "objective/train/original_loss": 3.294889450073242,
      "objective/train/theoretical_loss": 4.674244150168143,
      "objective/train/tokens_used": 136786400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24453645944595337,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050685167312622,
      "objective/train/weighted_lm_loss": 3.46191668510437,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0338950157165527,
      "theoretical_loss": 4.674244150168143,
      "tokens_seen": 116326400
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004872517552657974,
      "loss": 3.0672,
      "theoretical_loss": 4.674244150168143,
      "tokens_seen": 116326400
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004872417251755266,
      "loss": 3.571,
      "theoretical_loss": 4.673886928409454,
      "tokens_seen": 116391936
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048723169508525576,
      "loss": 3.7734,
      "theoretical_loss": 4.673529964015156,
      "tokens_seen": 116457472
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.000487221664994985,
      "loss": 3.5472,
      "theoretical_loss": 4.6731732566551525,
      "tokens_seen": 116523008
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004872116349047141,
      "loss": 3.3961,
      "theoretical_loss": 4.672816805999958,
      "tokens_seen": 116588544
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048720160481444336,
      "loss": 3.6973,
      "theoretical_loss": 4.672460611720694,
      "tokens_seen": 116654080
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048719157472417254,
      "loss": 3.7933,
      "theoretical_loss": 4.672104673489084,
      "tokens_seen": 116719616
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004871815446339017,
      "loss": 3.6317,
      "theoretical_loss": 4.67174899097746,
      "tokens_seen": 116785152
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004871715145436309,
      "loss": 3.6462,
      "theoretical_loss": 4.671393563858756,
      "tokens_seen": 116850688
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048716148445336014,
      "loss": 3.7168,
      "theoretical_loss": 4.671038391806507,
      "tokens_seen": 116916224
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048715145436308927,
      "loss": 3.7213,
      "theoretical_loss": 4.670683474494848,
      "tokens_seen": 116981760
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004871414242728185,
      "loss": 3.3068,
      "theoretical_loss": 4.670328811598512,
      "tokens_seen": 117047296
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048713139418254763,
      "loss": 3.8156,
      "theoretical_loss": 4.669974402792832,
      "tokens_seen": 117112832
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048712136409227686,
      "loss": 3.3729,
      "theoretical_loss": 4.6696202477537305,
      "tokens_seen": 117178368
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048711133400200604,
      "loss": 3.3147,
      "theoretical_loss": 4.669266346157732,
      "tokens_seen": 117243904
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004871013039117352,
      "loss": 3.752,
      "theoretical_loss": 4.668912697681949,
      "tokens_seen": 117309440
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004870912738214644,
      "loss": 3.6959,
      "theoretical_loss": 4.668559302004087,
      "tokens_seen": 117374976
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004870812437311936,
      "loss": 3.6343,
      "theoretical_loss": 4.668206158802439,
      "tokens_seen": 117440512
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048707121364092277,
      "loss": 3.6117,
      "theoretical_loss": 4.667853267755892,
      "tokens_seen": 117506048
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.000487061183550652,
      "loss": 3.4073,
      "theoretical_loss": 4.667500628543914,
      "tokens_seen": 117571584
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048705115346038113,
      "loss": 3.4449,
      "theoretical_loss": 4.667148240846565,
      "tokens_seen": 117637120
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048704112337011037,
      "loss": 3.6773,
      "theoretical_loss": 4.666796104344483,
      "tokens_seen": 117702656
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048703109327983955,
      "loss": 3.6765,
      "theoretical_loss": 4.666444218718894,
      "tokens_seen": 117768192
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048702106318956873,
      "loss": 3.5828,
      "theoretical_loss": 4.666092583651603,
      "tokens_seen": 117833728
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004870110330992979,
      "loss": 3.6428,
      "theoretical_loss": 4.665741198824998,
      "tokens_seen": 117899264
    },
    {
      "epoch": 0.04,
      "objective/train/advantage_avg": 0.49537116289138794,
      "objective/train/docs_used": 78354,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.918903350830078,
      "objective/train/original_loss": 2.9189035892486572,
      "objective/train/theoretical_loss": 4.665390063922041,
      "objective/train/tokens_used": 138424800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.245442733168602,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507848262786865,
      "objective/train/weighted_lm_loss": 3.0670511722564697,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0338950157165527,
      "theoretical_loss": 4.665390063922041,
      "tokens_seen": 117964800
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004870010030090271,
      "loss": 3.6994,
      "theoretical_loss": 4.665390063922041,
      "tokens_seen": 117964800
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048699097291875627,
      "loss": 3.3385,
      "theoretical_loss": 4.665039178626278,
      "tokens_seen": 118030336
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004869809428284855,
      "loss": 3.5874,
      "theoretical_loss": 4.664688542621825,
      "tokens_seen": 118095872
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048697091273821463,
      "loss": 3.5986,
      "theoretical_loss": 4.664338155593377,
      "tokens_seen": 118161408
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048696088264794387,
      "loss": 3.7081,
      "theoretical_loss": 4.6639880172262,
      "tokens_seen": 118226944
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.000486950852557673,
      "loss": 3.5448,
      "theoretical_loss": 4.6636381272061325,
      "tokens_seen": 118292480
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048694082246740223,
      "loss": 3.4524,
      "theoretical_loss": 4.663288485219585,
      "tokens_seen": 118358016
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004869307923771314,
      "loss": 3.6859,
      "theoretical_loss": 4.662939090953536,
      "tokens_seen": 118423552
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004869207622868606,
      "loss": 3.6192,
      "theoretical_loss": 4.662589944095533,
      "tokens_seen": 118489088
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004869107321965898,
      "loss": 3.6893,
      "theoretical_loss": 4.662241044333689,
      "tokens_seen": 118554624
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048690070210631896,
      "loss": 3.6706,
      "theoretical_loss": 4.661892391356682,
      "tokens_seen": 118620160
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048689067201604814,
      "loss": 3.0619,
      "theoretical_loss": 4.661543984853757,
      "tokens_seen": 118685696
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048688064192577737,
      "loss": 3.4726,
      "theoretical_loss": 4.661195824514717,
      "tokens_seen": 118751232
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004868706118355065,
      "loss": 3.4025,
      "theoretical_loss": 4.660847910029929,
      "tokens_seen": 118816768
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048686058174523573,
      "loss": 3.5865,
      "theoretical_loss": 4.660500241090322,
      "tokens_seen": 118882304
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004868505516549649,
      "loss": 3.471,
      "theoretical_loss": 4.660152817387377,
      "tokens_seen": 118947840
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004868405215646941,
      "loss": 3.5847,
      "theoretical_loss": 4.6598056386131415,
      "tokens_seen": 119013376
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004868304914744233,
      "loss": 3.4791,
      "theoretical_loss": 4.659458704460212,
      "tokens_seen": 119078912
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048682046138415246,
      "loss": 3.5631,
      "theoretical_loss": 4.6591120146217415,
      "tokens_seen": 119144448
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048681043129388164,
      "loss": 3.608,
      "theoretical_loss": 4.658765568791438,
      "tokens_seen": 119209984
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004868004012036109,
      "loss": 3.607,
      "theoretical_loss": 4.6584193666635585,
      "tokens_seen": 119275520
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048679037111334,
      "loss": 3.5626,
      "theoretical_loss": 4.658073407932915,
      "tokens_seen": 119341056
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048678034102306924,
      "loss": 3.6744,
      "theoretical_loss": 4.657727692294865,
      "tokens_seen": 119406592
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048677031093279836,
      "loss": 3.7958,
      "theoretical_loss": 4.657382219445317,
      "tokens_seen": 119472128
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004867602808425276,
      "loss": 3.5384,
      "theoretical_loss": 4.657036989080726,
      "tokens_seen": 119537664
    },
    {
      "epoch": 0.04,
      "objective/train/advantage_avg": 0.49456217885017395,
      "objective/train/docs_used": 78905,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.1790289878845215,
      "objective/train/original_loss": 3.1790294647216797,
      "objective/train/theoretical_loss": 4.656692000898091,
      "objective/train/tokens_used": 140063200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2446637600660324,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506999492645264,
      "objective/train/weighted_lm_loss": 3.3398995399475098,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0408108234405518,
      "theoretical_loss": 4.656692000898091,
      "tokens_seen": 119603200
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004867502507522568,
      "loss": 3.4968,
      "theoretical_loss": 4.656692000898091,
      "tokens_seen": 119603200
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048674022066198596,
      "loss": 3.6052,
      "theoretical_loss": 4.656347254594957,
      "tokens_seen": 119668736
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048673019057171514,
      "loss": 3.576,
      "theoretical_loss": 4.656002749869411,
      "tokens_seen": 119734272
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004867201604814443,
      "loss": 3.4327,
      "theoretical_loss": 4.655658486420084,
      "tokens_seen": 119799808
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004867101303911735,
      "loss": 3.6975,
      "theoretical_loss": 4.655314463946143,
      "tokens_seen": 119865344
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048670010030090274,
      "loss": 3.6398,
      "theoretical_loss": 4.654970682147299,
      "tokens_seen": 119930880
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048669007021063187,
      "loss": 3.3111,
      "theoretical_loss": 4.6546271407238,
      "tokens_seen": 119996416
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004866800401203611,
      "loss": 3.6706,
      "theoretical_loss": 4.654283839376431,
      "tokens_seen": 120061952
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004866700100300903,
      "loss": 3.3817,
      "theoretical_loss": 4.653940777806509,
      "tokens_seen": 120127488
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048665997993981947,
      "loss": 3.7451,
      "theoretical_loss": 4.653597955715888,
      "tokens_seen": 120193024
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048664994984954865,
      "loss": 3.6022,
      "theoretical_loss": 4.653255372806958,
      "tokens_seen": 120258560
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048663991975927783,
      "loss": 3.8613,
      "theoretical_loss": 4.652913028782635,
      "tokens_seen": 120324096
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.000486629889669007,
      "loss": 3.6821,
      "theoretical_loss": 4.652570923346367,
      "tokens_seen": 120389632
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048661985957873624,
      "loss": 3.5022,
      "theoretical_loss": 4.6522290562021364,
      "tokens_seen": 120455168
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048660982948846537,
      "loss": 3.6815,
      "theoretical_loss": 4.651887427054448,
      "tokens_seen": 120520704
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004865997993981946,
      "loss": 3.7667,
      "theoretical_loss": 4.651546035608336,
      "tokens_seen": 120586240
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048658976930792373,
      "loss": 3.9327,
      "theoretical_loss": 4.651204881569357,
      "tokens_seen": 120651776
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048657973921765297,
      "loss": 3.5794,
      "theoretical_loss": 4.650863964643597,
      "tokens_seen": 120717312
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048656970912738215,
      "loss": 3.6899,
      "theoretical_loss": 4.6505232845376625,
      "tokens_seen": 120782848
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048655967903711133,
      "loss": 3.6864,
      "theoretical_loss": 4.650182840958681,
      "tokens_seen": 120848384
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004865496489468405,
      "loss": 3.5708,
      "theoretical_loss": 4.649842633614304,
      "tokens_seen": 120913920
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048653961885656975,
      "loss": 3.6314,
      "theoretical_loss": 4.649502662212699,
      "tokens_seen": 120979456
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004865295887662989,
      "loss": 3.5018,
      "theoretical_loss": 4.649162926462553,
      "tokens_seen": 121044992
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004865195586760281,
      "loss": 3.8503,
      "theoretical_loss": 4.648823426073072,
      "tokens_seen": 121110528
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048650952858575724,
      "loss": 3.6371,
      "theoretical_loss": 4.648484160753975,
      "tokens_seen": 121176064
    },
    {
      "epoch": 0.04,
      "objective/train/advantage_avg": 0.49477940797805786,
      "objective/train/docs_used": 79692,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.9780516624450684,
      "objective/train/original_loss": 2.9780519008636475,
      "objective/train/theoretical_loss": 4.648145130215498,
      "objective/train/tokens_used": 141701600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24484485387802124,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050722599029541,
      "objective/train/weighted_lm_loss": 3.1290273666381836,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0457526445388794,
      "theoretical_loss": 4.648145130215498,
      "tokens_seen": 121241600
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048649949849548647,
      "loss": 3.6362,
      "theoretical_loss": 4.648145130215498,
      "tokens_seen": 121241600
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048648946840521565,
      "loss": 3.6149,
      "theoretical_loss": 4.64780633416839,
      "tokens_seen": 121307136
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048647943831494483,
      "loss": 3.8188,
      "theoretical_loss": 4.647467772323912,
      "tokens_seen": 121372672
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048646940822467407,
      "loss": 3.2255,
      "theoretical_loss": 4.647129444393837,
      "tokens_seen": 121438208
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004864593781344032,
      "loss": 3.7036,
      "theoretical_loss": 4.646791350090446,
      "tokens_seen": 121503744
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048644934804413243,
      "loss": 3.8621,
      "theoretical_loss": 4.646453489126531,
      "tokens_seen": 121569280
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004864393179538616,
      "loss": 3.4012,
      "theoretical_loss": 4.646115861215389,
      "tokens_seen": 121634816
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004864292878635908,
      "loss": 3.3391,
      "theoretical_loss": 4.645778466070829,
      "tokens_seen": 121700352
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048641925777332,
      "loss": 3.3744,
      "theoretical_loss": 4.64544130340716,
      "tokens_seen": 121765888
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048640922768304916,
      "loss": 3.4794,
      "theoretical_loss": 4.645104372939196,
      "tokens_seen": 121831424
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048639919759277834,
      "loss": 3.4479,
      "theoretical_loss": 4.644767674382256,
      "tokens_seen": 121896960
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048638916750250757,
      "loss": 3.7956,
      "theoretical_loss": 4.644431207452159,
      "tokens_seen": 121962496
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004863791374122367,
      "loss": 3.6267,
      "theoretical_loss": 4.6440949718652265,
      "tokens_seen": 122028032
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048636910732196593,
      "loss": 3.2961,
      "theoretical_loss": 4.643758967338277,
      "tokens_seen": 122093568
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004863590772316951,
      "loss": 3.5167,
      "theoretical_loss": 4.643423193588631,
      "tokens_seen": 122159104
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004863490471414243,
      "loss": 3.7848,
      "theoretical_loss": 4.643087650334102,
      "tokens_seen": 122224640
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004863390170511535,
      "loss": 3.4878,
      "theoretical_loss": 4.642752337293002,
      "tokens_seen": 122290176
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048632898696088266,
      "loss": 3.66,
      "theoretical_loss": 4.642417254184139,
      "tokens_seen": 122355712
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048631895687061184,
      "loss": 3.723,
      "theoretical_loss": 4.642082400726812,
      "tokens_seen": 122421248
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004863089267803411,
      "loss": 3.7939,
      "theoretical_loss": 4.641747776640816,
      "tokens_seen": 122486784
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004862988966900702,
      "loss": 3.6504,
      "theoretical_loss": 4.641413381646435,
      "tokens_seen": 122552320
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048628886659979944,
      "loss": 3.5785,
      "theoretical_loss": 4.641079215464444,
      "tokens_seen": 122617856
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048627883650952857,
      "loss": 3.6547,
      "theoretical_loss": 4.640745277816107,
      "tokens_seen": 122683392
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004862688064192578,
      "loss": 3.4997,
      "theoretical_loss": 4.640411568423177,
      "tokens_seen": 122748928
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.000486258776328987,
      "loss": 3.2168,
      "theoretical_loss": 4.640078087007897,
      "tokens_seen": 122814464
    },
    {
      "epoch": 0.04,
      "objective/train/advantage_avg": 0.4923751652240753,
      "objective/train/docs_used": 81051,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.282017707824707,
      "objective/train/original_loss": 3.282017469406128,
      "objective/train/theoretical_loss": 4.639744833292989,
      "objective/train/tokens_used": 143340000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24251699447631836,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050470232963562,
      "objective/train/weighted_lm_loss": 3.447389602661133,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0432155132293701,
      "theoretical_loss": 4.639744833292989,
      "tokens_seen": 122880000
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048624874623871616,
      "loss": 3.6865,
      "theoretical_loss": 4.639744833292989,
      "tokens_seen": 122880000
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048623871614844534,
      "loss": 3.6306,
      "theoretical_loss": 4.639411807001665,
      "tokens_seen": 122945536
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004862286860581745,
      "loss": 3.5801,
      "theoretical_loss": 4.639079007857621,
      "tokens_seen": 123011072
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004862186559679037,
      "loss": 3.5626,
      "theoretical_loss": 4.638746435585032,
      "tokens_seen": 123076608
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048620862587763294,
      "loss": 3.5216,
      "theoretical_loss": 4.638414089908558,
      "tokens_seen": 123142144
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048619859578736207,
      "loss": 3.4378,
      "theoretical_loss": 4.63808197055334,
      "tokens_seen": 123207680
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004861885656970913,
      "loss": 3.3797,
      "theoretical_loss": 4.637750077244993,
      "tokens_seen": 123273216
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004861785356068205,
      "loss": 3.4362,
      "theoretical_loss": 4.637418409709617,
      "tokens_seen": 123338752
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048616850551654967,
      "loss": 3.6062,
      "theoretical_loss": 4.637086967673785,
      "tokens_seen": 123404288
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048615847542627885,
      "loss": 3.4944,
      "theoretical_loss": 4.6367557508645465,
      "tokens_seen": 123469824
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048614844533600803,
      "loss": 3.7952,
      "theoretical_loss": 4.6364247590094285,
      "tokens_seen": 123535360
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004861384152457372,
      "loss": 3.6225,
      "theoretical_loss": 4.636093991836429,
      "tokens_seen": 123600896
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048612838515546644,
      "loss": 3.6744,
      "theoretical_loss": 4.635763449074021,
      "tokens_seen": 123666432
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048611835506519557,
      "loss": 3.3957,
      "theoretical_loss": 4.635433130451148,
      "tokens_seen": 123731968
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004861083249749248,
      "loss": 3.6043,
      "theoretical_loss": 4.635103035697226,
      "tokens_seen": 123797504
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048609829488465393,
      "loss": 3.5416,
      "theoretical_loss": 4.634773164542136,
      "tokens_seen": 123863040
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048608826479438317,
      "loss": 3.4396,
      "theoretical_loss": 4.634443516716235,
      "tokens_seen": 123928576
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048607823470411235,
      "loss": 3.5183,
      "theoretical_loss": 4.634114091950341,
      "tokens_seen": 123994112
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048606820461384153,
      "loss": 3.4985,
      "theoretical_loss": 4.633784889975743,
      "tokens_seen": 124059648
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004860581745235707,
      "loss": 3.3074,
      "theoretical_loss": 4.633455910524191,
      "tokens_seen": 124125184
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048604814443329995,
      "loss": 3.3869,
      "theoretical_loss": 4.633127153327905,
      "tokens_seen": 124190720
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004860381143430291,
      "loss": 3.78,
      "theoretical_loss": 4.632798618119562,
      "tokens_seen": 124256256
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004860280842527583,
      "loss": 3.5845,
      "theoretical_loss": 4.6324703046323075,
      "tokens_seen": 124321792
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048601805416248744,
      "loss": 3.6042,
      "theoretical_loss": 4.632142212599742,
      "tokens_seen": 124387328
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048600802407221667,
      "loss": 3.6129,
      "theoretical_loss": 4.631814341755933,
      "tokens_seen": 124452864
    },
    {
      "epoch": 0.04,
      "objective/train/advantage_avg": 0.49466538429260254,
      "objective/train/docs_used": 81846,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.6649770736694336,
      "objective/train/original_loss": 3.664978265762329,
      "objective/train/theoretical_loss": 4.631486691835402,
      "objective/train/tokens_used": 144978400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24475476145744324,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507107973098755,
      "objective/train/weighted_lm_loss": 3.8507893085479736,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0432155132293701,
      "theoretical_loss": 4.631486691835402,
      "tokens_seen": 124518400
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048599799398194585,
      "loss": 3.5991,
      "theoretical_loss": 4.631486691835402,
      "tokens_seen": 124518400
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048598796389167503,
      "loss": 3.663,
      "theoretical_loss": 4.631159262573131,
      "tokens_seen": 124583936
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004859779338014042,
      "loss": 3.706,
      "theoretical_loss": 4.630832053704557,
      "tokens_seen": 124649472
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004859679037111334,
      "loss": 3.3759,
      "theoretical_loss": 4.6305050649655755,
      "tokens_seen": 124715008
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004859578736208626,
      "loss": 3.4685,
      "theoretical_loss": 4.630178296092535,
      "tokens_seen": 124780544
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004859478435305918,
      "loss": 3.5424,
      "theoretical_loss": 4.629851746822241,
      "tokens_seen": 124846080
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048593781344032094,
      "loss": 3.623,
      "theoretical_loss": 4.6295254168919495,
      "tokens_seen": 124911616
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004859277833500502,
      "loss": 3.5506,
      "theoretical_loss": 4.629199306039368,
      "tokens_seen": 124977152
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004859177532597793,
      "loss": 3.5902,
      "theoretical_loss": 4.628873414002657,
      "tokens_seen": 125042688
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048590772316950854,
      "loss": 3.7212,
      "theoretical_loss": 4.6285477405204265,
      "tokens_seen": 125108224
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004858976930792377,
      "loss": 3.4427,
      "theoretical_loss": 4.628222285331734,
      "tokens_seen": 125173760
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004858876629889669,
      "loss": 3.2904,
      "theoretical_loss": 4.627897048176086,
      "tokens_seen": 125239296
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004858776328986961,
      "loss": 3.7141,
      "theoretical_loss": 4.627572028793438,
      "tokens_seen": 125304832
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004858676028084253,
      "loss": 3.4193,
      "theoretical_loss": 4.627247226924187,
      "tokens_seen": 125370368
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048585757271815444,
      "loss": 3.7142,
      "theoretical_loss": 4.626922642309179,
      "tokens_seen": 125435904
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004858475426278837,
      "loss": 3.7087,
      "theoretical_loss": 4.6265982746897,
      "tokens_seen": 125501440
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004858375125376128,
      "loss": 3.6295,
      "theoretical_loss": 4.626274123807484,
      "tokens_seen": 125566976
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048582748244734204,
      "loss": 3.2822,
      "theoretical_loss": 4.625950189404703,
      "tokens_seen": 125632512
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004858174523570712,
      "loss": 3.4699,
      "theoretical_loss": 4.625626471223972,
      "tokens_seen": 125698048
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004858074222668004,
      "loss": 3.5334,
      "theoretical_loss": 4.625302969008344,
      "tokens_seen": 125763584
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004857973921765296,
      "loss": 3.622,
      "theoretical_loss": 4.624979682501314,
      "tokens_seen": 125829120
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048578736208625877,
      "loss": 3.5481,
      "theoretical_loss": 4.624656611446811,
      "tokens_seen": 125894656
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048577733199598795,
      "loss": 3.6023,
      "theoretical_loss": 4.624333755589207,
      "tokens_seen": 125960192
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004857673019057172,
      "loss": 3.5516,
      "theoretical_loss": 4.624011114673303,
      "tokens_seen": 126025728
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004857572718154463,
      "loss": 3.6768,
      "theoretical_loss": 4.623688688444341,
      "tokens_seen": 126091264
    },
    {
      "epoch": 0.04,
      "objective/train/advantage_avg": 0.49431848526000977,
      "objective/train/docs_used": 83055,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.7771947383880615,
      "objective/train/original_loss": 3.7771952152252197,
      "objective/train/theoretical_loss": 4.623366476647994,
      "objective/train/tokens_used": 146616800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24441944062709808,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050674319267273,
      "objective/train/weighted_lm_loss": 3.9686665534973145,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0304546356201172,
      "theoretical_loss": 4.623366476647994,
      "tokens_seen": 126156800
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048574724172517554,
      "loss": 3.6504,
      "theoretical_loss": 4.623366476647994,
      "tokens_seen": 126156800
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048573721163490467,
      "loss": 3.5467,
      "theoretical_loss": 4.623044479030369,
      "tokens_seen": 126222336
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004857271815446339,
      "loss": 3.7043,
      "theoretical_loss": 4.622722695338005,
      "tokens_seen": 126287872
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048571715145436314,
      "loss": 3.3545,
      "theoretical_loss": 4.622401125317873,
      "tokens_seen": 126353408
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048570712136409227,
      "loss": 3.3494,
      "theoretical_loss": 4.622079768717375,
      "tokens_seen": 126418944
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004856970912738215,
      "loss": 3.5576,
      "theoretical_loss": 4.62175862528434,
      "tokens_seen": 126484480
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004856870611835507,
      "loss": 3.3641,
      "theoretical_loss": 4.621437694767026,
      "tokens_seen": 126550016
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048567703109327987,
      "loss": 3.6035,
      "theoretical_loss": 4.621116976914122,
      "tokens_seen": 126615552
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048566700100300905,
      "loss": 3.4448,
      "theoretical_loss": 4.620796471474737,
      "tokens_seen": 126681088
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048565697091273823,
      "loss": 3.647,
      "theoretical_loss": 4.620476178198411,
      "tokens_seen": 126746624
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004856469408224674,
      "loss": 3.2373,
      "theoretical_loss": 4.620156096835108,
      "tokens_seen": 126812160
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048563691073219664,
      "loss": 3.7293,
      "theoretical_loss": 4.619836227135212,
      "tokens_seen": 126877696
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048562688064192577,
      "loss": 3.7161,
      "theoretical_loss": 4.619516568849535,
      "tokens_seen": 126943232
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.000485616850551655,
      "loss": 3.6985,
      "theoretical_loss": 4.619197121729307,
      "tokens_seen": 127008768
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048560682046138413,
      "loss": 3.5833,
      "theoretical_loss": 4.61887788552618,
      "tokens_seen": 127074304
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048559679037111337,
      "loss": 3.5624,
      "theoretical_loss": 4.618558859992225,
      "tokens_seen": 127139840
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048558676028084255,
      "loss": 3.4075,
      "theoretical_loss": 4.6182400448799354,
      "tokens_seen": 127205376
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048557673019057173,
      "loss": 3.7672,
      "theoretical_loss": 4.617921439942219,
      "tokens_seen": 127270912
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004855667001003009,
      "loss": 3.1485,
      "theoretical_loss": 4.617603044932402,
      "tokens_seen": 127336448
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048555667001003015,
      "loss": 3.7475,
      "theoretical_loss": 4.617284859604227,
      "tokens_seen": 127401984
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004855466399197593,
      "loss": 3.6192,
      "theoretical_loss": 4.616966883711853,
      "tokens_seen": 127467520
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004855366098294885,
      "loss": 3.5069,
      "theoretical_loss": 4.616649117009851,
      "tokens_seen": 127533056
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048552657973921764,
      "loss": 3.6108,
      "theoretical_loss": 4.616331559253208,
      "tokens_seen": 127598592
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048551654964894687,
      "loss": 3.6585,
      "theoretical_loss": 4.6160142101973225,
      "tokens_seen": 127664128
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048550651955867605,
      "loss": 3.4267,
      "theoretical_loss": 4.6156970695980055,
      "tokens_seen": 127729664
    },
    {
      "epoch": 0.04,
      "objective/train/advantage_avg": 0.49420952796936035,
      "objective/train/docs_used": 83592,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.0786585807800293,
      "objective/train/original_loss": 3.0786588191986084,
      "objective/train/theoretical_loss": 4.615380137211477,
      "objective/train/tokens_used": 148255200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24435743689537048,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506631135940552,
      "objective/train/weighted_lm_loss": 3.234408378601074,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0168063640594482,
      "theoretical_loss": 4.615380137211477,
      "tokens_seen": 127795200
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048549648946840523,
      "loss": 3.3436,
      "theoretical_loss": 4.615380137211477,
      "tokens_seen": 127795200
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004854864593781344,
      "loss": 3.6201,
      "theoretical_loss": 4.615063412794369,
      "tokens_seen": 127860736
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004854764292878636,
      "loss": 3.8364,
      "theoretical_loss": 4.6147468961037195,
      "tokens_seen": 127926272
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004854663991975928,
      "loss": 3.5317,
      "theoretical_loss": 4.6144305868969795,
      "tokens_seen": 127991808
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.000485456369107322,
      "loss": 3.7012,
      "theoretical_loss": 4.614114484932002,
      "tokens_seen": 128057344
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048544633901705114,
      "loss": 3.3252,
      "theoretical_loss": 4.61379858996705,
      "tokens_seen": 128122880
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004854363089267804,
      "loss": 3.3984,
      "theoretical_loss": 4.613482901760792,
      "tokens_seen": 128188416
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004854262788365095,
      "loss": 3.7604,
      "theoretical_loss": 4.613167420072298,
      "tokens_seen": 128253952
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048541624874623874,
      "loss": 3.478,
      "theoretical_loss": 4.612852144661042,
      "tokens_seen": 128319488
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004854062186559679,
      "loss": 3.5069,
      "theoretical_loss": 4.612537075286906,
      "tokens_seen": 128385024
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004853961885656971,
      "loss": 3.8256,
      "theoretical_loss": 4.612222211710168,
      "tokens_seen": 128450560
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004853861584754263,
      "loss": 3.4809,
      "theoretical_loss": 4.611907553691509,
      "tokens_seen": 128516096
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004853761283851555,
      "loss": 3.5477,
      "theoretical_loss": 4.61159310099201,
      "tokens_seen": 128581632
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048536609829488464,
      "loss": 3.7691,
      "theoretical_loss": 4.611278853373152,
      "tokens_seen": 128647168
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004853560682046139,
      "loss": 3.583,
      "theoretical_loss": 4.6109648105968155,
      "tokens_seen": 128712704
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.000485346038114343,
      "loss": 3.6878,
      "theoretical_loss": 4.610650972425274,
      "tokens_seen": 128778240
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048533600802407224,
      "loss": 3.3567,
      "theoretical_loss": 4.610337338621205,
      "tokens_seen": 128843776
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004853259779338014,
      "loss": 3.6824,
      "theoretical_loss": 4.6100239089476736,
      "tokens_seen": 128909312
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004853159478435306,
      "loss": 3.5216,
      "theoretical_loss": 4.609710683168146,
      "tokens_seen": 128974848
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004853059177532598,
      "loss": 3.6159,
      "theoretical_loss": 4.6093976610464775,
      "tokens_seen": 129040384
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048529588766298897,
      "loss": 3.4138,
      "theoretical_loss": 4.609084842346924,
      "tokens_seen": 129105920
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048528585757271815,
      "loss": 3.3309,
      "theoretical_loss": 4.608772226834126,
      "tokens_seen": 129171456
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004852758274824474,
      "loss": 3.6673,
      "theoretical_loss": 4.60845981427312,
      "tokens_seen": 129236992
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004852657973921765,
      "loss": 3.597,
      "theoretical_loss": 4.60814760442933,
      "tokens_seen": 129302528
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048525576730190574,
      "loss": 3.6571,
      "theoretical_loss": 4.607835597068576,
      "tokens_seen": 129368064
    },
    {
      "epoch": 0.04,
      "objective/train/advantage_avg": 0.494545578956604,
      "objective/train/docs_used": 84814,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.5531492233276367,
      "objective/train/original_loss": 3.553149461746216,
      "objective/train/theoretical_loss": 4.607523791957059,
      "objective/train/tokens_used": 149893600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24464766681194305,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050698161125183,
      "objective/train/weighted_lm_loss": 3.7331182956695557,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0363596677780151,
      "theoretical_loss": 4.607523791957059,
      "tokens_seen": 129433600
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048524573721163487,
      "loss": 3.5945,
      "theoretical_loss": 4.607523791957059,
      "tokens_seen": 129433600
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004852357071213641,
      "loss": 3.7052,
      "theoretical_loss": 4.607212188861373,
      "tokens_seen": 129499136
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004852256770310933,
      "loss": 3.617,
      "theoretical_loss": 4.606900787548499,
      "tokens_seen": 129564672
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048521564694082247,
      "loss": 3.5905,
      "theoretical_loss": 4.606589587785802,
      "tokens_seen": 129630208
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048520561685055165,
      "loss": 3.689,
      "theoretical_loss": 4.606278589341037,
      "tokens_seen": 129695744
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004851955867602809,
      "loss": 3.7616,
      "theoretical_loss": 4.605967791982339,
      "tokens_seen": 129761280
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048518555667001,
      "loss": 3.608,
      "theoretical_loss": 4.60565719547823,
      "tokens_seen": 129826816
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048517552657973925,
      "loss": 3.573,
      "theoretical_loss": 4.605346799597614,
      "tokens_seen": 129892352
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004851654964894684,
      "loss": 3.5297,
      "theoretical_loss": 4.605036604109778,
      "tokens_seen": 129957888
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004851554663991976,
      "loss": 3.6223,
      "theoretical_loss": 4.604726608784391,
      "tokens_seen": 130023424
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004851454363089268,
      "loss": 3.6004,
      "theoretical_loss": 4.6044168133914996,
      "tokens_seen": 130088960
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048513540621865597,
      "loss": 3.5617,
      "theoretical_loss": 4.604107217701534,
      "tokens_seen": 130154496
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048512537612838515,
      "loss": 3.4561,
      "theoretical_loss": 4.6037978214853,
      "tokens_seen": 130220032
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048511534603811433,
      "loss": 3.5599,
      "theoretical_loss": 4.603488624513987,
      "tokens_seen": 130285568
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004851053159478435,
      "loss": 3.3541,
      "theoretical_loss": 4.603179626559155,
      "tokens_seen": 130351104
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048509528585757275,
      "loss": 3.6129,
      "theoretical_loss": 4.602870827392745,
      "tokens_seen": 130416640
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004850852557673019,
      "loss": 3.5001,
      "theoretical_loss": 4.6025622267870725,
      "tokens_seen": 130482176
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004850752256770311,
      "loss": 3.623,
      "theoretical_loss": 4.602253824514828,
      "tokens_seen": 130547712
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048506519558676024,
      "loss": 3.3791,
      "theoretical_loss": 4.601945620349077,
      "tokens_seen": 130613248
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004850551654964895,
      "loss": 3.3,
      "theoretical_loss": 4.601637614063257,
      "tokens_seen": 130678784
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048504513540621866,
      "loss": 3.5917,
      "theoretical_loss": 4.601329805431178,
      "tokens_seen": 130744320
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048503510531594784,
      "loss": 3.4741,
      "theoretical_loss": 4.601022194227024,
      "tokens_seen": 130809856
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.000485025075225677,
      "loss": 3.6766,
      "theoretical_loss": 4.600714780225347,
      "tokens_seen": 130875392
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048501504513540625,
      "loss": 3.5938,
      "theoretical_loss": 4.600407563201072,
      "tokens_seen": 130940928
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004850050150451354,
      "loss": 3.5169,
      "theoretical_loss": 4.60010054292949,
      "tokens_seen": 131006464
    },
    {
      "epoch": 0.04,
      "objective/train/advantage_avg": 0.49436497688293457,
      "objective/train/docs_used": 85560,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.45306396484375,
      "objective/train/original_loss": 3.453063726425171,
      "objective/train/theoretical_loss": 4.599793719186264,
      "objective/train/tokens_used": 151532000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24444147944450378,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050679087638855,
      "objective/train/weighted_lm_loss": 3.627986431121826,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0457526445388794,
      "theoretical_loss": 4.599793719186264,
      "tokens_seen": 131072000
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004849949849548646,
      "loss": 3.4764,
      "theoretical_loss": 4.599793719186264,
      "tokens_seen": 131072000
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004849849548645938,
      "loss": 3.5235,
      "theoretical_loss": 4.599487091747424,
      "tokens_seen": 131137536
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.000484974924774323,
      "loss": 3.7058,
      "theoretical_loss": 4.599180660389363,
      "tokens_seen": 131203072
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004849648946840522,
      "loss": 3.5799,
      "theoretical_loss": 4.598874424888848,
      "tokens_seen": 131268608
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048495486459378134,
      "loss": 3.2646,
      "theoretical_loss": 4.598568385023004,
      "tokens_seen": 131334144
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004849448345035106,
      "loss": 3.517,
      "theoretical_loss": 4.598262540569325,
      "tokens_seen": 131399680
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004849348044132397,
      "loss": 3.2602,
      "theoretical_loss": 4.597956891305667,
      "tokens_seen": 131465216
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048492477432296894,
      "loss": 3.3676,
      "theoretical_loss": 4.59765143701025,
      "tokens_seen": 131530752
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004849147442326981,
      "loss": 3.7521,
      "theoretical_loss": 4.597346177461656,
      "tokens_seen": 131596288
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004849047141424273,
      "loss": 3.8368,
      "theoretical_loss": 4.597041112438829,
      "tokens_seen": 131661824
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004848946840521565,
      "loss": 3.4516,
      "theoretical_loss": 4.5967362417210715,
      "tokens_seen": 131727360
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004848846539618857,
      "loss": 3.5047,
      "theoretical_loss": 4.596431565088048,
      "tokens_seen": 131792896
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048487462387161484,
      "loss": 3.4769,
      "theoretical_loss": 4.596127082319786,
      "tokens_seen": 131858432
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004848645937813441,
      "loss": 3.2792,
      "theoretical_loss": 4.595822793196664,
      "tokens_seen": 131923968
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004848545636910732,
      "loss": 3.5882,
      "theoretical_loss": 4.595518697499422,
      "tokens_seen": 131989504
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048484453360080244,
      "loss": 3.5949,
      "theoretical_loss": 4.595214795009159,
      "tokens_seen": 132055040
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004848345035105316,
      "loss": 3.5884,
      "theoretical_loss": 4.594911085507325,
      "tokens_seen": 132120576
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004848244734202608,
      "loss": 3.5266,
      "theoretical_loss": 4.594607568775732,
      "tokens_seen": 132186112
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048481444332999,
      "loss": 3.6206,
      "theoretical_loss": 4.594304244596541,
      "tokens_seen": 132251648
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048480441323971917,
      "loss": 3.2591,
      "theoretical_loss": 4.59400111275227,
      "tokens_seen": 132317184
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048479438314944835,
      "loss": 3.402,
      "theoretical_loss": 4.593698173025789,
      "tokens_seen": 132382720
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004847843530591776,
      "loss": 3.4452,
      "theoretical_loss": 4.593395425200322,
      "tokens_seen": 132448256
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004847743229689067,
      "loss": 3.4404,
      "theoretical_loss": 4.593092869059442,
      "tokens_seen": 132513792
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048476429287863594,
      "loss": 3.2882,
      "theoretical_loss": 4.592790504387073,
      "tokens_seen": 132579328
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048475426278836507,
      "loss": 3.6444,
      "theoretical_loss": 4.592488330967495,
      "tokens_seen": 132644864
    },
    {
      "epoch": 0.04,
      "objective/train/advantage_avg": 0.4951869249343872,
      "objective/train/docs_used": 86181,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.1502928733825684,
      "objective/train/original_loss": 3.1502931118011475,
      "objective/train/theoretical_loss": 4.592186348585331,
      "objective/train/tokens_used": 153170400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2452470064163208,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507653951644897,
      "objective/train/weighted_lm_loss": 3.3101487159729004,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0442860126495361,
      "theoretical_loss": 4.592186348585331,
      "tokens_seen": 132710400
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004847442326980943,
      "loss": 3.362,
      "theoretical_loss": 4.592186348585331,
      "tokens_seen": 132710400
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004847342026078235,
      "loss": 3.6262,
      "theoretical_loss": 4.591884557025555,
      "tokens_seen": 132775936
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048472417251755267,
      "loss": 3.3834,
      "theoretical_loss": 4.591582956073488,
      "tokens_seen": 132841472
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048471414242728185,
      "loss": 3.4256,
      "theoretical_loss": 4.5912815455148,
      "tokens_seen": 132907008
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004847041123370111,
      "loss": 3.6676,
      "theoretical_loss": 4.590980325135508,
      "tokens_seen": 132972544
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004846940822467402,
      "loss": 3.4671,
      "theoretical_loss": 4.590679294721973,
      "tokens_seen": 133038080
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048468405215646945,
      "loss": 3.6992,
      "theoretical_loss": 4.5903784540609,
      "tokens_seen": 133103616
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004846740220661986,
      "loss": 3.4428,
      "theoretical_loss": 4.5900778029393425,
      "tokens_seen": 133169152
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004846639919759278,
      "loss": 3.3156,
      "theoretical_loss": 4.5897773411446945,
      "tokens_seen": 133234688
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.000484653961885657,
      "loss": 3.5959,
      "theoretical_loss": 4.589477068464692,
      "tokens_seen": 133300224
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048464393179538617,
      "loss": 3.4034,
      "theoretical_loss": 4.589176984687415,
      "tokens_seen": 133365760
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048463390170511535,
      "loss": 3.4187,
      "theoretical_loss": 4.588877089601287,
      "tokens_seen": 133431296
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048462387161484453,
      "loss": 3.5189,
      "theoretical_loss": 4.5885773829950685,
      "tokens_seen": 133496832
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004846138415245737,
      "loss": 3.5248,
      "theoretical_loss": 4.5882778646578615,
      "tokens_seen": 133562368
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048460381143430295,
      "loss": 3.5755,
      "theoretical_loss": 4.587978534379106,
      "tokens_seen": 133627904
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004845937813440321,
      "loss": 3.354,
      "theoretical_loss": 4.587679391948584,
      "tokens_seen": 133693440
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004845837512537613,
      "loss": 3.8906,
      "theoretical_loss": 4.587380437156413,
      "tokens_seen": 133758976
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048457372116349044,
      "loss": 3.5652,
      "theoretical_loss": 4.587081669793047,
      "tokens_seen": 133824512
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004845636910732197,
      "loss": 3.5054,
      "theoretical_loss": 4.586783089649279,
      "tokens_seen": 133890048
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048455366098294886,
      "loss": 3.2784,
      "theoretical_loss": 4.586484696516234,
      "tokens_seen": 133955584
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048454363089267804,
      "loss": 3.4291,
      "theoretical_loss": 4.586186490185376,
      "tokens_seen": 134021120
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004845336008024072,
      "loss": 3.4347,
      "theoretical_loss": 4.585888470448501,
      "tokens_seen": 134086656
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048452357071213645,
      "loss": 3.3934,
      "theoretical_loss": 4.585590637097741,
      "tokens_seen": 134152192
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004845135406218656,
      "loss": 3.566,
      "theoretical_loss": 4.585292989925557,
      "tokens_seen": 134217728
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004845035105315948,
      "loss": 3.4804,
      "theoretical_loss": 4.584995528724745,
      "tokens_seen": 134283264
    },
    {
      "epoch": 0.04,
      "objective/train/advantage_avg": 0.4944709241390228,
      "objective/train/docs_used": 87535,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.1124672889709473,
      "objective/train/original_loss": 3.112466812133789,
      "objective/train/theoretical_loss": 4.584698253288433,
      "objective/train/tokens_used": 154808800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2445756047964096,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506902933120728,
      "objective/train/weighted_lm_loss": 3.2699925899505615,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0338950157165527,
      "theoretical_loss": 4.584698253288433,
      "tokens_seen": 134348800
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048449348044132394,
      "loss": 3.3234,
      "theoretical_loss": 4.584698253288433,
      "tokens_seen": 134348800
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004844834503510532,
      "loss": 3.5686,
      "theoretical_loss": 4.58440116341008,
      "tokens_seen": 134414336
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048447342026078236,
      "loss": 3.831,
      "theoretical_loss": 4.584104258883473,
      "tokens_seen": 134479872
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048446339017051154,
      "loss": 3.2735,
      "theoretical_loss": 4.58380753950273,
      "tokens_seen": 134545408
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004844533600802407,
      "loss": 3.3295,
      "theoretical_loss": 4.583511005062297,
      "tokens_seen": 134610944
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004844433299899699,
      "loss": 3.8233,
      "theoretical_loss": 4.583214655356951,
      "tokens_seen": 134676480
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004844332998996991,
      "loss": 3.7019,
      "theoretical_loss": 4.5829184901817905,
      "tokens_seen": 134742016
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004844232698094283,
      "loss": 3.6857,
      "theoretical_loss": 4.582622509332248,
      "tokens_seen": 134807552
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048441323971915745,
      "loss": 3.5476,
      "theoretical_loss": 4.582326712604077,
      "tokens_seen": 134873088
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004844032096288867,
      "loss": 3.3993,
      "theoretical_loss": 4.58203109979336,
      "tokens_seen": 134938624
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048439317953861586,
      "loss": 3.5692,
      "theoretical_loss": 4.581735670696499,
      "tokens_seen": 135004160
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048438314944834504,
      "loss": 3.3333,
      "theoretical_loss": 4.581440425110225,
      "tokens_seen": 135069696
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004843731193580742,
      "loss": 3.6261,
      "theoretical_loss": 4.581145362831592,
      "tokens_seen": 135135232
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004843630892678034,
      "loss": 3.56,
      "theoretical_loss": 4.580850483657974,
      "tokens_seen": 135200768
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004843530591775326,
      "loss": 3.4489,
      "theoretical_loss": 4.580555787387068,
      "tokens_seen": 135266304
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004843430290872618,
      "loss": 3.5932,
      "theoretical_loss": 4.580261273816895,
      "tokens_seen": 135331840
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048433299899699095,
      "loss": 3.688,
      "theoretical_loss": 4.579966942745792,
      "tokens_seen": 135397376
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004843229689067202,
      "loss": 3.5481,
      "theoretical_loss": 4.5796727939724216,
      "tokens_seen": 135462912
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004843129388164493,
      "loss": 3.5165,
      "theoretical_loss": 4.579378827295761,
      "tokens_seen": 135528448
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048430290872617855,
      "loss": 3.455,
      "theoretical_loss": 4.579085042515108,
      "tokens_seen": 135593984
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048429287863590773,
      "loss": 3.3142,
      "theoretical_loss": 4.578791439430079,
      "tokens_seen": 135659520
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004842828485456369,
      "loss": 3.257,
      "theoretical_loss": 4.578498017840608,
      "tokens_seen": 135725056
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004842728184553661,
      "loss": 3.5279,
      "theoretical_loss": 4.578204777546944,
      "tokens_seen": 135790592
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048426278836509527,
      "loss": 3.3612,
      "theoretical_loss": 4.577911718349654,
      "tokens_seen": 135856128
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048425275827482445,
      "loss": 3.614,
      "theoretical_loss": 4.5776188400496185,
      "tokens_seen": 135921664
    },
    {
      "epoch": 0.04,
      "objective/train/advantage_avg": 0.494077205657959,
      "objective/train/docs_used": 88189,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.0483860969543457,
      "objective/train/original_loss": 3.048386573791504,
      "objective/train/theoretical_loss": 4.577326142448035,
      "objective/train/tokens_used": 156447200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24418115615844727,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506490468978882,
      "objective/train/weighted_lm_loss": 3.202303409576416,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0408108234405518,
      "theoretical_loss": 4.577326142448035,
      "tokens_seen": 135987200
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004842427281845537,
      "loss": 3.3598,
      "theoretical_loss": 4.577326142448035,
      "tokens_seen": 135987200
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048423269809428287,
      "loss": 3.2862,
      "theoretical_loss": 4.577033625346415,
      "tokens_seen": 136052736
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048422266800401205,
      "loss": 3.1486,
      "theoretical_loss": 4.576741288546581,
      "tokens_seen": 136118272
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004842126379137413,
      "loss": 3.8291,
      "theoretical_loss": 4.576449131850669,
      "tokens_seen": 136183808
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004842026078234704,
      "loss": 3.4291,
      "theoretical_loss": 4.576157155061131,
      "tokens_seen": 136249344
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048419257773319965,
      "loss": 3.38,
      "theoretical_loss": 4.575865357980726,
      "tokens_seen": 136314880
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004841825476429288,
      "loss": 3.4041,
      "theoretical_loss": 4.575573740412527,
      "tokens_seen": 136380416
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.000484172517552658,
      "loss": 3.1819,
      "theoretical_loss": 4.575282302159915,
      "tokens_seen": 136445952
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004841624874623872,
      "loss": 3.4411,
      "theoretical_loss": 4.574991043026582,
      "tokens_seen": 136511488
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048415245737211637,
      "loss": 3.347,
      "theoretical_loss": 4.574699962816529,
      "tokens_seen": 136577024
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048414242728184555,
      "loss": 3.4874,
      "theoretical_loss": 4.574409061334064,
      "tokens_seen": 136642560
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048413239719157473,
      "loss": 3.5618,
      "theoretical_loss": 4.574118338383805,
      "tokens_seen": 136708096
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004841223671013039,
      "loss": 3.4416,
      "theoretical_loss": 4.5738277937706755,
      "tokens_seen": 136773632
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048411233701103315,
      "loss": 3.3989,
      "theoretical_loss": 4.573537427299906,
      "tokens_seen": 136839168
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004841023069207623,
      "loss": 3.8135,
      "theoretical_loss": 4.573247238777032,
      "tokens_seen": 136904704
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004840922768304915,
      "loss": 3.2826,
      "theoretical_loss": 4.572957228007897,
      "tokens_seen": 136970240
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048408224674022064,
      "loss": 3.3487,
      "theoretical_loss": 4.572667394798646,
      "tokens_seen": 137035776
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004840722166499499,
      "loss": 3.4457,
      "theoretical_loss": 4.572377738955729,
      "tokens_seen": 137101312
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048406218655967906,
      "loss": 3.6661,
      "theoretical_loss": 4.5720882602859,
      "tokens_seen": 137166848
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048405215646940824,
      "loss": 3.6471,
      "theoretical_loss": 4.571798958596215,
      "tokens_seen": 137232384
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004840421263791374,
      "loss": 3.3136,
      "theoretical_loss": 4.571509833694032,
      "tokens_seen": 137297920
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048403209628886665,
      "loss": 3.4755,
      "theoretical_loss": 4.571220885387013,
      "tokens_seen": 137363456
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004840220661985958,
      "loss": 3.7897,
      "theoretical_loss": 4.570932113483116,
      "tokens_seen": 137428992
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.000484012036108325,
      "loss": 3.7707,
      "theoretical_loss": 4.570643517790604,
      "tokens_seen": 137494528
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048400200601805414,
      "loss": 3.3767,
      "theoretical_loss": 4.570355098118037,
      "tokens_seen": 137560064
    },
    {
      "epoch": 0.04,
      "objective/train/advantage_avg": 0.495524138212204,
      "objective/train/docs_used": 88907,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.1107735633850098,
      "objective/train/original_loss": 3.1107733249664307,
      "objective/train/theoretical_loss": 4.570066854274275,
      "objective/train/tokens_used": 158085600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24560120701789856,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0508008003234863,
      "objective/train/weighted_lm_loss": 3.2685976028442383,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 4.570066854274275,
      "tokens_seen": 137625600
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004839919759277834,
      "loss": 3.4914,
      "theoretical_loss": 4.570066854274275,
      "tokens_seen": 137625600
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048398194583751256,
      "loss": 3.706,
      "theoretical_loss": 4.569778786068477,
      "tokens_seen": 137691136
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048397191574724174,
      "loss": 3.5393,
      "theoretical_loss": 4.569490893310098,
      "tokens_seen": 137756672
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004839618856569709,
      "loss": 3.3548,
      "theoretical_loss": 4.569203175808893,
      "tokens_seen": 137822208
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004839518555667001,
      "loss": 3.4905,
      "theoretical_loss": 4.5689156333749095,
      "tokens_seen": 137887744
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004839418254764293,
      "loss": 3.1102,
      "theoretical_loss": 4.5686282658184965,
      "tokens_seen": 137953280
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004839317953861585,
      "loss": 3.349,
      "theoretical_loss": 4.568341072950292,
      "tokens_seen": 138018816
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048392176529588765,
      "loss": 3.3688,
      "theoretical_loss": 4.568054054581236,
      "tokens_seen": 138084352
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004839117352056169,
      "loss": 3.5403,
      "theoretical_loss": 4.5677672105225575,
      "tokens_seen": 138149888
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048390170511534606,
      "loss": 3.6276,
      "theoretical_loss": 4.5674805405857795,
      "tokens_seen": 138215424
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048389167502507524,
      "loss": 3.5424,
      "theoretical_loss": 4.56719404458272,
      "tokens_seen": 138280960
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004838816449348044,
      "loss": 3.3991,
      "theoretical_loss": 4.56690772232549,
      "tokens_seen": 138346496
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004838716148445336,
      "loss": 3.4439,
      "theoretical_loss": 4.566621573626489,
      "tokens_seen": 138412032
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004838615847542628,
      "loss": 3.6006,
      "theoretical_loss": 4.56633559829841,
      "tokens_seen": 138477568
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.000483851554663992,
      "loss": 3.308,
      "theoretical_loss": 4.566049796154239,
      "tokens_seen": 138543104
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048384152457372115,
      "loss": 3.6462,
      "theoretical_loss": 4.565764167007247,
      "tokens_seen": 138608640
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004838314944834504,
      "loss": 3.754,
      "theoretical_loss": 4.565478710670996,
      "tokens_seen": 138674176
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004838214643931795,
      "loss": 3.563,
      "theoretical_loss": 4.565193426959341,
      "tokens_seen": 138739712
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048381143430290875,
      "loss": 3.6812,
      "theoretical_loss": 4.56490831568642,
      "tokens_seen": 138805248
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048380140421263793,
      "loss": 3.6877,
      "theoretical_loss": 4.5646233766666615,
      "tokens_seen": 138870784
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004837913741223671,
      "loss": 3.4363,
      "theoretical_loss": 4.564338609714781,
      "tokens_seen": 138936320
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004837813440320963,
      "loss": 3.592,
      "theoretical_loss": 4.56405401464578,
      "tokens_seen": 139001856
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048377131394182547,
      "loss": 3.5593,
      "theoretical_loss": 4.563769591274946,
      "tokens_seen": 139067392
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048376128385155465,
      "loss": 3.7644,
      "theoretical_loss": 4.563485339417852,
      "tokens_seen": 139132928
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004837512537612839,
      "loss": 3.5976,
      "theoretical_loss": 4.563201258890357,
      "tokens_seen": 139198464
    },
    {
      "epoch": 0.04,
      "objective/train/advantage_avg": 0.4939173460006714,
      "objective/train/docs_used": 90119,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.963639974594116,
      "objective/train/original_loss": 3.963639974594116,
      "objective/train/theoretical_loss": 4.562917349508602,
      "objective/train/tokens_used": 159724000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24404951930046082,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506322383880615,
      "objective/train/weighted_lm_loss": 4.164361953735352,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.042169213294983,
      "theoretical_loss": 4.562917349508602,
      "tokens_seen": 139264000
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.000483741223671013,
      "loss": 3.816,
      "theoretical_loss": 4.562917349508602,
      "tokens_seen": 139264000
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048373119358074225,
      "loss": 3.3637,
      "theoretical_loss": 4.562633611089014,
      "tokens_seen": 139329536
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048372116349047143,
      "loss": 3.4552,
      "theoretical_loss": 4.562350043448301,
      "tokens_seen": 139395072
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004837111334002006,
      "loss": 3.682,
      "theoretical_loss": 4.562066646403457,
      "tokens_seen": 139460608
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004837011033099298,
      "loss": 3.646,
      "theoretical_loss": 4.561783419771752,
      "tokens_seen": 139526144
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.000483691073219659,
      "loss": 3.3804,
      "theoretical_loss": 4.561500363370744,
      "tokens_seen": 139591680
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048368104312938816,
      "loss": 3.385,
      "theoretical_loss": 4.561217477018268,
      "tokens_seen": 139657216
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004836710130391174,
      "loss": 3.813,
      "theoretical_loss": 4.560934760532439,
      "tokens_seen": 139722752
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004836609829488465,
      "loss": 3.4686,
      "theoretical_loss": 4.560652213731656,
      "tokens_seen": 139788288
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048365095285857575,
      "loss": 3.6368,
      "theoretical_loss": 4.560369836434591,
      "tokens_seen": 139853824
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004836409227683049,
      "loss": 3.4482,
      "theoretical_loss": 4.5600876284602,
      "tokens_seen": 139919360
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004836308926780341,
      "loss": 3.2682,
      "theoretical_loss": 4.559805589627714,
      "tokens_seen": 139984896
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004836208625877633,
      "loss": 3.2677,
      "theoretical_loss": 4.559523719756642,
      "tokens_seen": 140050432
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004836108324974925,
      "loss": 3.4707,
      "theoretical_loss": 4.559242018666771,
      "tokens_seen": 140115968
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048360080240722166,
      "loss": 3.3458,
      "theoretical_loss": 4.558960486178163,
      "tokens_seen": 140181504
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048359077231695084,
      "loss": 3.4258,
      "theoretical_loss": 4.5586791221111564,
      "tokens_seen": 140247040
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048358074222668,
      "loss": 3.5789,
      "theoretical_loss": 4.558397926286366,
      "tokens_seen": 140312576
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048357071213640926,
      "loss": 3.444,
      "theoretical_loss": 4.5581168985246805,
      "tokens_seen": 140378112
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004835606820461384,
      "loss": 3.7812,
      "theoretical_loss": 4.557836038647261,
      "tokens_seen": 140443648
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004835506519558676,
      "loss": 3.4198,
      "theoretical_loss": 4.557555346475546,
      "tokens_seen": 140509184
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004835406218655968,
      "loss": 3.5029,
      "theoretical_loss": 4.557274821831242,
      "tokens_seen": 140574720
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.000483530591775326,
      "loss": 3.4955,
      "theoretical_loss": 4.556994464536334,
      "tokens_seen": 140640256
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048352056168505516,
      "loss": 3.4252,
      "theoretical_loss": 4.556714274413074,
      "tokens_seen": 140705792
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048351053159478434,
      "loss": 3.5268,
      "theoretical_loss": 4.556434251283989,
      "tokens_seen": 140771328
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004835005015045135,
      "loss": 3.4717,
      "theoretical_loss": 4.556154394971874,
      "tokens_seen": 140836864
    },
    {
      "epoch": 0.04,
      "objective/train/advantage_avg": 0.49193450808525085,
      "objective/train/docs_used": 90792,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.836174726486206,
      "objective/train/original_loss": 3.836174726486206,
      "objective/train/theoretical_loss": 4.5558747052997965,
      "objective/train/tokens_used": 161362400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24210110306739807,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0504240989685059,
      "objective/train/weighted_lm_loss": 4.0294084548950195,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0437887907028198,
      "theoretical_loss": 4.5558747052997965,
      "tokens_seen": 140902400
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048349047141424276,
      "loss": 3.5462,
      "theoretical_loss": 4.5558747052997965,
      "tokens_seen": 140902400
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048348044132397194,
      "loss": 3.7106,
      "theoretical_loss": 4.555595182091094,
      "tokens_seen": 140967936
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004834704112337011,
      "loss": 3.3865,
      "theoretical_loss": 4.555315825169371,
      "tokens_seen": 141033472
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004834603811434303,
      "loss": 3.3779,
      "theoretical_loss": 4.555036634358502,
      "tokens_seen": 141099008
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004834503510531595,
      "loss": 3.4289,
      "theoretical_loss": 4.554757609482632,
      "tokens_seen": 141164544
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004834403209628887,
      "loss": 3.3927,
      "theoretical_loss": 4.554478750366169,
      "tokens_seen": 141230080
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048343029087261785,
      "loss": 3.6633,
      "theoretical_loss": 4.5542000568337935,
      "tokens_seen": 141295616
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004834202607823471,
      "loss": 3.5973,
      "theoretical_loss": 4.553921528710448,
      "tokens_seen": 141361152
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048341023069207626,
      "loss": 3.4706,
      "theoretical_loss": 4.553643165821343,
      "tokens_seen": 141426688
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048340020060180544,
      "loss": 3.6421,
      "theoretical_loss": 4.553364967991957,
      "tokens_seen": 141492224
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004833901705115346,
      "loss": 3.6193,
      "theoretical_loss": 4.553086935048029,
      "tokens_seen": 141557760
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004833801404212638,
      "loss": 3.3472,
      "theoretical_loss": 4.552809066815566,
      "tokens_seen": 141623296
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.000483370110330993,
      "loss": 3.4077,
      "theoretical_loss": 4.552531363120837,
      "tokens_seen": 141688832
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004833600802407222,
      "loss": 3.4088,
      "theoretical_loss": 4.552253823790375,
      "tokens_seen": 141754368
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048335005015045135,
      "loss": 3.6293,
      "theoretical_loss": 4.5519764486509775,
      "tokens_seen": 141819904
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004833400200601806,
      "loss": 3.5724,
      "theoretical_loss": 4.551699237529702,
      "tokens_seen": 141885440
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004833299899699097,
      "loss": 3.5199,
      "theoretical_loss": 4.55142219025387,
      "tokens_seen": 141950976
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048331995987963895,
      "loss": 3.3614,
      "theoretical_loss": 4.551145306651064,
      "tokens_seen": 142016512
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048330992978936813,
      "loss": 3.2388,
      "theoretical_loss": 4.550868586549125,
      "tokens_seen": 142082048
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004832998996990973,
      "loss": 3.8,
      "theoretical_loss": 4.550592029776158,
      "tokens_seen": 142147584
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004832898696088265,
      "loss": 3.2818,
      "theoretical_loss": 4.550315636160526,
      "tokens_seen": 142213120
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048327983951855567,
      "loss": 3.5412,
      "theoretical_loss": 4.550039405530851,
      "tokens_seen": 142278656
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048326980942828485,
      "loss": 3.5927,
      "theoretical_loss": 4.549763337716014,
      "tokens_seen": 142344192
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004832597793380141,
      "loss": 3.3529,
      "theoretical_loss": 4.549487432545156,
      "tokens_seen": 142409728
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004832497492477432,
      "loss": 3.711,
      "theoretical_loss": 4.549211689847674,
      "tokens_seen": 142475264
    },
    {
      "epoch": 0.04,
      "objective/train/advantage_avg": 0.49578189849853516,
      "objective/train/docs_used": 92175,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.860226631164551,
      "objective/train/original_loss": 2.86022686958313,
      "objective/train/theoretical_loss": 4.548936109453223,
      "objective/train/tokens_used": 163000800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24584046006202698,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050827980041504,
      "objective/train/weighted_lm_loss": 3.0055363178253174,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0304546356201172,
      "theoretical_loss": 4.548936109453223,
      "tokens_seen": 142540800
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048323971915747245,
      "loss": 3.413,
      "theoretical_loss": 4.548936109453223,
      "tokens_seen": 142540800
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048322968906720163,
      "loss": 3.5168,
      "theoretical_loss": 4.5486606911917145,
      "tokens_seen": 142606336
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004832196589769308,
      "loss": 3.5327,
      "theoretical_loss": 4.548385434893316,
      "tokens_seen": 142671872
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048320962888666,
      "loss": 3.6024,
      "theoretical_loss": 4.548110340388453,
      "tokens_seen": 142737408
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004831995987963892,
      "loss": 3.4211,
      "theoretical_loss": 4.547835407507803,
      "tokens_seen": 142802944
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048318956870611836,
      "loss": 3.3099,
      "theoretical_loss": 4.5475606360822995,
      "tokens_seen": 142868480
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004831795386158476,
      "loss": 3.8352,
      "theoretical_loss": 4.547286025943132,
      "tokens_seen": 142934016
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004831695085255767,
      "loss": 3.4629,
      "theoretical_loss": 4.547011576921742,
      "tokens_seen": 142999552
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048315947843530595,
      "loss": 3.5423,
      "theoretical_loss": 4.546737288849825,
      "tokens_seen": 143065088
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004831494483450351,
      "loss": 3.6474,
      "theoretical_loss": 4.546463161559329,
      "tokens_seen": 143130624
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004831394182547643,
      "loss": 3.4136,
      "theoretical_loss": 4.546189194882455,
      "tokens_seen": 143196160
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004831293881644935,
      "loss": 3.5102,
      "theoretical_loss": 4.545915388651654,
      "tokens_seen": 143261696
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004831193580742227,
      "loss": 3.4249,
      "theoretical_loss": 4.54564174269963,
      "tokens_seen": 143327232
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048310932798395186,
      "loss": 3.7411,
      "theoretical_loss": 4.545368256859338,
      "tokens_seen": 143392768
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048309929789368104,
      "loss": 3.5779,
      "theoretical_loss": 4.545094930963982,
      "tokens_seen": 143458304
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004830892678034102,
      "loss": 3.5933,
      "theoretical_loss": 4.5448217648470175,
      "tokens_seen": 143523840
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048307923771313946,
      "loss": 3.3503,
      "theoretical_loss": 4.544548758342148,
      "tokens_seen": 143589376
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004830692076228686,
      "loss": 3.7908,
      "theoretical_loss": 4.544275911283326,
      "tokens_seen": 143654912
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004830591775325978,
      "loss": 3.5777,
      "theoretical_loss": 4.5440032235047525,
      "tokens_seen": 143720448
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.000483049147442327,
      "loss": 3.4707,
      "theoretical_loss": 4.543730694840878,
      "tokens_seen": 143785984
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004830391173520562,
      "loss": 3.5128,
      "theoretical_loss": 4.543458325126398,
      "tokens_seen": 143851520
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048302908726178536,
      "loss": 3.4662,
      "theoretical_loss": 4.543186114196256,
      "tokens_seen": 143917056
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048301905717151454,
      "loss": 3.3641,
      "theoretical_loss": 4.542914061885643,
      "tokens_seen": 143982592
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004830090270812437,
      "loss": 3.6053,
      "theoretical_loss": 4.542642168029994,
      "tokens_seen": 144048128
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048299899699097296,
      "loss": 3.5902,
      "theoretical_loss": 4.54237043246499,
      "tokens_seen": 144113664
    },
    {
      "epoch": 0.04,
      "objective/train/advantage_avg": 0.4952732026576996,
      "objective/train/docs_used": 92798,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.8385159969329834,
      "objective/train/original_loss": 2.838515520095825,
      "objective/train/theoretical_loss": 4.542098855026559,
      "objective/train/tokens_used": 164639200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24533838033676147,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507744550704956,
      "objective/train/weighted_lm_loss": 2.9823014736175537,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0363596677780151,
      "theoretical_loss": 4.542098855026559,
      "tokens_seen": 144179200
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004829889669007021,
      "loss": 3.4161,
      "theoretical_loss": 4.542098855026559,
      "tokens_seen": 144179200
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004829789368104313,
      "loss": 3.6375,
      "theoretical_loss": 4.54182743555087,
      "tokens_seen": 144244736
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048296890672016045,
      "loss": 3.5452,
      "theoretical_loss": 4.54155617387434,
      "tokens_seen": 144310272
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004829588766298897,
      "loss": 3.3869,
      "theoretical_loss": 4.541285069833626,
      "tokens_seen": 144375808
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048294884653961887,
      "loss": 3.376,
      "theoretical_loss": 4.541014123265632,
      "tokens_seen": 144441344
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048293881644934805,
      "loss": 3.3103,
      "theoretical_loss": 4.5407433340075,
      "tokens_seen": 144506880
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048292878635907723,
      "loss": 3.4463,
      "theoretical_loss": 4.5404727018966184,
      "tokens_seen": 144572416
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048291875626880646,
      "loss": 3.6557,
      "theoretical_loss": 4.5402022267706155,
      "tokens_seen": 144637952
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004829087261785356,
      "loss": 3.524,
      "theoretical_loss": 4.539931908467359,
      "tokens_seen": 144703488
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004828986960882648,
      "loss": 3.412,
      "theoretical_loss": 4.539661746824962,
      "tokens_seen": 144769024
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048288866599799395,
      "loss": 3.4516,
      "theoretical_loss": 4.539391741681774,
      "tokens_seen": 144834560
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004828786359077232,
      "loss": 3.3718,
      "theoretical_loss": 4.539121892876385,
      "tokens_seen": 144900096
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048286860581745237,
      "loss": 3.5573,
      "theoretical_loss": 4.538852200247625,
      "tokens_seen": 144965632
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048285857572718155,
      "loss": 3.7104,
      "theoretical_loss": 4.538582663634564,
      "tokens_seen": 145031168
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048284854563691073,
      "loss": 3.7741,
      "theoretical_loss": 4.5383132828765085,
      "tokens_seen": 145096704
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004828385155466399,
      "loss": 3.7424,
      "theoretical_loss": 4.538044057813003,
      "tokens_seen": 145162240
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004828284854563691,
      "loss": 3.502,
      "theoretical_loss": 4.5377749882838305,
      "tokens_seen": 145227776
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048281845536609833,
      "loss": 3.7115,
      "theoretical_loss": 4.537506074129013,
      "tokens_seen": 145293312
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048280842527582746,
      "loss": 3.4845,
      "theoretical_loss": 4.537237315188804,
      "tokens_seen": 145358848
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004827983951855567,
      "loss": 3.3949,
      "theoretical_loss": 4.536968711303699,
      "tokens_seen": 145424384
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004827883650952858,
      "loss": 3.5439,
      "theoretical_loss": 4.536700262314425,
      "tokens_seen": 145489920
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048277833500501505,
      "loss": 3.5539,
      "theoretical_loss": 4.536431968061946,
      "tokens_seen": 145555456
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048276830491474423,
      "loss": 3.3937,
      "theoretical_loss": 4.536163828387459,
      "tokens_seen": 145620992
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004827582748244734,
      "loss": 3.6023,
      "theoretical_loss": 4.535895843132399,
      "tokens_seen": 145686528
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004827482447342026,
      "loss": 3.7428,
      "theoretical_loss": 4.53562801213843,
      "tokens_seen": 145752064
    },
    {
      "epoch": 0.04,
      "objective/train/advantage_avg": 0.49418872594833374,
      "objective/train/docs_used": 94036,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.0065338611602783,
      "objective/train/original_loss": 3.006533622741699,
      "objective/train/theoretical_loss": 4.535360335247455,
      "objective/train/tokens_used": 166277600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24429532885551453,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506607294082642,
      "objective/train/weighted_lm_loss": 3.1586456298828125,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0338950157165527,
      "theoretical_loss": 4.535360335247455,
      "tokens_seen": 145817600
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048273821464393183,
      "loss": 3.2932,
      "theoretical_loss": 4.535360335247455,
      "tokens_seen": 145817600
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.000482728184553661,
      "loss": 3.5763,
      "theoretical_loss": 4.535092812301606,
      "tokens_seen": 145883136
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004827181544633902,
      "loss": 3.4034,
      "theoretical_loss": 4.534825443143248,
      "tokens_seen": 145948672
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004827081243731194,
      "loss": 3.5839,
      "theoretical_loss": 4.53455822761498,
      "tokens_seen": 146014208
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048269809428284856,
      "loss": 3.4611,
      "theoretical_loss": 4.53429116555963,
      "tokens_seen": 146079744
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004826880641925778,
      "loss": 3.3895,
      "theoretical_loss": 4.53402425682026,
      "tokens_seen": 146145280
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004826780341023069,
      "loss": 3.611,
      "theoretical_loss": 4.53375750124016,
      "tokens_seen": 146210816
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048266800401203615,
      "loss": 3.4491,
      "theoretical_loss": 4.533490898662853,
      "tokens_seen": 146276352
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004826579739217653,
      "loss": 3.4751,
      "theoretical_loss": 4.53322444893209,
      "tokens_seen": 146341888
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004826479438314945,
      "loss": 3.2838,
      "theoretical_loss": 4.532958151891852,
      "tokens_seen": 146407424
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004826379137412237,
      "loss": 3.4448,
      "theoretical_loss": 4.53269200738635,
      "tokens_seen": 146472960
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004826278836509529,
      "loss": 3.4537,
      "theoretical_loss": 4.532426015260021,
      "tokens_seen": 146538496
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048261785356068206,
      "loss": 3.5616,
      "theoretical_loss": 4.532160175357533,
      "tokens_seen": 146604032
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048260782347041124,
      "loss": 3.3224,
      "theoretical_loss": 4.531894487523781,
      "tokens_seen": 146669568
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004825977933801404,
      "loss": 3.4069,
      "theoretical_loss": 4.531628951603885,
      "tokens_seen": 146735104
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048258776328986966,
      "loss": 3.4698,
      "theoretical_loss": 4.531363567443194,
      "tokens_seen": 146800640
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004825777331995988,
      "loss": 3.6557,
      "theoretical_loss": 4.531098334887284,
      "tokens_seen": 146866176
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.000482567703109328,
      "loss": 3.4692,
      "theoretical_loss": 4.530833253781955,
      "tokens_seen": 146931712
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004825576730190572,
      "loss": 3.4661,
      "theoretical_loss": 4.530568323973234,
      "tokens_seen": 146997248
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004825476429287864,
      "loss": 3.8025,
      "theoretical_loss": 4.530303545307371,
      "tokens_seen": 147062784
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048253761283851556,
      "loss": 3.5818,
      "theoretical_loss": 4.530038917630844,
      "tokens_seen": 147128320
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048252758274824474,
      "loss": 3.6769,
      "theoretical_loss": 4.529774440790353,
      "tokens_seen": 147193856
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004825175526579739,
      "loss": 3.7287,
      "theoretical_loss": 4.529510114632821,
      "tokens_seen": 147259392
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048250752256770316,
      "loss": 3.4588,
      "theoretical_loss": 4.529245939005397,
      "tokens_seen": 147324928
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004824974924774323,
      "loss": 3.5035,
      "theoretical_loss": 4.528981913755452,
      "tokens_seen": 147390464
    },
    {
      "epoch": 0.04,
      "objective/train/advantage_avg": 0.49612343311309814,
      "objective/train/docs_used": 94726,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.955519676208496,
      "objective/train/original_loss": 2.955519437789917,
      "objective/train/theoretical_loss": 4.5287180387305765,
      "objective/train/tokens_used": 167916000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24617838859558105,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0508638620376587,
      "objective/train/weighted_lm_loss": 3.10575008392334,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.042169213294983,
      "theoretical_loss": 4.5287180387305765,
      "tokens_seen": 147456000
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004824874623871615,
      "loss": 3.6152,
      "theoretical_loss": 4.5287180387305765,
      "tokens_seen": 147456000
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048247743229689065,
      "loss": 3.8394,
      "theoretical_loss": 4.528454313778588,
      "tokens_seen": 147521536
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004824674022066199,
      "loss": 3.4768,
      "theoretical_loss": 4.528190738747523,
      "tokens_seen": 147587072
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048245737211634907,
      "loss": 3.457,
      "theoretical_loss": 4.527927313485639,
      "tokens_seen": 147652608
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048244734202607825,
      "loss": 3.4467,
      "theoretical_loss": 4.527664037841416,
      "tokens_seen": 147718144
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048243731193580743,
      "loss": 3.0931,
      "theoretical_loss": 4.527400911663553,
      "tokens_seen": 147783680
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048242728184553666,
      "loss": 3.4782,
      "theoretical_loss": 4.527137934800969,
      "tokens_seen": 147849216
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004824172517552658,
      "loss": 3.6918,
      "theoretical_loss": 4.526875107102803,
      "tokens_seen": 147914752
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.000482407221664995,
      "loss": 3.5004,
      "theoretical_loss": 4.526612428418414,
      "tokens_seen": 147980288
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048239719157472415,
      "loss": 3.4875,
      "theoretical_loss": 4.5263498985973785,
      "tokens_seen": 148045824
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004823871614844534,
      "loss": 3.6611,
      "theoretical_loss": 4.526087517489492,
      "tokens_seen": 148111360
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048237713139418257,
      "loss": 3.2393,
      "theoretical_loss": 4.5258252849447675,
      "tokens_seen": 148176896
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048236710130391175,
      "loss": 3.4027,
      "theoretical_loss": 4.525563200813437,
      "tokens_seen": 148242432
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.00048235707121364093,
      "loss": 3.6409,
      "theoretical_loss": 4.525301264945947,
      "tokens_seen": 148307968
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004823470411233701,
      "loss": 3.6154,
      "theoretical_loss": 4.525039477192962,
      "tokens_seen": 148373504
    },
    {
      "epoch": 0.04,
      "learning_rate": 0.0004823370110330993,
      "loss": 3.5781,
      "theoretical_loss": 4.524777837405364,
      "tokens_seen": 148439040
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00048232698094282853,
      "loss": 3.1691,
      "theoretical_loss": 4.5245163454342485,
      "tokens_seen": 148504576
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00048231695085255766,
      "loss": 3.4685,
      "theoretical_loss": 4.524255001130929,
      "tokens_seen": 148570112
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004823069207622869,
      "loss": 3.6188,
      "theoretical_loss": 4.523993804346933,
      "tokens_seen": 148635648
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.000482296890672016,
      "loss": 3.3929,
      "theoretical_loss": 4.5237327549340005,
      "tokens_seen": 148701184
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00048228686058174525,
      "loss": 3.5603,
      "theoretical_loss": 4.523471852744089,
      "tokens_seen": 148766720
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00048227683049147443,
      "loss": 3.5671,
      "theoretical_loss": 4.52321109762937,
      "tokens_seen": 148832256
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004822668004012036,
      "loss": 3.3246,
      "theoretical_loss": 4.522950489442225,
      "tokens_seen": 148897792
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004822567703109328,
      "loss": 3.3642,
      "theoretical_loss": 4.522690028035251,
      "tokens_seen": 148963328
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00048224674022066203,
      "loss": 3.6808,
      "theoretical_loss": 4.522429713261258,
      "tokens_seen": 149028864
    },
    {
      "epoch": 0.05,
      "objective/train/advantage_avg": 0.49372151494026184,
      "objective/train/docs_used": 95869,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.4808573722839355,
      "objective/train/original_loss": 3.4808578491210938,
      "objective/train/theoretical_loss": 4.522169544973266,
      "objective/train/tokens_used": 169554400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24387165904045105,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506118535995483,
      "objective/train/weighted_lm_loss": 3.6567330360412598,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0338950157165527,
      "theoretical_loss": 4.522169544973266,
      "tokens_seen": 149094400
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00048223671013039116,
      "loss": 3.482,
      "theoretical_loss": 4.522169544973266,
      "tokens_seen": 149094400
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004822266800401204,
      "loss": 3.5388,
      "theoretical_loss": 4.52190952302451,
      "tokens_seen": 149159936
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004822166499498495,
      "loss": 3.528,
      "theoretical_loss": 4.521649647268434,
      "tokens_seen": 149225472
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00048220661985957876,
      "loss": 3.5071,
      "theoretical_loss": 4.521389917558694,
      "tokens_seen": 149291008
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00048219658976930794,
      "loss": 3.5666,
      "theoretical_loss": 4.521130333749157,
      "tokens_seen": 149356544
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004821865596790371,
      "loss": 3.4972,
      "theoretical_loss": 4.520870895693898,
      "tokens_seen": 149422080
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004821765295887663,
      "loss": 3.6578,
      "theoretical_loss": 4.520611603247206,
      "tokens_seen": 149487616
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004821664994984955,
      "loss": 3.3646,
      "theoretical_loss": 4.520352456263574,
      "tokens_seen": 149553152
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00048215646940822466,
      "loss": 3.3945,
      "theoretical_loss": 4.520093454597711,
      "tokens_seen": 149618688
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004821464393179539,
      "loss": 3.2941,
      "theoretical_loss": 4.519834598104527,
      "tokens_seen": 149684224
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.000482136409227683,
      "loss": 3.5809,
      "theoretical_loss": 4.5195758866391476,
      "tokens_seen": 149749760
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00048212637913741226,
      "loss": 3.3155,
      "theoretical_loss": 4.519317320056901,
      "tokens_seen": 149815296
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004821163490471414,
      "loss": 3.3261,
      "theoretical_loss": 4.519058898213325,
      "tokens_seen": 149880832
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004821063189568706,
      "loss": 3.2887,
      "theoretical_loss": 4.518800620964164,
      "tokens_seen": 149946368
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004820962888665998,
      "loss": 3.5263,
      "theoretical_loss": 4.51854248816537,
      "tokens_seen": 150011904
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.000482086258776329,
      "loss": 3.5012,
      "theoretical_loss": 4.5182844996731015,
      "tokens_seen": 150077440
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00048207622868605816,
      "loss": 3.5302,
      "theoretical_loss": 4.5180266553437205,
      "tokens_seen": 150142976
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004820661985957874,
      "loss": 3.5738,
      "theoretical_loss": 4.517768955033798,
      "tokens_seen": 150208512
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00048205616850551653,
      "loss": 3.606,
      "theoretical_loss": 4.5175113986001065,
      "tokens_seen": 150274048
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00048204613841524576,
      "loss": 3.4885,
      "theoretical_loss": 4.517253985899628,
      "tokens_seen": 150339584
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004820361083249749,
      "loss": 3.3201,
      "theoretical_loss": 4.516996716789544,
      "tokens_seen": 150405120
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004820260782347041,
      "loss": 3.457,
      "theoretical_loss": 4.516739591127244,
      "tokens_seen": 150470656
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004820160481444333,
      "loss": 3.1456,
      "theoretical_loss": 4.516482608770319,
      "tokens_seen": 150536192
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004820060180541625,
      "loss": 3.5659,
      "theoretical_loss": 4.516225769576564,
      "tokens_seen": 150601728
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00048199598796389167,
      "loss": 3.2205,
      "theoretical_loss": 4.515969073403976,
      "tokens_seen": 150667264
    },
    {
      "epoch": 0.05,
      "objective/train/advantage_avg": 0.49585628509521484,
      "objective/train/docs_used": 96442,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.439760446548462,
      "objective/train/original_loss": 3.439760446548462,
      "objective/train/theoretical_loss": 4.515712520110756,
      "objective/train/tokens_used": 171192800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24592286348342896,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0508357286453247,
      "objective/train/weighted_lm_loss": 3.6146368980407715,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0432155132293701,
      "theoretical_loss": 4.515712520110756,
      "tokens_seen": 150732800
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00048198595787362085,
      "loss": 3.4775,
      "theoretical_loss": 4.515712520110756,
      "tokens_seen": 150732800
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004819759277833501,
      "loss": 3.5096,
      "theoretical_loss": 4.515456109555306,
      "tokens_seen": 150798336
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00048196589769307927,
      "loss": 3.4557,
      "theoretical_loss": 4.515199841596231,
      "tokens_seen": 150863872
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00048195586760280845,
      "loss": 3.2938,
      "theoretical_loss": 4.514943716092336,
      "tokens_seen": 150929408
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00048194583751253763,
      "loss": 3.5835,
      "theoretical_loss": 4.514687732902628,
      "tokens_seen": 150994944
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00048193580742226686,
      "loss": 3.2485,
      "theoretical_loss": 4.514431891886314,
      "tokens_seen": 151060480
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.000481925777331996,
      "loss": 3.6166,
      "theoretical_loss": 4.514176192902802,
      "tokens_seen": 151126016
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004819157472417252,
      "loss": 3.5709,
      "theoretical_loss": 4.513920635811699,
      "tokens_seen": 151191552
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00048190571715145435,
      "loss": 3.4013,
      "theoretical_loss": 4.513665220472813,
      "tokens_seen": 151257088
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004818956870611836,
      "loss": 3.5024,
      "theoretical_loss": 4.513409946746149,
      "tokens_seen": 151322624
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00048188565697091277,
      "loss": 3.7843,
      "theoretical_loss": 4.513154814491914,
      "tokens_seen": 151388160
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00048187562688064195,
      "loss": 3.3515,
      "theoretical_loss": 4.5128998235705104,
      "tokens_seen": 151453696
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00048186559679037113,
      "loss": 3.4757,
      "theoretical_loss": 4.512644973842541,
      "tokens_seen": 151519232
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004818555667001003,
      "loss": 3.3592,
      "theoretical_loss": 4.512390265168802,
      "tokens_seen": 151584768
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004818455366098295,
      "loss": 3.289,
      "theoretical_loss": 4.512135697410294,
      "tokens_seen": 151650304
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00048183550651955873,
      "loss": 3.5237,
      "theoretical_loss": 4.511881270428207,
      "tokens_seen": 151715840
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00048182547642928786,
      "loss": 3.398,
      "theoretical_loss": 4.511626984083934,
      "tokens_seen": 151781376
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004818154463390171,
      "loss": 3.4929,
      "theoretical_loss": 4.51137283823906,
      "tokens_seen": 151846912
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004818054162487462,
      "loss": 3.335,
      "theoretical_loss": 4.511118832755368,
      "tokens_seen": 151912448
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00048179538615847545,
      "loss": 3.3739,
      "theoretical_loss": 4.510864967494836,
      "tokens_seen": 151977984
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00048178535606820463,
      "loss": 3.2394,
      "theoretical_loss": 4.510611242319637,
      "tokens_seen": 152043520
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004817753259779338,
      "loss": 3.3228,
      "theoretical_loss": 4.51035765709214,
      "tokens_seen": 152109056
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.000481765295887663,
      "loss": 3.5928,
      "theoretical_loss": 4.510104211674904,
      "tokens_seen": 152174592
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00048175526579739223,
      "loss": 3.2652,
      "theoretical_loss": 4.509850905930689,
      "tokens_seen": 152240128
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00048174523570712136,
      "loss": 3.5293,
      "theoretical_loss": 4.509597739722444,
      "tokens_seen": 152305664
    },
    {
      "epoch": 0.05,
      "objective/train/advantage_avg": 0.49441292881965637,
      "objective/train/docs_used": 97778,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.802095890045166,
      "objective/train/original_loss": 2.802095413208008,
      "objective/train/theoretical_loss": 4.509344712913311,
      "objective/train/tokens_used": 172831200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24452422559261322,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506843328475952,
      "objective/train/weighted_lm_loss": 2.944071054458618,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0304546356201172,
      "theoretical_loss": 4.509344712913311,
      "tokens_seen": 152371200
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004817352056168506,
      "loss": 3.4597,
      "theoretical_loss": 4.509344712913311,
      "tokens_seen": 152371200
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004817251755265797,
      "loss": 3.5686,
      "theoretical_loss": 4.509091825366629,
      "tokens_seen": 152436736
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00048171514543630896,
      "loss": 3.5266,
      "theoretical_loss": 4.508839076945925,
      "tokens_seen": 152502272
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00048170511534603814,
      "loss": 3.3371,
      "theoretical_loss": 4.508586467514924,
      "tokens_seen": 152567808
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004816950852557673,
      "loss": 3.5079,
      "theoretical_loss": 4.508333996937535,
      "tokens_seen": 152633344
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004816850551654965,
      "loss": 3.4987,
      "theoretical_loss": 4.5080816650778655,
      "tokens_seen": 152698880
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004816750250752257,
      "loss": 3.6936,
      "theoretical_loss": 4.5078294718002105,
      "tokens_seen": 152764416
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00048166499498495486,
      "loss": 3.3043,
      "theoretical_loss": 4.50757741696906,
      "tokens_seen": 152829952
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004816549648946841,
      "loss": 3.3471,
      "theoretical_loss": 4.507325500449089,
      "tokens_seen": 152895488
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004816449348044132,
      "loss": 3.5877,
      "theoretical_loss": 4.507073722105167,
      "tokens_seen": 152961024
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00048163490471414246,
      "loss": 3.7289,
      "theoretical_loss": 4.506822081802351,
      "tokens_seen": 153026560
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004816248746238716,
      "loss": 3.3053,
      "theoretical_loss": 4.506570579405888,
      "tokens_seen": 153092096
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004816148445336008,
      "loss": 3.3464,
      "theoretical_loss": 4.506319214781215,
      "tokens_seen": 153157632
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00048160481444333,
      "loss": 3.495,
      "theoretical_loss": 4.5060679877939585,
      "tokens_seen": 153223168
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004815947843530592,
      "loss": 3.3312,
      "theoretical_loss": 4.505816898309931,
      "tokens_seen": 153288704
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00048158475426278837,
      "loss": 3.5656,
      "theoretical_loss": 4.505565946195134,
      "tokens_seen": 153354240
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004815747241725176,
      "loss": 3.708,
      "theoretical_loss": 4.505315131315757,
      "tokens_seen": 153419776
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00048156469408224673,
      "loss": 3.7638,
      "theoretical_loss": 4.5050644535381785,
      "tokens_seen": 153485312
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00048155466399197596,
      "loss": 3.5609,
      "theoretical_loss": 4.504813912728961,
      "tokens_seen": 153550848
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004815446339017051,
      "loss": 3.3579,
      "theoretical_loss": 4.504563508754857,
      "tokens_seen": 153616384
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004815346038114343,
      "loss": 3.361,
      "theoretical_loss": 4.504313241482803,
      "tokens_seen": 153681920
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004815245737211635,
      "loss": 3.525,
      "theoretical_loss": 4.5040631107799225,
      "tokens_seen": 153747456
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004815145436308927,
      "loss": 3.6776,
      "theoretical_loss": 4.503813116513525,
      "tokens_seen": 153812992
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00048150451354062187,
      "loss": 3.2419,
      "theoretical_loss": 4.503563258551105,
      "tokens_seen": 153878528
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00048149448345035105,
      "loss": 3.3474,
      "theoretical_loss": 4.503313536760341,
      "tokens_seen": 153944064
    },
    {
      "epoch": 0.05,
      "objective/train/advantage_avg": 0.4937589168548584,
      "objective/train/docs_used": 98099,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.3721837997436523,
      "objective/train/original_loss": 3.3721835613250732,
      "objective/train/theoretical_loss": 4.503063951009098,
      "objective/train/tokens_used": 174469600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24388092756271362,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050615668296814,
      "objective/train/weighted_lm_loss": 3.5428178310394287,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0363596677780151,
      "theoretical_loss": 4.503063951009098,
      "tokens_seen": 154009600
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00048148445336008023,
      "loss": 3.604,
      "theoretical_loss": 4.503063951009098,
      "tokens_seen": 154009600
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00048147442326980947,
      "loss": 3.5015,
      "theoretical_loss": 4.502814501165425,
      "tokens_seen": 154075136
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004814643931795386,
      "loss": 3.5659,
      "theoretical_loss": 4.502565187097554,
      "tokens_seen": 154140672
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00048145436308926783,
      "loss": 3.4027,
      "theoretical_loss": 4.502316008673902,
      "tokens_seen": 154206208
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00048144433299899696,
      "loss": 3.1147,
      "theoretical_loss": 4.502066965763069,
      "tokens_seen": 154271744
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004814343029087262,
      "loss": 3.4261,
      "theoretical_loss": 4.5018180582338365,
      "tokens_seen": 154337280
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00048142427281845537,
      "loss": 3.5209,
      "theoretical_loss": 4.501569285955171,
      "tokens_seen": 154402816
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00048141424272818455,
      "loss": 3.1747,
      "theoretical_loss": 4.50132064879622,
      "tokens_seen": 154468352
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00048140421263791373,
      "loss": 3.5698,
      "theoretical_loss": 4.5010721466263135,
      "tokens_seen": 154533888
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00048139418254764297,
      "loss": 3.6406,
      "theoretical_loss": 4.500823779314965,
      "tokens_seen": 154599424
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004813841524573721,
      "loss": 3.4173,
      "theoretical_loss": 4.500575546731865,
      "tokens_seen": 154664960
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00048137412236710133,
      "loss": 3.5286,
      "theoretical_loss": 4.500327448746887,
      "tokens_seen": 154730496
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00048136409227683046,
      "loss": 3.6468,
      "theoretical_loss": 4.5000794852300885,
      "tokens_seen": 154796032
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004813540621865597,
      "loss": 3.4764,
      "theoretical_loss": 4.499831656051704,
      "tokens_seen": 154861568
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004813440320962889,
      "loss": 3.4454,
      "theoretical_loss": 4.499583961082147,
      "tokens_seen": 154927104
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00048133400200601806,
      "loss": 3.7611,
      "theoretical_loss": 4.499336400192014,
      "tokens_seen": 154992640
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00048132397191574724,
      "loss": 3.429,
      "theoretical_loss": 4.49908897325208,
      "tokens_seen": 155058176
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004813139418254764,
      "loss": 3.2862,
      "theoretical_loss": 4.498841680133298,
      "tokens_seen": 155123712
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004813039117352056,
      "loss": 3.5066,
      "theoretical_loss": 4.498594520706801,
      "tokens_seen": 155189248
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00048129388164493483,
      "loss": 3.4153,
      "theoretical_loss": 4.498347494843897,
      "tokens_seen": 155254784
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00048128385155466396,
      "loss": 3.4468,
      "theoretical_loss": 4.498100602416079,
      "tokens_seen": 155320320
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004812738214643932,
      "loss": 3.6878,
      "theoretical_loss": 4.49785384329501,
      "tokens_seen": 155385856
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004812637913741223,
      "loss": 3.732,
      "theoretical_loss": 4.497607217352536,
      "tokens_seen": 155451392
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00048125376128385156,
      "loss": 3.383,
      "theoretical_loss": 4.497360724460679,
      "tokens_seen": 155516928
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00048124373119358074,
      "loss": 3.3907,
      "theoretical_loss": 4.497114364491635,
      "tokens_seen": 155582464
    },
    {
      "epoch": 0.05,
      "objective/train/advantage_avg": 0.4935433268547058,
      "objective/train/docs_used": 99496,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.3396732807159424,
      "objective/train/original_loss": 3.3396730422973633,
      "objective/train/theoretical_loss": 4.496868137317781,
      "objective/train/tokens_used": 176108000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24367322027683258,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505928993225098,
      "objective/train/weighted_lm_loss": 3.5085935592651367,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0392107963562012,
      "theoretical_loss": 4.496868137317781,
      "tokens_seen": 155648000
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004812337011033099,
      "loss": 3.1379,
      "theoretical_loss": 4.496868137317781,
      "tokens_seen": 155648000
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00048122367101303916,
      "loss": 3.543,
      "theoretical_loss": 4.496622042811667,
      "tokens_seen": 155713536
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00048121364092276834,
      "loss": 3.4175,
      "theoretical_loss": 4.496376080846021,
      "tokens_seen": 155779072
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004812036108324975,
      "loss": 3.5759,
      "theoretical_loss": 4.496130251293744,
      "tokens_seen": 155844608
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004811935807422267,
      "loss": 3.5886,
      "theoretical_loss": 4.4958845540279135,
      "tokens_seen": 155910144
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004811835506519559,
      "loss": 3.3216,
      "theoretical_loss": 4.495638988921783,
      "tokens_seen": 155975680
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00048117352056168506,
      "loss": 3.5736,
      "theoretical_loss": 4.495393555848779,
      "tokens_seen": 156041216
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004811634904714143,
      "loss": 3.3189,
      "theoretical_loss": 4.495148254682505,
      "tokens_seen": 156106752
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004811534603811434,
      "loss": 3.7279,
      "theoretical_loss": 4.494903085296733,
      "tokens_seen": 156172288
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00048114343029087266,
      "loss": 3.5747,
      "theoretical_loss": 4.494658047565416,
      "tokens_seen": 156237824
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004811334002006018,
      "loss": 3.6717,
      "theoretical_loss": 4.494413141362674,
      "tokens_seen": 156303360
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.000481123370110331,
      "loss": 3.5059,
      "theoretical_loss": 4.494168366562805,
      "tokens_seen": 156368896
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004811133400200602,
      "loss": 3.5633,
      "theoretical_loss": 4.493923723040275,
      "tokens_seen": 156434432
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004811033099297894,
      "loss": 3.2011,
      "theoretical_loss": 4.493679210669727,
      "tokens_seen": 156499968
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00048109327983951857,
      "loss": 3.3097,
      "theoretical_loss": 4.493434829325974,
      "tokens_seen": 156565504
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004810832497492478,
      "loss": 3.3113,
      "theoretical_loss": 4.493190578883999,
      "tokens_seen": 156631040
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00048107321965897693,
      "loss": 3.0799,
      "theoretical_loss": 4.49294645921896,
      "tokens_seen": 156696576
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00048106318956870616,
      "loss": 3.5264,
      "theoretical_loss": 4.492702470206184,
      "tokens_seen": 156762112
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004810531594784353,
      "loss": 3.5027,
      "theoretical_loss": 4.49245861172117,
      "tokens_seen": 156827648
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004810431293881645,
      "loss": 3.3555,
      "theoretical_loss": 4.492214883639586,
      "tokens_seen": 156893184
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004810330992978937,
      "loss": 3.4403,
      "theoretical_loss": 4.491971285837273,
      "tokens_seen": 156958720
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004810230692076229,
      "loss": 3.468,
      "theoretical_loss": 4.49172781819024,
      "tokens_seen": 157024256
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00048101303911735207,
      "loss": 3.4509,
      "theoretical_loss": 4.491484480574666,
      "tokens_seen": 157089792
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00048100300902708125,
      "loss": 3.6654,
      "theoretical_loss": 4.4912412728668984,
      "tokens_seen": 157155328
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00048099297893681043,
      "loss": 3.4109,
      "theoretical_loss": 4.490998194943456,
      "tokens_seen": 157220864
    },
    {
      "epoch": 0.05,
      "objective/train/advantage_avg": 0.49436303973197937,
      "objective/train/docs_used": 100000,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.4820659160614014,
      "objective/train/original_loss": 3.4820661544799805,
      "objective/train/theoretical_loss": 4.490755246681026,
      "objective/train/tokens_used": 177746400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24443647265434265,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506788492202759,
      "objective/train/weighted_lm_loss": 3.6586177349090576,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0408108234405518,
      "theoretical_loss": 4.490755246681026,
      "tokens_seen": 157286400
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00048098294884653967,
      "loss": 3.3307,
      "theoretical_loss": 4.490755246681026,
      "tokens_seen": 157286400
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004809729187562688,
      "loss": 3.5335,
      "theoretical_loss": 4.490512427956462,
      "tokens_seen": 157351936
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00048096288866599803,
      "loss": 3.591,
      "theoretical_loss": 4.490269738646786,
      "tokens_seen": 157417472
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00048095285857572716,
      "loss": 3.523,
      "theoretical_loss": 4.490027178629192,
      "tokens_seen": 157483008
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004809428284854564,
      "loss": 3.4958,
      "theoretical_loss": 4.4897847477810355,
      "tokens_seen": 157548544
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00048093279839518557,
      "loss": 3.3983,
      "theoretical_loss": 4.489542445979844,
      "tokens_seen": 157614080
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00048092276830491475,
      "loss": 3.5395,
      "theoretical_loss": 4.489300273103311,
      "tokens_seen": 157679616
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00048091273821464393,
      "loss": 3.5374,
      "theoretical_loss": 4.489058229029293,
      "tokens_seen": 157745152
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00048090270812437317,
      "loss": 3.3385,
      "theoretical_loss": 4.4888163136358195,
      "tokens_seen": 157810688
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004808926780341023,
      "loss": 3.5738,
      "theoretical_loss": 4.48857452680108,
      "tokens_seen": 157876224
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00048088264794383153,
      "loss": 3.3575,
      "theoretical_loss": 4.4883328684034325,
      "tokens_seen": 157941760
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00048087261785356066,
      "loss": 3.5976,
      "theoretical_loss": 4.488091338321402,
      "tokens_seen": 158007296
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004808625877632899,
      "loss": 3.3101,
      "theoretical_loss": 4.487849936433676,
      "tokens_seen": 158072832
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004808525576730191,
      "loss": 3.3155,
      "theoretical_loss": 4.487608662619107,
      "tokens_seen": 158138368
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00048084252758274826,
      "loss": 3.5933,
      "theoretical_loss": 4.487367516756715,
      "tokens_seen": 158203904
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00048083249749247744,
      "loss": 3.6469,
      "theoretical_loss": 4.487126498725679,
      "tokens_seen": 158269440
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004808224674022066,
      "loss": 3.634,
      "theoretical_loss": 4.48688560840535,
      "tokens_seen": 158334976
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004808124373119358,
      "loss": 3.3629,
      "theoretical_loss": 4.486644845675235,
      "tokens_seen": 158400512
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00048080240722166503,
      "loss": 3.3318,
      "theoretical_loss": 4.48640421041501,
      "tokens_seen": 158466048
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00048079237713139416,
      "loss": 3.4746,
      "theoretical_loss": 4.486163702504509,
      "tokens_seen": 158531584
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004807823470411234,
      "loss": 3.3411,
      "theoretical_loss": 4.4859233218237335,
      "tokens_seen": 158597120
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004807723169508525,
      "loss": 3.35,
      "theoretical_loss": 4.485683068252847,
      "tokens_seen": 158662656
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00048076228686058176,
      "loss": 3.6398,
      "theoretical_loss": 4.485442941672172,
      "tokens_seen": 158728192
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00048075225677031094,
      "loss": 3.5667,
      "theoretical_loss": 4.485202941962196,
      "tokens_seen": 158793728
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004807422266800401,
      "loss": 3.3531,
      "theoretical_loss": 4.484963069003568,
      "tokens_seen": 158859264
    },
    {
      "epoch": 0.05,
      "objective/train/advantage_avg": 0.49529367685317993,
      "objective/train/docs_used": 101476,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.3036837577819824,
      "objective/train/original_loss": 3.3036842346191406,
      "objective/train/theoretical_loss": 4.484723322677097,
      "objective/train/tokens_used": 179384800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24535971879959106,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050776720046997,
      "objective/train/weighted_lm_loss": 3.471553087234497,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0428944826126099,
      "theoretical_loss": 4.484723322677097,
      "tokens_seen": 158924800
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004807321965897693,
      "loss": 3.431,
      "theoretical_loss": 4.484723322677097,
      "tokens_seen": 158924800
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00048072216649949854,
      "loss": 3.6562,
      "theoretical_loss": 4.484483702863755,
      "tokens_seen": 158990336
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00048071213640922766,
      "loss": 3.6208,
      "theoretical_loss": 4.4842442094446735,
      "tokens_seen": 159055872
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004807021063189569,
      "loss": 3.4032,
      "theoretical_loss": 4.484004842301145,
      "tokens_seen": 159121408
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.000480692076228686,
      "loss": 3.3958,
      "theoretical_loss": 4.483765601314621,
      "tokens_seen": 159186944
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00048068204613841526,
      "loss": 3.6535,
      "theoretical_loss": 4.483526486366717,
      "tokens_seen": 159252480
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00048067201604814444,
      "loss": 3.3515,
      "theoretical_loss": 4.483287497339205,
      "tokens_seen": 159318016
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004806619859578736,
      "loss": 3.4659,
      "theoretical_loss": 4.483048634114016,
      "tokens_seen": 159383552
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004806519558676028,
      "loss": 3.2996,
      "theoretical_loss": 4.482809896573242,
      "tokens_seen": 159449088
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.000480641925777332,
      "loss": 3.3367,
      "theoretical_loss": 4.482571284599134,
      "tokens_seen": 159514624
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00048063189568706117,
      "loss": 3.5892,
      "theoretical_loss": 4.482332798074099,
      "tokens_seen": 159580160
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004806218655967904,
      "loss": 3.6079,
      "theoretical_loss": 4.482094436880706,
      "tokens_seen": 159645696
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00048061183550651953,
      "loss": 3.6944,
      "theoretical_loss": 4.481856200901679,
      "tokens_seen": 159711232
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00048060180541624877,
      "loss": 3.8891,
      "theoretical_loss": 4.481618090019903,
      "tokens_seen": 159776768
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00048059177532597795,
      "loss": 3.5406,
      "theoretical_loss": 4.481380104118417,
      "tokens_seen": 159842304
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00048058174523570713,
      "loss": 3.5644,
      "theoretical_loss": 4.481142243080419,
      "tokens_seen": 159907840
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004805717151454363,
      "loss": 3.217,
      "theoretical_loss": 4.480904506789265,
      "tokens_seen": 159973376
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004805616850551655,
      "loss": 3.4773,
      "theoretical_loss": 4.480666895128465,
      "tokens_seen": 160038912
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00048055165496489467,
      "loss": 3.2264,
      "theoretical_loss": 4.480429407981688,
      "tokens_seen": 160104448
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004805416248746239,
      "loss": 3.92,
      "theoretical_loss": 4.4801920452327595,
      "tokens_seen": 160169984
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00048053159478435303,
      "loss": 3.5036,
      "theoretical_loss": 4.479954806765657,
      "tokens_seen": 160235520
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00048052156469408227,
      "loss": 3.1436,
      "theoretical_loss": 4.479717692464517,
      "tokens_seen": 160301056
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004805115346038114,
      "loss": 3.47,
      "theoretical_loss": 4.479480702213631,
      "tokens_seen": 160366592
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00048050150451354063,
      "loss": 3.3881,
      "theoretical_loss": 4.479243835897444,
      "tokens_seen": 160432128
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004804914744232698,
      "loss": 3.3933,
      "theoretical_loss": 4.479007093400558,
      "tokens_seen": 160497664
    },
    {
      "epoch": 0.05,
      "objective/train/advantage_avg": 0.49491098523139954,
      "objective/train/docs_used": 102054,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.2796711921691895,
      "objective/train/original_loss": 3.2796711921691895,
      "objective/train/theoretical_loss": 4.478770474607726,
      "objective/train/tokens_used": 181023200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24500545859336853,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050736665725708,
      "objective/train/weighted_lm_loss": 3.4458789825439453,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0396549701690674,
      "theoretical_loss": 4.478770474607726,
      "tokens_seen": 160563200
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.000480481444332999,
      "loss": 3.3783,
      "theoretical_loss": 4.478770474607726,
      "tokens_seen": 160563200
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00048047141424272823,
      "loss": 3.5345,
      "theoretical_loss": 4.478533979403862,
      "tokens_seen": 160628736
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00048046138415245736,
      "loss": 3.4893,
      "theoretical_loss": 4.478297607674024,
      "tokens_seen": 160694272
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004804513540621866,
      "loss": 3.4277,
      "theoretical_loss": 4.4780613593034335,
      "tokens_seen": 160759808
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00048044132397191577,
      "loss": 3.3443,
      "theoretical_loss": 4.47782523417746,
      "tokens_seen": 160825344
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00048043129388164495,
      "loss": 3.3301,
      "theoretical_loss": 4.477589232181627,
      "tokens_seen": 160890880
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00048042126379137413,
      "loss": 3.342,
      "theoretical_loss": 4.47735335320161,
      "tokens_seen": 160956416
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00048041123370110337,
      "loss": 3.5606,
      "theoretical_loss": 4.47711759712324,
      "tokens_seen": 161021952
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004804012036108325,
      "loss": 3.3574,
      "theoretical_loss": 4.476881963832499,
      "tokens_seen": 161087488
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00048039117352056173,
      "loss": 3.3906,
      "theoretical_loss": 4.476646453215521,
      "tokens_seen": 161153024
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00048038114343029086,
      "loss": 3.2207,
      "theoretical_loss": 4.476411065158589,
      "tokens_seen": 161218560
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004803711133400201,
      "loss": 3.7256,
      "theoretical_loss": 4.476175799548144,
      "tokens_seen": 161284096
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004803610832497493,
      "loss": 3.476,
      "theoretical_loss": 4.475940656270772,
      "tokens_seen": 161349632
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00048035105315947846,
      "loss": 3.5265,
      "theoretical_loss": 4.475705635213215,
      "tokens_seen": 161415168
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00048034102306920764,
      "loss": 3.4927,
      "theoretical_loss": 4.475470736262361,
      "tokens_seen": 161480704
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004803309929789368,
      "loss": 3.3691,
      "theoretical_loss": 4.475235959305252,
      "tokens_seen": 161546240
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.000480320962888666,
      "loss": 3.2592,
      "theoretical_loss": 4.47500130422908,
      "tokens_seen": 161611776
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00048031093279839523,
      "loss": 3.4047,
      "theoretical_loss": 4.4747667709211845,
      "tokens_seen": 161677312
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00048030090270812436,
      "loss": 3.6386,
      "theoretical_loss": 4.474532359269059,
      "tokens_seen": 161742848
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004802908726178536,
      "loss": 3.1313,
      "theoretical_loss": 4.474298069160341,
      "tokens_seen": 161808384
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004802808425275827,
      "loss": 3.5661,
      "theoretical_loss": 4.4740639004828235,
      "tokens_seen": 161873920
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00048027081243731196,
      "loss": 3.1891,
      "theoretical_loss": 4.4738298531244425,
      "tokens_seen": 161939456
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00048026078234704114,
      "loss": 3.6931,
      "theoretical_loss": 4.473595926973286,
      "tokens_seen": 162004992
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004802507522567703,
      "loss": 3.4435,
      "theoretical_loss": 4.473362121917591,
      "tokens_seen": 162070528
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004802407221664995,
      "loss": 3.4266,
      "theoretical_loss": 4.473128437845741,
      "tokens_seen": 162136064
    },
    {
      "epoch": 0.05,
      "objective/train/advantage_avg": 0.49375954270362854,
      "objective/train/docs_used": 102718,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.605020046234131,
      "objective/train/original_loss": 3.60502028465271,
      "objective/train/theoretical_loss": 4.4728948746462684,
      "objective/train/tokens_used": 182661600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24387326836585999,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050615668296814,
      "objective/train/weighted_lm_loss": 3.7875723838806152,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0425469875335693,
      "theoretical_loss": 4.4728948746462684,
      "tokens_seen": 162201600
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00048023069207622874,
      "loss": 3.6002,
      "theoretical_loss": 4.4728948746462684,
      "tokens_seen": 162201600
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00048022066198595786,
      "loss": 3.2531,
      "theoretical_loss": 4.47266143220785,
      "tokens_seen": 162267136
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004802106318956871,
      "loss": 3.2391,
      "theoretical_loss": 4.472428110419317,
      "tokens_seen": 162332672
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00048020060180541623,
      "loss": 3.1836,
      "theoretical_loss": 4.472194909169643,
      "tokens_seen": 162398208
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00048019057171514546,
      "loss": 3.5938,
      "theoretical_loss": 4.471961828347947,
      "tokens_seen": 162463744
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00048018054162487464,
      "loss": 3.4451,
      "theoretical_loss": 4.471728867843497,
      "tokens_seen": 162529280
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004801705115346038,
      "loss": 3.3494,
      "theoretical_loss": 4.47149602754571,
      "tokens_seen": 162594816
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.000480160481444333,
      "loss": 3.7168,
      "theoretical_loss": 4.4712633073441435,
      "tokens_seen": 162660352
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004801504513540622,
      "loss": 3.2493,
      "theoretical_loss": 4.471030707128506,
      "tokens_seen": 162725888
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00048014042126379137,
      "loss": 3.2841,
      "theoretical_loss": 4.470798226788649,
      "tokens_seen": 162791424
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004801303911735206,
      "loss": 3.5582,
      "theoretical_loss": 4.470565866214568,
      "tokens_seen": 162856960
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00048012036108324973,
      "loss": 3.2854,
      "theoretical_loss": 4.4703336252964085,
      "tokens_seen": 162922496
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00048011033099297897,
      "loss": 3.4537,
      "theoretical_loss": 4.470101503924456,
      "tokens_seen": 162988032
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00048010030090270815,
      "loss": 3.5539,
      "theoretical_loss": 4.469869501989144,
      "tokens_seen": 163053568
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00048009027081243733,
      "loss": 3.3577,
      "theoretical_loss": 4.469637619381047,
      "tokens_seen": 163119104
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004800802407221665,
      "loss": 3.5634,
      "theoretical_loss": 4.469405855990889,
      "tokens_seen": 163184640
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004800702106318957,
      "loss": 3.4557,
      "theoretical_loss": 4.469174211709531,
      "tokens_seen": 163250176
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00048006018054162487,
      "loss": 3.3329,
      "theoretical_loss": 4.468942686427985,
      "tokens_seen": 163315712
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004800501504513541,
      "loss": 3.3671,
      "theoretical_loss": 4.4687112800373985,
      "tokens_seen": 163381248
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00048004012036108323,
      "loss": 3.6252,
      "theoretical_loss": 4.46847999242907,
      "tokens_seen": 163446784
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00048003009027081247,
      "loss": 3.2167,
      "theoretical_loss": 4.468248823494436,
      "tokens_seen": 163512320
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004800200601805416,
      "loss": 3.4799,
      "theoretical_loss": 4.4680177731250765,
      "tokens_seen": 163577856
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00048001003009027083,
      "loss": 3.4754,
      "theoretical_loss": 4.467786841212715,
      "tokens_seen": 163643392
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00048,
      "loss": 3.3587,
      "theoretical_loss": 4.467556027649216,
      "tokens_seen": 163708928
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004799899699097292,
      "loss": 3.2968,
      "theoretical_loss": 4.467325332326588,
      "tokens_seen": 163774464
    },
    {
      "debugging/Self-BLEU-5": 0.4072555680918346,
      "debugging/distinct-1-grams": 0.8137531241311615,
      "debugging/distinct-2-grams": 0.9866498101723051,
      "debugging/entropy-1-grams": 5.4466195562752855,
      "debugging/entropy-2-grams": 6.104764435286928,
      "debugging/length": 544.5,
      "debugging/num_segments": 6,
      "debugging/raw_token_scores_avg": 0.003607792081311345,
      "debugging/raw_token_scores_std": 0.005498606711626053,
      "debugging/score": 0.002729943707102512,
      "debugging/score_std": 0.0025596098515178914,
      "epoch": 0.05,
      "objective/train/advantage_avg": 0.49639105796813965,
      "objective/train/docs_used": 104069,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.1349663734436035,
      "objective/train/original_loss": 3.1349668502807617,
      "objective/train/theoretical_loss": 4.467094755136979,
      "objective/train/tokens_used": 184300000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24643433094024658,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0508918762207031,
      "objective/train/weighted_lm_loss": 3.2943742275238037,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0468999147415161,
      "theoretical_loss": 4.467094755136979,
      "tokens_seen": 163840000
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004799799398194584,
      "loss": 3.534,
      "theoretical_loss": 4.467094755136979,
      "tokens_seen": 163840000
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047996990972918756,
      "loss": 3.4204,
      "theoretical_loss": 4.4668642959726785,
      "tokens_seen": 163905536
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047995987963891674,
      "loss": 3.6429,
      "theoretical_loss": 4.466633954726119,
      "tokens_seen": 163971072
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047994984954864597,
      "loss": 3.2499,
      "theoretical_loss": 4.466403731289873,
      "tokens_seen": 164036608
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004799398194583751,
      "loss": 3.3139,
      "theoretical_loss": 4.466173625556653,
      "tokens_seen": 164102144
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047992978936810433,
      "loss": 3.5283,
      "theoretical_loss": 4.465943637419313,
      "tokens_seen": 164167680
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004799197592778335,
      "loss": 3.1685,
      "theoretical_loss": 4.465713766770847,
      "tokens_seen": 164233216
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004799097291875627,
      "loss": 3.4739,
      "theoretical_loss": 4.465484013504389,
      "tokens_seen": 164298752
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004798996990972919,
      "loss": 3.2571,
      "theoretical_loss": 4.4652543775132125,
      "tokens_seen": 164364288
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047988966900702106,
      "loss": 3.3555,
      "theoretical_loss": 4.465024858690731,
      "tokens_seen": 164429824
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047987963891675024,
      "loss": 3.55,
      "theoretical_loss": 4.4647954569304975,
      "tokens_seen": 164495360
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004798696088264795,
      "loss": 3.2888,
      "theoretical_loss": 4.464566172126203,
      "tokens_seen": 164560896
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004798595787362086,
      "loss": 3.4083,
      "theoretical_loss": 4.464337004171679,
      "tokens_seen": 164626432
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047984954864593784,
      "loss": 3.2287,
      "theoretical_loss": 4.464107952960893,
      "tokens_seen": 164691968
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047983951855566696,
      "loss": 3.3855,
      "theoretical_loss": 4.463879018387954,
      "tokens_seen": 164757504
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004798294884653962,
      "loss": 3.3134,
      "theoretical_loss": 4.463650200347106,
      "tokens_seen": 164823040
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004798194583751254,
      "loss": 3.2988,
      "theoretical_loss": 4.4634214987327345,
      "tokens_seen": 164888576
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047980942828485456,
      "loss": 3.3721,
      "theoretical_loss": 4.463192913439359,
      "tokens_seen": 164954112
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047979939819458374,
      "loss": 3.6735,
      "theoretical_loss": 4.4629644443616385,
      "tokens_seen": 165019648
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004797893681043129,
      "loss": 3.5188,
      "theoretical_loss": 4.46273609139437,
      "tokens_seen": 165085184
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004797793380140421,
      "loss": 3.3081,
      "theoretical_loss": 4.462507854432483,
      "tokens_seen": 165150720
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047976930792377134,
      "loss": 3.4414,
      "theoretical_loss": 4.46227973337105,
      "tokens_seen": 165216256
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047975927783350047,
      "loss": 3.5987,
      "theoretical_loss": 4.462051728105276,
      "tokens_seen": 165281792
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004797492477432297,
      "loss": 3.567,
      "theoretical_loss": 4.461823838530503,
      "tokens_seen": 165347328
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004797392176529589,
      "loss": 3.5135,
      "theoretical_loss": 4.461596064542208,
      "tokens_seen": 165412864
    },
    {
      "epoch": 0.05,
      "objective/train/advantage_avg": 0.49306467175483704,
      "objective/train/docs_used": 105300,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.002910614013672,
      "objective/train/original_loss": 3.002910614013672,
      "objective/train/theoretical_loss": 4.461368406036007,
      "objective/train/tokens_used": 185938400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2431906759738922,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505425930023193,
      "objective/train/weighted_lm_loss": 3.1547579765319824,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0363596677780151,
      "theoretical_loss": 4.461368406036007,
      "tokens_seen": 165478400
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047972918756268807,
      "loss": 3.1977,
      "theoretical_loss": 4.461368406036007,
      "tokens_seen": 165478400
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004797191574724173,
      "loss": 3.451,
      "theoretical_loss": 4.461140862907648,
      "tokens_seen": 165543936
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047970912738214643,
      "loss": 3.4912,
      "theoretical_loss": 4.4609134350530155,
      "tokens_seen": 165609472
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047969909729187566,
      "loss": 3.77,
      "theoretical_loss": 4.460686122368132,
      "tokens_seen": 165675008
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047968906720160484,
      "loss": 3.3778,
      "theoretical_loss": 4.460458924749149,
      "tokens_seen": 165740544
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.000479679037111334,
      "loss": 3.407,
      "theoretical_loss": 4.460231842092359,
      "tokens_seen": 165806080
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004796690070210632,
      "loss": 3.3434,
      "theoretical_loss": 4.460004874294185,
      "tokens_seen": 165871616
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004796589769307924,
      "loss": 3.4497,
      "theoretical_loss": 4.459778021251183,
      "tokens_seen": 165937152
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047964894684052157,
      "loss": 3.4431,
      "theoretical_loss": 4.459551282860048,
      "tokens_seen": 166002688
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004796389167502508,
      "loss": 3.466,
      "theoretical_loss": 4.459324659017604,
      "tokens_seen": 166068224
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047962888665997993,
      "loss": 3.4582,
      "theoretical_loss": 4.459098149620811,
      "tokens_seen": 166133760
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047961885656970917,
      "loss": 3.3992,
      "theoretical_loss": 4.458871754566761,
      "tokens_seen": 166199296
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047960882647943835,
      "loss": 3.2188,
      "theoretical_loss": 4.458645473752681,
      "tokens_seen": 166264832
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047959879638916753,
      "loss": 3.46,
      "theoretical_loss": 4.458419307075928,
      "tokens_seen": 166330368
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004795887662988967,
      "loss": 3.7723,
      "theoretical_loss": 4.458193254433994,
      "tokens_seen": 166395904
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004795787362086259,
      "loss": 3.7305,
      "theoretical_loss": 4.4579673157245026,
      "tokens_seen": 166461440
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047956870611835507,
      "loss": 3.2343,
      "theoretical_loss": 4.457741490845209,
      "tokens_seen": 166526976
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004795586760280843,
      "loss": 3.4758,
      "theoretical_loss": 4.457515779694001,
      "tokens_seen": 166592512
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047954864593781343,
      "loss": 3.1311,
      "theoretical_loss": 4.457290182168899,
      "tokens_seen": 166658048
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047953861584754267,
      "loss": 3.3415,
      "theoretical_loss": 4.457064698168051,
      "tokens_seen": 166723584
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004795285857572718,
      "loss": 3.284,
      "theoretical_loss": 4.456839327589742,
      "tokens_seen": 166789120
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047951855566700103,
      "loss": 3.4982,
      "theoretical_loss": 4.456614070332385,
      "tokens_seen": 166854656
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004795085255767302,
      "loss": 3.1079,
      "theoretical_loss": 4.456388926294522,
      "tokens_seen": 166920192
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004794984954864594,
      "loss": 3.244,
      "theoretical_loss": 4.45616389537483,
      "tokens_seen": 166985728
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004794884653961886,
      "loss": 3.5831,
      "theoretical_loss": 4.455938977472112,
      "tokens_seen": 167051264
    },
    {
      "epoch": 0.05,
      "objective/train/advantage_avg": 0.493816614151001,
      "objective/train/docs_used": 106148,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.7108566761016846,
      "objective/train/original_loss": 3.7108559608459473,
      "objective/train/theoretical_loss": 4.455714172485305,
      "objective/train/tokens_used": 187576800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24392615258693695,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506216287612915,
      "objective/train/weighted_lm_loss": 3.898561954498291,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0408108234405518,
      "theoretical_loss": 4.455714172485305,
      "tokens_seen": 167116800
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047947843530591776,
      "loss": 3.568,
      "theoretical_loss": 4.455714172485305,
      "tokens_seen": 167116800
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047946840521564694,
      "loss": 3.4664,
      "theoretical_loss": 4.4554894803134735,
      "tokens_seen": 167182336
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047945837512537617,
      "loss": 3.5439,
      "theoretical_loss": 4.455264900855813,
      "tokens_seen": 167247872
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004794483450351053,
      "loss": 3.5604,
      "theoretical_loss": 4.4550404340116465,
      "tokens_seen": 167313408
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047943831494483453,
      "loss": 3.4373,
      "theoretical_loss": 4.454816079680429,
      "tokens_seen": 167378944
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004794282848545637,
      "loss": 3.3157,
      "theoretical_loss": 4.454591837761743,
      "tokens_seen": 167444480
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004794182547642929,
      "loss": 3.2997,
      "theoretical_loss": 4.454367708155301,
      "tokens_seen": 167510016
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004794082246740221,
      "loss": 3.4866,
      "theoretical_loss": 4.454143690760943,
      "tokens_seen": 167575552
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047939819458375126,
      "loss": 3.194,
      "theoretical_loss": 4.453919785478638,
      "tokens_seen": 167641088
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047938816449348044,
      "loss": 3.1672,
      "theoretical_loss": 4.453695992208482,
      "tokens_seen": 167706624
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004793781344032097,
      "loss": 3.4815,
      "theoretical_loss": 4.453472310850701,
      "tokens_seen": 167772160
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004793681043129388,
      "loss": 3.4221,
      "theoretical_loss": 4.453248741305649,
      "tokens_seen": 167837696
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047935807422266804,
      "loss": 3.616,
      "theoretical_loss": 4.4530252834738056,
      "tokens_seen": 167903232
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047934804413239716,
      "loss": 3.3521,
      "theoretical_loss": 4.452801937255778,
      "tokens_seen": 167968768
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004793380140421264,
      "loss": 3.4454,
      "theoretical_loss": 4.4525787025523025,
      "tokens_seen": 168034304
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004793279839518556,
      "loss": 3.4192,
      "theoretical_loss": 4.452355579264241,
      "tokens_seen": 168099840
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047931795386158476,
      "loss": 3.5488,
      "theoretical_loss": 4.452132567292581,
      "tokens_seen": 168165376
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047930792377131394,
      "loss": 3.6895,
      "theoretical_loss": 4.451909666538439,
      "tokens_seen": 168230912
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004792978936810431,
      "loss": 3.4376,
      "theoretical_loss": 4.4516868769030555,
      "tokens_seen": 168296448
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004792878635907723,
      "loss": 3.4283,
      "theoretical_loss": 4.4514641982878,
      "tokens_seen": 168361984
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047927783350050154,
      "loss": 3.4137,
      "theoretical_loss": 4.451241630594166,
      "tokens_seen": 168427520
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047926780341023067,
      "loss": 3.1494,
      "theoretical_loss": 4.451019173723771,
      "tokens_seen": 168493056
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004792577733199599,
      "loss": 3.554,
      "theoretical_loss": 4.450796827578361,
      "tokens_seen": 168558592
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004792477432296891,
      "loss": 3.5715,
      "theoretical_loss": 4.450574592059806,
      "tokens_seen": 168624128
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047923771313941827,
      "loss": 3.5396,
      "theoretical_loss": 4.450352467070101,
      "tokens_seen": 168689664
    },
    {
      "epoch": 0.05,
      "objective/train/advantage_avg": 0.4956451952457428,
      "objective/train/docs_used": 106755,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.595431089401245,
      "objective/train/original_loss": 2.595431089401245,
      "objective/train/theoretical_loss": 4.450130452511366,
      "objective/train/tokens_used": 189215200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24570877850055695,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0508135557174683,
      "objective/train/weighted_lm_loss": 2.7270896434783936,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0408108234405518,
      "theoretical_loss": 4.450130452511366,
      "tokens_seen": 168755200
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047922768304914745,
      "loss": 3.1621,
      "theoretical_loss": 4.450130452511366,
      "tokens_seen": 168755200
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047921765295887663,
      "loss": 3.3973,
      "theoretical_loss": 4.449908548285846,
      "tokens_seen": 168820736
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004792076228686058,
      "loss": 3.1535,
      "theoretical_loss": 4.449686754295909,
      "tokens_seen": 168886272
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047919759277833504,
      "loss": 3.2878,
      "theoretical_loss": 4.44946507044405,
      "tokens_seen": 168951808
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047918756268806417,
      "loss": 3.6079,
      "theoretical_loss": 4.449243496632885,
      "tokens_seen": 169017344
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004791775325977934,
      "loss": 3.097,
      "theoretical_loss": 4.449022032765155,
      "tokens_seen": 169082880
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047916750250752253,
      "loss": 3.4705,
      "theoretical_loss": 4.448800678743726,
      "tokens_seen": 169148416
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047915747241725177,
      "loss": 3.4138,
      "theoretical_loss": 4.448579434471585,
      "tokens_seen": 169213952
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047914744232698095,
      "loss": 3.3962,
      "theoretical_loss": 4.4483582998518445,
      "tokens_seen": 169279488
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047913741223671013,
      "loss": 3.4523,
      "theoretical_loss": 4.448137274787738,
      "tokens_seen": 169345024
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004791273821464393,
      "loss": 3.3502,
      "theoretical_loss": 4.447916359182623,
      "tokens_seen": 169410560
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047911735205616855,
      "loss": 3.8021,
      "theoretical_loss": 4.4476955529399795,
      "tokens_seen": 169476096
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004791073219658977,
      "loss": 3.4705,
      "theoretical_loss": 4.44747485596341,
      "tokens_seen": 169541632
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004790972918756269,
      "loss": 3.5778,
      "theoretical_loss": 4.447254268156637,
      "tokens_seen": 169607168
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047908726178535604,
      "loss": 3.5024,
      "theoretical_loss": 4.44703378942351,
      "tokens_seen": 169672704
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047907723169508527,
      "loss": 3.5782,
      "theoretical_loss": 4.446813419667994,
      "tokens_seen": 169738240
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047906720160481445,
      "loss": 3.2833,
      "theoretical_loss": 4.446593158794181,
      "tokens_seen": 169803776
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047905717151454363,
      "loss": 3.4609,
      "theoretical_loss": 4.446373006706281,
      "tokens_seen": 169869312
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004790471414242728,
      "loss": 3.3786,
      "theoretical_loss": 4.446152963308625,
      "tokens_seen": 169934848
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.000479037111334002,
      "loss": 3.4031,
      "theoretical_loss": 4.445933028505671,
      "tokens_seen": 170000384
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004790270812437312,
      "loss": 3.4541,
      "theoretical_loss": 4.445713202201988,
      "tokens_seen": 170065920
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004790170511534604,
      "loss": 3.5257,
      "theoretical_loss": 4.445493484302272,
      "tokens_seen": 170131456
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047900702106318954,
      "loss": 3.5678,
      "theoretical_loss": 4.44527387471134,
      "tokens_seen": 170196992
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004789969909729188,
      "loss": 3.5501,
      "theoretical_loss": 4.445054373334125,
      "tokens_seen": 170262528
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004789869608826479,
      "loss": 3.67,
      "theoretical_loss": 4.444834980075683,
      "tokens_seen": 170328064
    },
    {
      "epoch": 0.05,
      "objective/train/advantage_avg": 0.4946400225162506,
      "objective/train/docs_used": 107830,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.4795050621032715,
      "objective/train/original_loss": 3.4795048236846924,
      "objective/train/theoretical_loss": 4.44461569484119,
      "objective/train/tokens_used": 190853600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24474002420902252,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507080554962158,
      "objective/train/weighted_lm_loss": 3.6560089588165283,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0253151655197144,
      "theoretical_loss": 4.44461569484119,
      "tokens_seen": 170393600
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047897693079237714,
      "loss": 3.4345,
      "theoretical_loss": 4.44461569484119,
      "tokens_seen": 170393600
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047896690070210637,
      "loss": 3.203,
      "theoretical_loss": 4.444396517535939,
      "tokens_seen": 170459136
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004789568706118355,
      "loss": 3.419,
      "theoretical_loss": 4.444177448065345,
      "tokens_seen": 170524672
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047894684052156473,
      "loss": 3.637,
      "theoretical_loss": 4.443958486334941,
      "tokens_seen": 170590208
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004789368104312939,
      "loss": 3.0977,
      "theoretical_loss": 4.443739632250379,
      "tokens_seen": 170655744
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004789267803410231,
      "loss": 3.563,
      "theoretical_loss": 4.443520885717431,
      "tokens_seen": 170721280
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004789167502507523,
      "loss": 3.3739,
      "theoretical_loss": 4.443302246641985,
      "tokens_seen": 170786816
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047890672016048146,
      "loss": 3.2141,
      "theoretical_loss": 4.443083714930051,
      "tokens_seen": 170852352
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047889669007021064,
      "loss": 3.4,
      "theoretical_loss": 4.442865290487752,
      "tokens_seen": 170917888
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004788866599799399,
      "loss": 3.5408,
      "theoretical_loss": 4.442646973221336,
      "tokens_seen": 170983424
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.000478876629889669,
      "loss": 3.4155,
      "theoretical_loss": 4.442428763037165,
      "tokens_seen": 171048960
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047886659979939824,
      "loss": 3.3987,
      "theoretical_loss": 4.442210659841717,
      "tokens_seen": 171114496
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047885656970912736,
      "loss": 3.215,
      "theoretical_loss": 4.441992663541589,
      "tokens_seen": 171180032
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004788465396188566,
      "loss": 3.4477,
      "theoretical_loss": 4.4417747740434965,
      "tokens_seen": 171245568
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004788365095285858,
      "loss": 3.5214,
      "theoretical_loss": 4.441556991254271,
      "tokens_seen": 171311104
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047882647943831496,
      "loss": 3.2282,
      "theoretical_loss": 4.441339315080861,
      "tokens_seen": 171376640
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047881644934804414,
      "loss": 3.3898,
      "theoretical_loss": 4.441121745430332,
      "tokens_seen": 171442176
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004788064192577733,
      "loss": 3.4668,
      "theoretical_loss": 4.440904282209865,
      "tokens_seen": 171507712
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004787963891675025,
      "loss": 3.3984,
      "theoretical_loss": 4.440686925326759,
      "tokens_seen": 171573248
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047878635907723174,
      "loss": 3.4297,
      "theoretical_loss": 4.440469674688426,
      "tokens_seen": 171638784
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047877632898696087,
      "loss": 3.2384,
      "theoretical_loss": 4.4402525302023985,
      "tokens_seen": 171704320
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004787662988966901,
      "loss": 3.5769,
      "theoretical_loss": 4.440035491776321,
      "tokens_seen": 171769856
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004787562688064193,
      "loss": 3.5081,
      "theoretical_loss": 4.439818559317954,
      "tokens_seen": 171835392
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047874623871614847,
      "loss": 3.1559,
      "theoretical_loss": 4.439601732735176,
      "tokens_seen": 171900928
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047873620862587765,
      "loss": 3.3807,
      "theoretical_loss": 4.439385011935977,
      "tokens_seen": 171966464
    },
    {
      "epoch": 0.05,
      "objective/train/advantage_avg": 0.4959370791912079,
      "objective/train/docs_used": 108471,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.180680274963379,
      "objective/train/original_loss": 3.180680513381958,
      "objective/train/theoretical_loss": 4.439168396828466,
      "objective/train/tokens_used": 192492000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24603313207626343,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0508445501327515,
      "objective/train/weighted_lm_loss": 3.342397689819336,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.038975715637207,
      "theoretical_loss": 4.439168396828466,
      "tokens_seen": 172032000
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047872617853560683,
      "loss": 3.3567,
      "theoretical_loss": 4.439168396828466,
      "tokens_seen": 172032000
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.000478716148445336,
      "loss": 3.3422,
      "theoretical_loss": 4.438951887320862,
      "tokens_seen": 172097536
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047870611835506524,
      "loss": 3.4383,
      "theoretical_loss": 4.438735483321503,
      "tokens_seen": 172163072
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047869608826479437,
      "loss": 3.1744,
      "theoretical_loss": 4.438519184738839,
      "tokens_seen": 172228608
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004786860581745236,
      "loss": 3.3648,
      "theoretical_loss": 4.438302991481436,
      "tokens_seen": 172294144
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047867602808425273,
      "loss": 3.3488,
      "theoretical_loss": 4.43808690345797,
      "tokens_seen": 172359680
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047866599799398197,
      "loss": 3.1324,
      "theoretical_loss": 4.437870920577237,
      "tokens_seen": 172425216
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047865596790371115,
      "loss": 3.1514,
      "theoretical_loss": 4.437655042748141,
      "tokens_seen": 172490752
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047864593781344033,
      "loss": 3.3295,
      "theoretical_loss": 4.437439269879704,
      "tokens_seen": 172556288
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004786359077231695,
      "loss": 3.4425,
      "theoretical_loss": 4.437223601881057,
      "tokens_seen": 172621824
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047862587763289875,
      "loss": 3.4472,
      "theoretical_loss": 4.437008038661449,
      "tokens_seen": 172687360
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004786158475426279,
      "loss": 3.5052,
      "theoretical_loss": 4.436792580130238,
      "tokens_seen": 172752896
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004786058174523571,
      "loss": 3.225,
      "theoretical_loss": 4.436577226196894,
      "tokens_seen": 172818432
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047859578736208624,
      "loss": 3.3333,
      "theoretical_loss": 4.436361976771005,
      "tokens_seen": 172883968
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047858575727181547,
      "loss": 3.2362,
      "theoretical_loss": 4.436146831762267,
      "tokens_seen": 172949504
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047857572718154465,
      "loss": 3.392,
      "theoretical_loss": 4.435931791080489,
      "tokens_seen": 173015040
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047856569709127383,
      "loss": 3.3698,
      "theoretical_loss": 4.435716854635592,
      "tokens_seen": 173080576
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.000478555667001003,
      "loss": 3.2848,
      "theoretical_loss": 4.43550202233761,
      "tokens_seen": 173146112
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004785456369107322,
      "loss": 3.3584,
      "theoretical_loss": 4.435287294096685,
      "tokens_seen": 173211648
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004785356068204614,
      "loss": 3.6607,
      "theoretical_loss": 4.435072669823078,
      "tokens_seen": 173277184
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004785255767301906,
      "loss": 3.4742,
      "theoretical_loss": 4.434858149427153,
      "tokens_seen": 173342720
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047851554663991974,
      "loss": 3.3386,
      "theoretical_loss": 4.4346437328193895,
      "tokens_seen": 173408256
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.000478505516549649,
      "loss": 3.4489,
      "theoretical_loss": 4.434429419910377,
      "tokens_seen": 173473792
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004784954864593781,
      "loss": 3.2974,
      "theoretical_loss": 4.434215210610818,
      "tokens_seen": 173539328
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047848545636910734,
      "loss": 3.6732,
      "theoretical_loss": 4.43400110483152,
      "tokens_seen": 173604864
    },
    {
      "epoch": 0.05,
      "objective/train/advantage_avg": 0.4947871267795563,
      "objective/train/docs_used": 109645,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.281594753265381,
      "objective/train/original_loss": 3.281594753265381,
      "objective/train/theoretical_loss": 4.433787102483406,
      "objective/train/tokens_used": 194130400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2448694407939911,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507235527038574,
      "objective/train/weighted_lm_loss": 3.4479472637176514,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0413066148757935,
      "theoretical_loss": 4.433787102483406,
      "tokens_seen": 173670400
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004784754262788365,
      "loss": 3.4388,
      "theoretical_loss": 4.433787102483406,
      "tokens_seen": 173670400
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004784653961885657,
      "loss": 3.4421,
      "theoretical_loss": 4.4335732034775095,
      "tokens_seen": 173735936
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004784553660982949,
      "loss": 3.401,
      "theoretical_loss": 4.43335940772497,
      "tokens_seen": 173801472
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004784453360080241,
      "loss": 3.299,
      "theoretical_loss": 4.433145715137038,
      "tokens_seen": 173867008
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047843530591775324,
      "loss": 3.7232,
      "theoretical_loss": 4.4329321256250775,
      "tokens_seen": 173932544
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004784252758274825,
      "loss": 3.4855,
      "theoretical_loss": 4.432718639100558,
      "tokens_seen": 173998080
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004784152457372116,
      "loss": 3.6101,
      "theoretical_loss": 4.43250525547506,
      "tokens_seen": 174063616
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047840521564694084,
      "loss": 3.3999,
      "theoretical_loss": 4.432291974660272,
      "tokens_seen": 174129152
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047839518555667,
      "loss": 3.2669,
      "theoretical_loss": 4.432078796567993,
      "tokens_seen": 174194688
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004783851554663992,
      "loss": 3.4361,
      "theoretical_loss": 4.43186572111013,
      "tokens_seen": 174260224
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004783751253761284,
      "loss": 3.3236,
      "theoretical_loss": 4.431652748198697,
      "tokens_seen": 174325760
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047836509528585756,
      "loss": 3.2636,
      "theoretical_loss": 4.431439877745822,
      "tokens_seen": 174391296
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047835506519558675,
      "loss": 2.9865,
      "theoretical_loss": 4.431227109663735,
      "tokens_seen": 174456832
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.000478345035105316,
      "loss": 3.628,
      "theoretical_loss": 4.431014443864776,
      "tokens_seen": 174522368
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004783350050150451,
      "loss": 3.3055,
      "theoretical_loss": 4.430801880261395,
      "tokens_seen": 174587904
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047832497492477434,
      "loss": 3.4472,
      "theoretical_loss": 4.430589418766148,
      "tokens_seen": 174653440
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047831494483450347,
      "loss": 3.3376,
      "theoretical_loss": 4.430377059291699,
      "tokens_seen": 174718976
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004783049147442327,
      "loss": 3.6125,
      "theoretical_loss": 4.430164801750817,
      "tokens_seen": 174784512
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004782948846539619,
      "loss": 3.3456,
      "theoretical_loss": 4.4299526460563845,
      "tokens_seen": 174850048
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047828485456369107,
      "loss": 3.2753,
      "theoretical_loss": 4.429740592121385,
      "tokens_seen": 174915584
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047827482447342025,
      "loss": 3.4458,
      "theoretical_loss": 4.429528639858911,
      "tokens_seen": 174981120
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004782647943831495,
      "loss": 3.4735,
      "theoretical_loss": 4.429316789182162,
      "tokens_seen": 175046656
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004782547642928786,
      "loss": 3.4697,
      "theoretical_loss": 4.429105040004445,
      "tokens_seen": 175112192
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047824473420260785,
      "loss": 3.3508,
      "theoretical_loss": 4.428893392239171,
      "tokens_seen": 175177728
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.000478234704112337,
      "loss": 3.3988,
      "theoretical_loss": 4.428681845799859,
      "tokens_seen": 175243264
    },
    {
      "epoch": 0.05,
      "objective/train/advantage_avg": 0.49201929569244385,
      "objective/train/docs_used": 110166,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.418017625808716,
      "objective/train/original_loss": 3.4180173873901367,
      "objective/train/theoretical_loss": 4.4284704006001325,
      "objective/train/tokens_used": 195768800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24221739172935486,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0504330396652222,
      "objective/train/weighted_lm_loss": 3.59049391746521,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0338950157165527,
      "theoretical_loss": 4.4284704006001325,
      "tokens_seen": 175308800
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004782246740220662,
      "loss": 3.3964,
      "theoretical_loss": 4.4284704006001325,
      "tokens_seen": 175308800
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047821464393179544,
      "loss": 3.6092,
      "theoretical_loss": 4.428259056553724,
      "tokens_seen": 175374336
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047820461384152457,
      "loss": 3.4793,
      "theoretical_loss": 4.428047813574468,
      "tokens_seen": 175439872
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004781945837512538,
      "loss": 3.3566,
      "theoretical_loss": 4.427836671576307,
      "tokens_seen": 175505408
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047818455366098293,
      "loss": 3.6182,
      "theoretical_loss": 4.427625630473288,
      "tokens_seen": 175570944
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047817452357071217,
      "loss": 3.5185,
      "theoretical_loss": 4.427414690179562,
      "tokens_seen": 175636480
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047816449348044135,
      "loss": 3.19,
      "theoretical_loss": 4.42720385060939,
      "tokens_seen": 175702016
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047815446339017053,
      "loss": 3.3679,
      "theoretical_loss": 4.426993111677129,
      "tokens_seen": 175767552
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004781444332998997,
      "loss": 3.678,
      "theoretical_loss": 4.42678247329725,
      "tokens_seen": 175833088
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047813440320962895,
      "loss": 3.5091,
      "theoretical_loss": 4.426571935384323,
      "tokens_seen": 175898624
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004781243731193581,
      "loss": 3.2368,
      "theoretical_loss": 4.426361497853024,
      "tokens_seen": 175964160
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004781143430290873,
      "loss": 3.4174,
      "theoretical_loss": 4.426151160618133,
      "tokens_seen": 176029696
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047810431293881644,
      "loss": 3.3621,
      "theoretical_loss": 4.425940923594533,
      "tokens_seen": 176095232
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047809428284854567,
      "loss": 3.3926,
      "theoretical_loss": 4.4257307866972155,
      "tokens_seen": 176160768
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047808425275827485,
      "loss": 3.4274,
      "theoretical_loss": 4.425520749841268,
      "tokens_seen": 176226304
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047807422266800403,
      "loss": 3.4326,
      "theoretical_loss": 4.425310812941888,
      "tokens_seen": 176291840
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004780641925777332,
      "loss": 3.1517,
      "theoretical_loss": 4.425100975914372,
      "tokens_seen": 176357376
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004780541624874624,
      "loss": 3.4734,
      "theoretical_loss": 4.424891238674125,
      "tokens_seen": 176422912
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004780441323971916,
      "loss": 3.2721,
      "theoretical_loss": 4.424681601136649,
      "tokens_seen": 176488448
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004780341023069208,
      "loss": 3.2057,
      "theoretical_loss": 4.4244720632175545,
      "tokens_seen": 176553984
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047802407221664994,
      "loss": 3.3691,
      "theoretical_loss": 4.42426262483255,
      "tokens_seen": 176619520
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004780140421263792,
      "loss": 3.5622,
      "theoretical_loss": 4.424053285897449,
      "tokens_seen": 176685056
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004780040120361083,
      "loss": 3.5559,
      "theoretical_loss": 4.423844046328166,
      "tokens_seen": 176750592
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047799398194583754,
      "loss": 3.2575,
      "theoretical_loss": 4.4236349060407205,
      "tokens_seen": 176816128
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004779839518555667,
      "loss": 3.5452,
      "theoretical_loss": 4.423425864951231,
      "tokens_seen": 176881664
    },
    {
      "epoch": 0.05,
      "objective/train/advantage_avg": 0.49467232823371887,
      "objective/train/docs_used": 110915,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.211423635482788,
      "objective/train/original_loss": 3.211423635482788,
      "objective/train/theoretical_loss": 4.42321692297592,
      "objective/train/tokens_used": 197407200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24474696815013885,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507113933563232,
      "objective/train/weighted_lm_loss": 3.374145269393921,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.047196388244629,
      "theoretical_loss": 4.42321692297592,
      "tokens_seen": 176947200
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004779739217652959,
      "loss": 3.3183,
      "theoretical_loss": 4.42321692297592,
      "tokens_seen": 176947200
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004779638916750251,
      "loss": 3.2714,
      "theoretical_loss": 4.42300808003111,
      "tokens_seen": 177012736
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004779538615847543,
      "loss": 3.4906,
      "theoretical_loss": 4.422799336033226,
      "tokens_seen": 177078272
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047794383149448344,
      "loss": 3.4468,
      "theoretical_loss": 4.4225906908987955,
      "tokens_seen": 177143808
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004779338014042127,
      "loss": 3.4765,
      "theoretical_loss": 4.422382144544446,
      "tokens_seen": 177209344
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004779237713139418,
      "loss": 3.3183,
      "theoretical_loss": 4.4221736968869045,
      "tokens_seen": 177274880
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047791374122367104,
      "loss": 3.3391,
      "theoretical_loss": 4.421965347843003,
      "tokens_seen": 177340416
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004779037111334002,
      "loss": 3.2415,
      "theoretical_loss": 4.421757097329672,
      "tokens_seen": 177405952
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004778936810431294,
      "loss": 3.2882,
      "theoretical_loss": 4.421548945263941,
      "tokens_seen": 177471488
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004778836509528586,
      "loss": 3.5512,
      "theoretical_loss": 4.421340891562943,
      "tokens_seen": 177537024
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047787362086258776,
      "loss": 3.2885,
      "theoretical_loss": 4.421132936143909,
      "tokens_seen": 177602560
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047786359077231695,
      "loss": 3.2393,
      "theoretical_loss": 4.420925078924172,
      "tokens_seen": 177668096
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004778535606820462,
      "loss": 3.5259,
      "theoretical_loss": 4.420717319821163,
      "tokens_seen": 177733632
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004778435305917753,
      "loss": 3.4351,
      "theoretical_loss": 4.420509658752415,
      "tokens_seen": 177799168
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047783350050150454,
      "loss": 3.1694,
      "theoretical_loss": 4.420302095635559,
      "tokens_seen": 177864704
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047782347041123367,
      "loss": 3.4997,
      "theoretical_loss": 4.420094630388327,
      "tokens_seen": 177930240
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004778134403209629,
      "loss": 3.2944,
      "theoretical_loss": 4.419887262928549,
      "tokens_seen": 177995776
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004778034102306921,
      "loss": 3.3215,
      "theoretical_loss": 4.419679993174155,
      "tokens_seen": 178061312
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047779338014042127,
      "loss": 3.226,
      "theoretical_loss": 4.419472821043172,
      "tokens_seen": 178126848
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047778335005015045,
      "loss": 3.4448,
      "theoretical_loss": 4.41926574645373,
      "tokens_seen": 178192384
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004777733199598797,
      "loss": 3.6291,
      "theoretical_loss": 4.419058769324055,
      "tokens_seen": 178257920
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004777632898696088,
      "loss": 3.5838,
      "theoretical_loss": 4.41885188957247,
      "tokens_seen": 178323456
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047775325977933805,
      "loss": 3.5965,
      "theoretical_loss": 4.418645107117401,
      "tokens_seen": 178388992
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004777432296890672,
      "loss": 3.3358,
      "theoretical_loss": 4.418438421877369,
      "tokens_seen": 178454528
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004777331995987964,
      "loss": 3.4101,
      "theoretical_loss": 4.418231833770992,
      "tokens_seen": 178520064
    },
    {
      "epoch": 0.05,
      "objective/train/advantage_avg": 0.4948647916316986,
      "objective/train/docs_used": 112413,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.0979092121124268,
      "objective/train/original_loss": 3.0979092121124268,
      "objective/train/theoretical_loss": 4.418025342716991,
      "objective/train/tokens_used": 199045600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.244956836104393,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507316589355469,
      "objective/train/weighted_lm_loss": 3.2549819946289062,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0253151655197144,
      "theoretical_loss": 4.418025342716991,
      "tokens_seen": 178585600
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004777231695085256,
      "loss": 3.2454,
      "theoretical_loss": 4.418025342716991,
      "tokens_seen": 178585600
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047771313941825477,
      "loss": 3.1971,
      "theoretical_loss": 4.417818948634178,
      "tokens_seen": 178651136
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047770310932798395,
      "loss": 3.5005,
      "theoretical_loss": 4.41761265144147,
      "tokens_seen": 178716672
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047769307923771313,
      "loss": 3.3436,
      "theoretical_loss": 4.417406451057875,
      "tokens_seen": 178782208
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004776830491474423,
      "loss": 3.1508,
      "theoretical_loss": 4.417200347402503,
      "tokens_seen": 178847744
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047767301905717155,
      "loss": 3.2575,
      "theoretical_loss": 4.416994340394557,
      "tokens_seen": 178913280
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004776629889669007,
      "loss": 3.3917,
      "theoretical_loss": 4.416788429953341,
      "tokens_seen": 178978816
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004776529588766299,
      "loss": 3.5485,
      "theoretical_loss": 4.416582615998255,
      "tokens_seen": 179044352
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047764292878635904,
      "loss": 3.2773,
      "theoretical_loss": 4.416376898448793,
      "tokens_seen": 179109888
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004776328986960883,
      "loss": 3.255,
      "theoretical_loss": 4.416171277224549,
      "tokens_seen": 179175424
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047762286860581746,
      "loss": 3.4578,
      "theoretical_loss": 4.415965752245213,
      "tokens_seen": 179240960
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047761283851554664,
      "loss": 3.3194,
      "theoretical_loss": 4.415760323430568,
      "tokens_seen": 179306496
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004776028084252758,
      "loss": 3.2296,
      "theoretical_loss": 4.415554990700496,
      "tokens_seen": 179372032
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047759277833500505,
      "loss": 3.1707,
      "theoretical_loss": 4.415349753974977,
      "tokens_seen": 179437568
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004775827482447342,
      "loss": 3.3703,
      "theoretical_loss": 4.41514461317408,
      "tokens_seen": 179503104
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004775727181544634,
      "loss": 3.3497,
      "theoretical_loss": 4.414939568217978,
      "tokens_seen": 179568640
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047756268806419254,
      "loss": 3.5743,
      "theoretical_loss": 4.414734619026934,
      "tokens_seen": 179634176
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004775526579739218,
      "loss": 3.3655,
      "theoretical_loss": 4.4145297655213085,
      "tokens_seen": 179699712
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047754262788365096,
      "loss": 3.4367,
      "theoretical_loss": 4.414325007621555,
      "tokens_seen": 179765248
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047753259779338014,
      "loss": 3.3675,
      "theoretical_loss": 4.414120345248225,
      "tokens_seen": 179830784
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004775225677031093,
      "loss": 3.2201,
      "theoretical_loss": 4.4139157783219645,
      "tokens_seen": 179896320
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004775125376128385,
      "loss": 3.4657,
      "theoretical_loss": 4.4137113067635125,
      "tokens_seen": 179961856
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004775025075225677,
      "loss": 3.4253,
      "theoretical_loss": 4.413506930493703,
      "tokens_seen": 180027392
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004774924774322969,
      "loss": 3.3257,
      "theoretical_loss": 4.413302649433467,
      "tokens_seen": 180092928
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047748244734202605,
      "loss": 3.3707,
      "theoretical_loss": 4.413098463503827,
      "tokens_seen": 180158464
    },
    {
      "epoch": 0.05,
      "objective/train/advantage_avg": 0.4950984716415405,
      "objective/train/docs_used": 112982,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.1033313274383545,
      "objective/train/original_loss": 3.1033318042755127,
      "objective/train/theoretical_loss": 4.412894372625901,
      "objective/train/tokens_used": 200684000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2451845407485962,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507563352584839,
      "objective/train/weighted_lm_loss": 3.2606964111328125,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 4.412894372625901,
      "tokens_seen": 180224000
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004774724172517553,
      "loss": 3.4106,
      "theoretical_loss": 4.412894372625901,
      "tokens_seen": 180224000
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004774623871614845,
      "loss": 3.6212,
      "theoretical_loss": 4.412690376720901,
      "tokens_seen": 180289536
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047745235707121364,
      "loss": 3.5214,
      "theoretical_loss": 4.412486475710132,
      "tokens_seen": 180355072
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004774423269809429,
      "loss": 3.2988,
      "theoretical_loss": 4.412282669514994,
      "tokens_seen": 180420608
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.000477432296890672,
      "loss": 3.5009,
      "theoretical_loss": 4.412078958056981,
      "tokens_seen": 180486144
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047742226680040124,
      "loss": 3.1817,
      "theoretical_loss": 4.411875341257677,
      "tokens_seen": 180551680
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004774122367101304,
      "loss": 3.6172,
      "theoretical_loss": 4.411671819038765,
      "tokens_seen": 180617216
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004774022066198596,
      "loss": 3.2969,
      "theoretical_loss": 4.411468391322017,
      "tokens_seen": 180682752
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004773921765295888,
      "loss": 3.1682,
      "theoretical_loss": 4.411265058029299,
      "tokens_seen": 180748288
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047738214643931797,
      "loss": 3.5001,
      "theoretical_loss": 4.411061819082569,
      "tokens_seen": 180813824
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047737211634904715,
      "loss": 3.4226,
      "theoretical_loss": 4.4108586744038805,
      "tokens_seen": 180879360
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004773620862587764,
      "loss": 3.4242,
      "theoretical_loss": 4.410655623915377,
      "tokens_seen": 180944896
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004773520561685055,
      "loss": 3.3079,
      "theoretical_loss": 4.410452667539296,
      "tokens_seen": 181010432
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047734202607823474,
      "loss": 3.2552,
      "theoretical_loss": 4.410249805197967,
      "tokens_seen": 181075968
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047733199598796387,
      "loss": 3.5285,
      "theoretical_loss": 4.410047036813811,
      "tokens_seen": 181141504
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004773219658976931,
      "loss": 3.1154,
      "theoretical_loss": 4.409844362309341,
      "tokens_seen": 181207040
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004773119358074223,
      "loss": 3.2822,
      "theoretical_loss": 4.409641781607164,
      "tokens_seen": 181272576
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047730190571715147,
      "loss": 3.2936,
      "theoretical_loss": 4.4094392946299745,
      "tokens_seen": 181338112
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.00047729187562688065,
      "loss": 3.311,
      "theoretical_loss": 4.409236901300563,
      "tokens_seen": 181403648
    },
    {
      "epoch": 0.05,
      "learning_rate": 0.0004772818455366099,
      "loss": 3.4846,
      "theoretical_loss": 4.40903460154181,
      "tokens_seen": 181469184
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.000477271815446339,
      "loss": 3.2595,
      "theoretical_loss": 4.408832395276686,
      "tokens_seen": 181534720
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047726178535606825,
      "loss": 3.4071,
      "theoretical_loss": 4.408630282428255,
      "tokens_seen": 181600256
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004772517552657974,
      "loss": 3.1771,
      "theoretical_loss": 4.408428262919669,
      "tokens_seen": 181665792
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004772417251755266,
      "loss": 3.333,
      "theoretical_loss": 4.408226336674173,
      "tokens_seen": 181731328
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004772316950852558,
      "loss": 3.437,
      "theoretical_loss": 4.4080245036151045,
      "tokens_seen": 181796864
    },
    {
      "epoch": 0.06,
      "objective/train/advantage_avg": 0.49566391110420227,
      "objective/train/docs_used": 114216,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.073647975921631,
      "objective/train/original_loss": 3.073647975921631,
      "objective/train/theoretical_loss": 4.407822763665887,
      "objective/train/tokens_used": 202322400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2457331418991089,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0508155822753906,
      "objective/train/weighted_lm_loss": 3.2298130989074707,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0396549701690674,
      "theoretical_loss": 4.407822763665887,
      "tokens_seen": 181862400
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047722166499498497,
      "loss": 3.24,
      "theoretical_loss": 4.407822763665887,
      "tokens_seen": 181862400
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047721163490471415,
      "loss": 3.4355,
      "theoretical_loss": 4.407621116750039,
      "tokens_seen": 181927936
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047720160481444333,
      "loss": 3.4819,
      "theoretical_loss": 4.4074195627911665,
      "tokens_seen": 181993472
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004771915747241725,
      "loss": 3.4768,
      "theoretical_loss": 4.407218101712966,
      "tokens_seen": 182059008
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047718154463390175,
      "loss": 3.437,
      "theoretical_loss": 4.407016733439226,
      "tokens_seen": 182124544
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004771715145436309,
      "loss": 3.2575,
      "theoretical_loss": 4.406815457893824,
      "tokens_seen": 182190080
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004771614844533601,
      "loss": 3.3491,
      "theoretical_loss": 4.4066142750007256,
      "tokens_seen": 182255616
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047715145436308924,
      "loss": 3.2039,
      "theoretical_loss": 4.406413184683987,
      "tokens_seen": 182321152
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004771414242728185,
      "loss": 3.2685,
      "theoretical_loss": 4.406212186867756,
      "tokens_seen": 182386688
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047713139418254766,
      "loss": 3.2634,
      "theoretical_loss": 4.406011281476267,
      "tokens_seen": 182452224
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047712136409227684,
      "loss": 3.2118,
      "theoretical_loss": 4.405810468433844,
      "tokens_seen": 182517760
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.000477111334002006,
      "loss": 3.3822,
      "theoretical_loss": 4.4056097476649025,
      "tokens_seen": 182583296
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047710130391173525,
      "loss": 3.288,
      "theoretical_loss": 4.405409119093944,
      "tokens_seen": 182648832
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004770912738214644,
      "loss": 3.2546,
      "theoretical_loss": 4.405208582645562,
      "tokens_seen": 182714368
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004770812437311936,
      "loss": 3.6179,
      "theoretical_loss": 4.405008138244434,
      "tokens_seen": 182779904
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047707121364092274,
      "loss": 3.5899,
      "theoretical_loss": 4.404807785815331,
      "tokens_seen": 182845440
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.000477061183550652,
      "loss": 3.5229,
      "theoretical_loss": 4.40460752528311,
      "tokens_seen": 182910976
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047705115346038116,
      "loss": 3.5056,
      "theoretical_loss": 4.404407356572715,
      "tokens_seen": 182976512
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047704112337011034,
      "loss": 3.2369,
      "theoretical_loss": 4.404207279609182,
      "tokens_seen": 183042048
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004770310932798395,
      "loss": 3.4716,
      "theoretical_loss": 4.404007294317632,
      "tokens_seen": 183107584
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004770210631895687,
      "loss": 3.3636,
      "theoretical_loss": 4.403807400623273,
      "tokens_seen": 183173120
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004770110330992979,
      "loss": 3.3666,
      "theoretical_loss": 4.403607598451404,
      "tokens_seen": 183238656
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004770010030090271,
      "loss": 3.3698,
      "theoretical_loss": 4.40340788772741,
      "tokens_seen": 183304192
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047699097291875625,
      "loss": 3.4237,
      "theoretical_loss": 4.403208268376764,
      "tokens_seen": 183369728
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004769809428284855,
      "loss": 3.5203,
      "theoretical_loss": 4.403008740325024,
      "tokens_seen": 183435264
    },
    {
      "epoch": 0.06,
      "objective/train/advantage_avg": 0.4938001334667206,
      "objective/train/docs_used": 114775,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.201855421066284,
      "objective/train/original_loss": 3.201855421066284,
      "objective/train/theoretical_loss": 4.40280930349784,
      "objective/train/tokens_used": 203960800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2439144253730774,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506199598312378,
      "objective/train/weighted_lm_loss": 3.363767385482788,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0396549701690674,
      "theoretical_loss": 4.40280930349784,
      "tokens_seen": 183500800
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047697091273821466,
      "loss": 3.4764,
      "theoretical_loss": 4.40280930349784,
      "tokens_seen": 183500800
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047696088264794384,
      "loss": 3.1116,
      "theoretical_loss": 4.4026099578209426,
      "tokens_seen": 183566336
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.000476950852557673,
      "loss": 3.1319,
      "theoretical_loss": 4.402410703220154,
      "tokens_seen": 183631872
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004769408224674022,
      "loss": 3.5832,
      "theoretical_loss": 4.4022115396213835,
      "tokens_seen": 183697408
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004769307923771314,
      "loss": 3.6015,
      "theoretical_loss": 4.402012466950625,
      "tokens_seen": 183762944
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004769207622868606,
      "loss": 3.3855,
      "theoretical_loss": 4.4018134851339585,
      "tokens_seen": 183828480
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047691073219658975,
      "loss": 3.5521,
      "theoretical_loss": 4.401614594097554,
      "tokens_seen": 183894016
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.000476900702106319,
      "loss": 3.5072,
      "theoretical_loss": 4.401415793767662,
      "tokens_seen": 183959552
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004768906720160481,
      "loss": 3.3454,
      "theoretical_loss": 4.401217084070625,
      "tokens_seen": 184025088
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047688064192577735,
      "loss": 3.4143,
      "theoretical_loss": 4.401018464932868,
      "tokens_seen": 184090624
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047687061183550653,
      "loss": 3.2606,
      "theoretical_loss": 4.400819936280903,
      "tokens_seen": 184156160
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004768605817452357,
      "loss": 3.3518,
      "theoretical_loss": 4.400621498041327,
      "tokens_seen": 184221696
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004768505516549649,
      "loss": 3.2543,
      "theoretical_loss": 4.400423150140825,
      "tokens_seen": 184287232
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047684052156469407,
      "loss": 3.3838,
      "theoretical_loss": 4.400224892506165,
      "tokens_seen": 184352768
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047683049147442325,
      "loss": 3.5664,
      "theoretical_loss": 4.400026725064199,
      "tokens_seen": 184418304
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004768204613841525,
      "loss": 3.5153,
      "theoretical_loss": 4.39982864774187,
      "tokens_seen": 184483840
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004768104312938816,
      "loss": 3.4489,
      "theoretical_loss": 4.3996306604662,
      "tokens_seen": 184549376
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047680040120361085,
      "loss": 3.5846,
      "theoretical_loss": 4.399432763164299,
      "tokens_seen": 184614912
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047679037111334003,
      "loss": 3.4092,
      "theoretical_loss": 4.399234955763362,
      "tokens_seen": 184680448
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004767803410230692,
      "loss": 3.2862,
      "theoretical_loss": 4.399037238190667,
      "tokens_seen": 184745984
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004767703109327984,
      "loss": 3.3632,
      "theoretical_loss": 4.398839610373579,
      "tokens_seen": 184811520
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004767602808425276,
      "loss": 3.302,
      "theoretical_loss": 4.398642072239545,
      "tokens_seen": 184877056
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047675025075225676,
      "loss": 3.413,
      "theoretical_loss": 4.398444623716097,
      "tokens_seen": 184942592
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.000476740220661986,
      "loss": 3.4363,
      "theoretical_loss": 4.398247264730853,
      "tokens_seen": 185008128
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004767301905717151,
      "loss": 3.5951,
      "theoretical_loss": 4.398049995211513,
      "tokens_seen": 185073664
    },
    {
      "epoch": 0.06,
      "objective/train/advantage_avg": 0.49490630626678467,
      "objective/train/docs_used": 115939,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.4280598163604736,
      "objective/train/original_loss": 3.4280598163604736,
      "objective/train/theoretical_loss": 4.397852815085862,
      "objective/train/tokens_used": 205599200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2449771761894226,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507360696792603,
      "objective/train/weighted_lm_loss": 3.6018779277801514,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0457526445388794,
      "theoretical_loss": 4.397852815085862,
      "tokens_seen": 185139200
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047672016048144435,
      "loss": 3.2268,
      "theoretical_loss": 4.397852815085862,
      "tokens_seen": 185139200
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047671013039117353,
      "loss": 3.387,
      "theoretical_loss": 4.397655724281767,
      "tokens_seen": 185204736
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004767001003009027,
      "loss": 3.5387,
      "theoretical_loss": 4.397458722727183,
      "tokens_seen": 185270272
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047669007021063195,
      "loss": 3.393,
      "theoretical_loss": 4.397261810350145,
      "tokens_seen": 185335808
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004766800401203611,
      "loss": 3.2963,
      "theoretical_loss": 4.397064987078769,
      "tokens_seen": 185401344
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004766700100300903,
      "loss": 3.1234,
      "theoretical_loss": 4.396868252841262,
      "tokens_seen": 185466880
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047665997993981944,
      "loss": 3.2432,
      "theoretical_loss": 4.396671607565907,
      "tokens_seen": 185532416
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004766499498495487,
      "loss": 3.4353,
      "theoretical_loss": 4.396475051181074,
      "tokens_seen": 185597952
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047663991975927786,
      "loss": 3.3209,
      "theoretical_loss": 4.3962785836152145,
      "tokens_seen": 185663488
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047662988966900704,
      "loss": 3.333,
      "theoretical_loss": 4.396082204796863,
      "tokens_seen": 185729024
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004766198595787362,
      "loss": 3.5329,
      "theoretical_loss": 4.395885914654636,
      "tokens_seen": 185794560
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047660982948846545,
      "loss": 3.4285,
      "theoretical_loss": 4.395689713117235,
      "tokens_seen": 185860096
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004765997993981946,
      "loss": 3.2443,
      "theoretical_loss": 4.395493600113442,
      "tokens_seen": 185925632
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004765897693079238,
      "loss": 3.3204,
      "theoretical_loss": 4.39529757557212,
      "tokens_seen": 185991168
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047657973921765294,
      "loss": 3.6103,
      "theoretical_loss": 4.395101639422217,
      "tokens_seen": 186056704
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004765697091273822,
      "loss": 3.2795,
      "theoretical_loss": 4.3949057915927625,
      "tokens_seen": 186122240
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047655967903711136,
      "loss": 3.5055,
      "theoretical_loss": 4.394710032012865,
      "tokens_seen": 186187776
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047654964894684054,
      "loss": 3.3572,
      "theoretical_loss": 4.39451436061172,
      "tokens_seen": 186253312
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004765396188565697,
      "loss": 3.3234,
      "theoretical_loss": 4.3943187773186,
      "tokens_seen": 186318848
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004765295887662989,
      "loss": 3.1564,
      "theoretical_loss": 4.3941232820628615,
      "tokens_seen": 186384384
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004765195586760281,
      "loss": 3.1574,
      "theoretical_loss": 4.393927874773942,
      "tokens_seen": 186449920
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004765095285857573,
      "loss": 3.0581,
      "theoretical_loss": 4.3937325553813595,
      "tokens_seen": 186515456
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047649949849548645,
      "loss": 3.4308,
      "theoretical_loss": 4.393537323814715,
      "tokens_seen": 186580992
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004764894684052157,
      "loss": 3.3811,
      "theoretical_loss": 4.393342180003689,
      "tokens_seen": 186646528
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047647943831494486,
      "loss": 2.9604,
      "theoretical_loss": 4.393147123878043,
      "tokens_seen": 186712064
    },
    {
      "epoch": 0.06,
      "objective/train/advantage_avg": 0.49547553062438965,
      "objective/train/docs_used": 116562,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.2502243518829346,
      "objective/train/original_loss": 3.2502248287200928,
      "objective/train/theoretical_loss": 4.392952155367621,
      "objective/train/tokens_used": 207237600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24558210372924805,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507959127426147,
      "objective/train/weighted_lm_loss": 3.4151992797851562,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0408108234405518,
      "theoretical_loss": 4.392952155367621,
      "tokens_seen": 186777600
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047646940822467404,
      "loss": 3.42,
      "theoretical_loss": 4.392952155367621,
      "tokens_seen": 186777600
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004764593781344032,
      "loss": 3.4218,
      "theoretical_loss": 4.392757274402346,
      "tokens_seen": 186843136
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004764493480441324,
      "loss": 3.4846,
      "theoretical_loss": 4.392562480912222,
      "tokens_seen": 186908672
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004764393179538616,
      "loss": 3.4077,
      "theoretical_loss": 4.392367774827333,
      "tokens_seen": 186974208
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004764292878635908,
      "loss": 3.2395,
      "theoretical_loss": 4.392173156077845,
      "tokens_seen": 187039744
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047641925777331995,
      "loss": 3.5905,
      "theoretical_loss": 4.391978624594003,
      "tokens_seen": 187105280
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004764092276830492,
      "loss": 3.3666,
      "theoretical_loss": 4.391784180306133,
      "tokens_seen": 187170816
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004763991975927783,
      "loss": 3.3127,
      "theoretical_loss": 4.391589823144638,
      "tokens_seen": 187236352
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047638916750250755,
      "loss": 3.1385,
      "theoretical_loss": 4.391395553040007,
      "tokens_seen": 187301888
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047637913741223673,
      "loss": 3.3103,
      "theoretical_loss": 4.391201369922801,
      "tokens_seen": 187367424
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004763691073219659,
      "loss": 3.5164,
      "theoretical_loss": 4.391007273723668,
      "tokens_seen": 187432960
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004763590772316951,
      "loss": 3.5073,
      "theoretical_loss": 4.39081326437333,
      "tokens_seen": 187498496
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047634904714142427,
      "loss": 3.4578,
      "theoretical_loss": 4.390619341802592,
      "tokens_seen": 187564032
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047633901705115345,
      "loss": 3.2434,
      "theoretical_loss": 4.390425505942336,
      "tokens_seen": 187629568
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004763289869608827,
      "loss": 3.3691,
      "theoretical_loss": 4.390231756723523,
      "tokens_seen": 187695104
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004763189568706118,
      "loss": 3.2496,
      "theoretical_loss": 4.390038094077196,
      "tokens_seen": 187760640
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047630892678034105,
      "loss": 3.504,
      "theoretical_loss": 4.389844517934475,
      "tokens_seen": 187826176
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047629889669007023,
      "loss": 3.4626,
      "theoretical_loss": 4.389651028226558,
      "tokens_seen": 187891712
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004762888665997994,
      "loss": 3.3644,
      "theoretical_loss": 4.389457624884722,
      "tokens_seen": 187957248
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004762788365095286,
      "loss": 3.5134,
      "theoretical_loss": 4.389264307840324,
      "tokens_seen": 188022784
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004762688064192578,
      "loss": 3.3614,
      "theoretical_loss": 4.389071077024798,
      "tokens_seen": 188088320
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047625877632898696,
      "loss": 3.4139,
      "theoretical_loss": 4.388877932369657,
      "tokens_seen": 188153856
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004762487462387162,
      "loss": 3.22,
      "theoretical_loss": 4.3886848738064925,
      "tokens_seen": 188219392
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004762387161484453,
      "loss": 3.2898,
      "theoretical_loss": 4.388491901266973,
      "tokens_seen": 188284928
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047622868605817455,
      "loss": 3.3795,
      "theoretical_loss": 4.388299014682847,
      "tokens_seen": 188350464
    },
    {
      "epoch": 0.06,
      "objective/train/advantage_avg": 0.49448370933532715,
      "objective/train/docs_used": 117858,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.1107797622680664,
      "objective/train/original_loss": 3.1107795238494873,
      "objective/train/theoretical_loss": 4.388106213985938,
      "objective/train/tokens_used": 208876000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24456298351287842,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506914854049683,
      "objective/train/weighted_lm_loss": 3.2684762477874756,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0338950157165527,
      "theoretical_loss": 4.388106213985938,
      "tokens_seen": 188416000
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004762186559679037,
      "loss": 3.1496,
      "theoretical_loss": 4.388106213985938,
      "tokens_seen": 188416000
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004762086258776329,
      "loss": 3.1387,
      "theoretical_loss": 4.3879134991081505,
      "tokens_seen": 188481536
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004761985957873621,
      "loss": 3.125,
      "theoretical_loss": 4.387720869981464,
      "tokens_seen": 188547072
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004761885656970913,
      "loss": 3.504,
      "theoretical_loss": 4.387528326537936,
      "tokens_seen": 188612608
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047617853560682046,
      "loss": 3.2213,
      "theoretical_loss": 4.3873358687097035,
      "tokens_seen": 188678144
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047616850551654964,
      "loss": 3.3343,
      "theoretical_loss": 4.387143496428978,
      "tokens_seen": 188743680
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004761584754262788,
      "loss": 3.4957,
      "theoretical_loss": 4.38695120962805,
      "tokens_seen": 188809216
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047614844533600806,
      "loss": 3.3756,
      "theoretical_loss": 4.386759008239285,
      "tokens_seen": 188874752
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004761384152457372,
      "loss": 3.3262,
      "theoretical_loss": 4.38656689219513,
      "tokens_seen": 188940288
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004761283851554664,
      "loss": 3.3286,
      "theoretical_loss": 4.386374861428104,
      "tokens_seen": 189005824
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004761183550651956,
      "loss": 3.4415,
      "theoretical_loss": 4.386182915870804,
      "tokens_seen": 189071360
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004761083249749248,
      "loss": 3.4304,
      "theoretical_loss": 4.3859910554559045,
      "tokens_seen": 189136896
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047609829488465396,
      "loss": 3.5458,
      "theoretical_loss": 4.385799280116157,
      "tokens_seen": 189202432
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047608826479438314,
      "loss": 3.3824,
      "theoretical_loss": 4.385607589784388,
      "tokens_seen": 189267968
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004760782347041123,
      "loss": 3.4198,
      "theoretical_loss": 4.3854159843935,
      "tokens_seen": 189333504
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047606820461384156,
      "loss": 3.1586,
      "theoretical_loss": 4.385224463876474,
      "tokens_seen": 189399040
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004760581745235707,
      "loss": 3.4116,
      "theoretical_loss": 4.385033028166365,
      "tokens_seen": 189464576
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004760481444332999,
      "loss": 3.3576,
      "theoretical_loss": 4.384841677196304,
      "tokens_seen": 189530112
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047603811434302905,
      "loss": 3.3158,
      "theoretical_loss": 4.384650410899498,
      "tokens_seen": 189595648
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004760280842527583,
      "loss": 3.6079,
      "theoretical_loss": 4.384459229209231,
      "tokens_seen": 189661184
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047601805416248746,
      "loss": 3.4535,
      "theoretical_loss": 4.38426813205886,
      "tokens_seen": 189726720
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047600802407221665,
      "loss": 3.466,
      "theoretical_loss": 4.384077119381821,
      "tokens_seen": 189792256
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004759979939819458,
      "loss": 3.1677,
      "theoretical_loss": 4.383886191111623,
      "tokens_seen": 189857792
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047598796389167506,
      "loss": 3.4814,
      "theoretical_loss": 4.383695347181849,
      "tokens_seen": 189923328
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004759779338014042,
      "loss": 3.3499,
      "theoretical_loss": 4.383504587526161,
      "tokens_seen": 189988864
    },
    {
      "epoch": 0.06,
      "objective/train/advantage_avg": 0.4934444725513458,
      "objective/train/docs_used": 118609,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.1113767623901367,
      "objective/train/original_loss": 3.1113765239715576,
      "objective/train/theoretical_loss": 4.383313912078293,
      "objective/train/tokens_used": 210514400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24356885254383087,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505824089050293,
      "objective/train/weighted_lm_loss": 3.2688117027282715,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0408108234405518,
      "theoretical_loss": 4.383313912078293,
      "tokens_seen": 190054400
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004759679037111334,
      "loss": 3.556,
      "theoretical_loss": 4.383313912078293,
      "tokens_seen": 190054400
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004759578736208626,
      "loss": 3.0268,
      "theoretical_loss": 4.383123320772055,
      "tokens_seen": 190119936
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004759478435305918,
      "loss": 3.6297,
      "theoretical_loss": 4.382932813541331,
      "tokens_seen": 190185472
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.000475937813440321,
      "loss": 3.2196,
      "theoretical_loss": 4.382742390320081,
      "tokens_seen": 190251008
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047592778335005015,
      "loss": 3.379,
      "theoretical_loss": 4.382552051042339,
      "tokens_seen": 190316544
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004759177532597794,
      "loss": 3.3182,
      "theoretical_loss": 4.382361795642213,
      "tokens_seen": 190382080
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004759077231695085,
      "loss": 3.2449,
      "theoretical_loss": 4.3821716240538855,
      "tokens_seen": 190447616
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047589769307923775,
      "loss": 3.2045,
      "theoretical_loss": 4.381981536211615,
      "tokens_seen": 190513152
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047588766298896693,
      "loss": 3.3895,
      "theoretical_loss": 4.381791532049733,
      "tokens_seen": 190578688
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004758776328986961,
      "loss": 3.5757,
      "theoretical_loss": 4.381601611502643,
      "tokens_seen": 190644224
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004758676028084253,
      "loss": 3.2043,
      "theoretical_loss": 4.381411774504825,
      "tokens_seen": 190709760
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047585757271815447,
      "loss": 3.5252,
      "theoretical_loss": 4.381222020990832,
      "tokens_seen": 190775296
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047584754262788365,
      "loss": 3.3334,
      "theoretical_loss": 4.381032350895292,
      "tokens_seen": 190840832
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004758375125376129,
      "loss": 3.3918,
      "theoretical_loss": 4.380842764152904,
      "tokens_seen": 190906368
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.000475827482447342,
      "loss": 3.4295,
      "theoretical_loss": 4.380653260698443,
      "tokens_seen": 190971904
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047581745235707125,
      "loss": 3.1268,
      "theoretical_loss": 4.380463840466755,
      "tokens_seen": 191037440
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047580742226680043,
      "loss": 3.3916,
      "theoretical_loss": 4.380274503392762,
      "tokens_seen": 191102976
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004757973921765296,
      "loss": 3.7235,
      "theoretical_loss": 4.380085249411458,
      "tokens_seen": 191168512
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004757873620862588,
      "loss": 3.3881,
      "theoretical_loss": 4.379896078457908,
      "tokens_seen": 191234048
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.000475777331995988,
      "loss": 3.3506,
      "theoretical_loss": 4.379706990467254,
      "tokens_seen": 191299584
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047576730190571716,
      "loss": 3.4007,
      "theoretical_loss": 4.379517985374708,
      "tokens_seen": 191365120
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004757572718154464,
      "loss": 3.1795,
      "theoretical_loss": 4.379329063115555,
      "tokens_seen": 191430656
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004757472417251755,
      "loss": 3.156,
      "theoretical_loss": 4.3791402236251535,
      "tokens_seen": 191496192
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047573721163490475,
      "loss": 3.5919,
      "theoretical_loss": 4.378951466838935,
      "tokens_seen": 191561728
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004757271815446339,
      "loss": 3.3665,
      "theoretical_loss": 4.378762792692402,
      "tokens_seen": 191627264
    },
    {
      "epoch": 0.06,
      "objective/train/advantage_avg": 0.49388450384140015,
      "objective/train/docs_used": 119339,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.186603546142578,
      "objective/train/original_loss": 3.186603546142578,
      "objective/train/theoretical_loss": 4.37857420112113,
      "objective/train/tokens_used": 212152800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24403183162212372,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506290197372437,
      "objective/train/weighted_lm_loss": 3.3478922843933105,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 4.37857420112113,
      "tokens_seen": 191692800
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004757171514543631,
      "loss": 3.1559,
      "theoretical_loss": 4.37857420112113,
      "tokens_seen": 191692800
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004757071213640923,
      "loss": 3.248,
      "theoretical_loss": 4.378385692060767,
      "tokens_seen": 191758336
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004756970912738215,
      "loss": 3.4246,
      "theoretical_loss": 4.378197265447032,
      "tokens_seen": 191823872
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047568706118355066,
      "loss": 3.3472,
      "theoretical_loss": 4.378008921215717,
      "tokens_seen": 191889408
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047567703109327984,
      "loss": 3.3461,
      "theoretical_loss": 4.377820659302687,
      "tokens_seen": 191954944
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.000475667001003009,
      "loss": 3.3929,
      "theoretical_loss": 4.377632479643877,
      "tokens_seen": 192020480
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047565697091273826,
      "loss": 3.394,
      "theoretical_loss": 4.377444382175294,
      "tokens_seen": 192086016
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004756469408224674,
      "loss": 3.2366,
      "theoretical_loss": 4.377256366833016,
      "tokens_seen": 192151552
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004756369107321966,
      "loss": 3.3089,
      "theoretical_loss": 4.3770684335531955,
      "tokens_seen": 192217088
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004756268806419258,
      "loss": 3.2558,
      "theoretical_loss": 4.376880582272053,
      "tokens_seen": 192282624
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.000475616850551655,
      "loss": 3.6578,
      "theoretical_loss": 4.3766928129258815,
      "tokens_seen": 192348160
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047560682046138416,
      "loss": 3.2968,
      "theoretical_loss": 4.376505125451045,
      "tokens_seen": 192413696
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047559679037111334,
      "loss": 3.4571,
      "theoretical_loss": 4.37631751978398,
      "tokens_seen": 192479232
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004755867602808425,
      "loss": 3.7011,
      "theoretical_loss": 4.376129995861191,
      "tokens_seen": 192544768
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047557673019057176,
      "loss": 3.2704,
      "theoretical_loss": 4.3759425536192555,
      "tokens_seen": 192610304
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004755667001003009,
      "loss": 3.3213,
      "theoretical_loss": 4.3757551929948235,
      "tokens_seen": 192675840
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004755566700100301,
      "loss": 3.4359,
      "theoretical_loss": 4.375567913924611,
      "tokens_seen": 192741376
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047554663991975925,
      "loss": 3.3473,
      "theoretical_loss": 4.375380716345409,
      "tokens_seen": 192806912
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004755366098294885,
      "loss": 3.1867,
      "theoretical_loss": 4.375193600194075,
      "tokens_seen": 192872448
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047552657973921766,
      "loss": 3.2316,
      "theoretical_loss": 4.375006565407541,
      "tokens_seen": 192937984
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047551654964894685,
      "loss": 3.2016,
      "theoretical_loss": 4.374819611922806,
      "tokens_seen": 193003520
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047550651955867603,
      "loss": 3.6866,
      "theoretical_loss": 4.374632739676942,
      "tokens_seen": 193069056
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047549648946840526,
      "loss": 3.4715,
      "theoretical_loss": 4.374445948607087,
      "tokens_seen": 193134592
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004754864593781344,
      "loss": 3.177,
      "theoretical_loss": 4.374259238650452,
      "tokens_seen": 193200128
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004754764292878636,
      "loss": 3.2715,
      "theoretical_loss": 4.374072609744318,
      "tokens_seen": 193265664
    },
    {
      "epoch": 0.06,
      "objective/train/advantage_avg": 0.4934258759021759,
      "objective/train/docs_used": 120055,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.32948637008667,
      "objective/train/original_loss": 3.329486131668091,
      "objective/train/theoretical_loss": 4.373886061826036,
      "objective/train/tokens_used": 213791200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2435128539800644,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050580382347107,
      "objective/train/weighted_lm_loss": 3.4976999759674072,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.047311544418335,
      "theoretical_loss": 4.373886061826036,
      "tokens_seen": 193331200
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047546639919759275,
      "loss": 3.5391,
      "theoretical_loss": 4.373886061826036,
      "tokens_seen": 193331200
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.000475456369107322,
      "loss": 3.424,
      "theoretical_loss": 4.373699594833022,
      "tokens_seen": 193396736
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047544633901705117,
      "loss": 3.2935,
      "theoretical_loss": 4.373513208702769,
      "tokens_seen": 193462272
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047543630892678035,
      "loss": 3.318,
      "theoretical_loss": 4.373326903372833,
      "tokens_seen": 193527808
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047542627883650953,
      "loss": 3.3387,
      "theoretical_loss": 4.3731406787808424,
      "tokens_seen": 193593344
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004754162487462387,
      "loss": 3.3715,
      "theoretical_loss": 4.372954534864496,
      "tokens_seen": 193658880
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004754062186559679,
      "loss": 3.4774,
      "theoretical_loss": 4.372768471561558,
      "tokens_seen": 193724416
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047539618856569713,
      "loss": 3.4791,
      "theoretical_loss": 4.372582488809863,
      "tokens_seen": 193789952
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047538615847542625,
      "loss": 3.4937,
      "theoretical_loss": 4.372396586547319,
      "tokens_seen": 193855488
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004753761283851555,
      "loss": 3.4428,
      "theoretical_loss": 4.372210764711896,
      "tokens_seen": 193921024
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004753660982948846,
      "loss": 3.3989,
      "theoretical_loss": 4.372025023241637,
      "tokens_seen": 193986560
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047535606820461385,
      "loss": 3.5315,
      "theoretical_loss": 4.371839362074652,
      "tokens_seen": 194052096
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047534603811434303,
      "loss": 3.4993,
      "theoretical_loss": 4.37165378114912,
      "tokens_seen": 194117632
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004753360080240722,
      "loss": 3.5458,
      "theoretical_loss": 4.371468280403288,
      "tokens_seen": 194183168
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004753259779338014,
      "loss": 3.3869,
      "theoretical_loss": 4.371282859775474,
      "tokens_seen": 194248704
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047531594784353063,
      "loss": 3.2841,
      "theoretical_loss": 4.371097519204059,
      "tokens_seen": 194314240
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047530591775325976,
      "loss": 3.3183,
      "theoretical_loss": 4.370912258627498,
      "tokens_seen": 194379776
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.000475295887662989,
      "loss": 3.413,
      "theoretical_loss": 4.370727077984309,
      "tokens_seen": 194445312
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004752858575727181,
      "loss": 3.2564,
      "theoretical_loss": 4.370541977213082,
      "tokens_seen": 194510848
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047527582748244736,
      "loss": 3.46,
      "theoretical_loss": 4.370356956252472,
      "tokens_seen": 194576384
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047526579739217654,
      "loss": 3.3302,
      "theoretical_loss": 4.370172015041203,
      "tokens_seen": 194641920
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004752557673019057,
      "loss": 3.4173,
      "theoretical_loss": 4.369987153518067,
      "tokens_seen": 194707456
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004752457372116349,
      "loss": 3.5199,
      "theoretical_loss": 4.369802371621922,
      "tokens_seen": 194772992
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004752357071213641,
      "loss": 3.1885,
      "theoretical_loss": 4.369617669291695,
      "tokens_seen": 194838528
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047522567703109326,
      "loss": 3.2085,
      "theoretical_loss": 4.36943304646638,
      "tokens_seen": 194904064
    },
    {
      "epoch": 0.06,
      "objective/train/advantage_avg": 0.4935552179813385,
      "objective/train/docs_used": 121265,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.2108192443847656,
      "objective/train/original_loss": 3.210819959640503,
      "objective/train/theoretical_loss": 4.369248503085039,
      "objective/train/tokens_used": 215429600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2436530590057373,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505939722061157,
      "objective/train/weighted_lm_loss": 3.3733160495758057,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0425469875335693,
      "theoretical_loss": 4.369248503085039,
      "tokens_seen": 194969600
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004752156469408225,
      "loss": 3.1944,
      "theoretical_loss": 4.369248503085039,
      "tokens_seen": 194969600
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004752056168505517,
      "loss": 3.6773,
      "theoretical_loss": 4.3690640390867985,
      "tokens_seen": 195035136
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047519558676028086,
      "loss": 3.3246,
      "theoretical_loss": 4.368879654410855,
      "tokens_seen": 195100672
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047518555667001004,
      "loss": 3.4913,
      "theoretical_loss": 4.368695348996471,
      "tokens_seen": 195166208
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004751755265797392,
      "loss": 3.4969,
      "theoretical_loss": 4.368511122782975,
      "tokens_seen": 195231744
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047516549648946846,
      "loss": 3.1788,
      "theoretical_loss": 4.3683269757097625,
      "tokens_seen": 195297280
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004751554663991976,
      "loss": 3.493,
      "theoretical_loss": 4.368142907716297,
      "tokens_seen": 195362816
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004751454363089268,
      "loss": 3.483,
      "theoretical_loss": 4.36795891874211,
      "tokens_seen": 195428352
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.000475135406218656,
      "loss": 3.4587,
      "theoretical_loss": 4.3677750087267935,
      "tokens_seen": 195493888
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004751253761283852,
      "loss": 3.6759,
      "theoretical_loss": 4.367591177610011,
      "tokens_seen": 195559424
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047511534603811436,
      "loss": 3.2671,
      "theoretical_loss": 4.367407425331493,
      "tokens_seen": 195624960
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047510531594784354,
      "loss": 3.4684,
      "theoretical_loss": 4.367223751831032,
      "tokens_seen": 195690496
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004750952858575727,
      "loss": 3.459,
      "theoretical_loss": 4.367040157048489,
      "tokens_seen": 195756032
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047508525576730196,
      "loss": 3.454,
      "theoretical_loss": 4.3668566409237926,
      "tokens_seen": 195821568
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004750752256770311,
      "loss": 3.2356,
      "theoretical_loss": 4.3666732033969335,
      "tokens_seen": 195887104
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004750651955867603,
      "loss": 3.3176,
      "theoretical_loss": 4.366489844407972,
      "tokens_seen": 195952640
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047505516549648945,
      "loss": 3.446,
      "theoretical_loss": 4.36630656389703,
      "tokens_seen": 196018176
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004750451354062187,
      "loss": 3.366,
      "theoretical_loss": 4.366123361804301,
      "tokens_seen": 196083712
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047503510531594787,
      "loss": 3.2443,
      "theoretical_loss": 4.3659402380700385,
      "tokens_seen": 196149248
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047502507522567705,
      "loss": 3.5757,
      "theoretical_loss": 4.365757192634564,
      "tokens_seen": 196214784
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047501504513540623,
      "loss": 3.1939,
      "theoretical_loss": 4.365574225438263,
      "tokens_seen": 196280320
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047500501504513546,
      "loss": 3.4778,
      "theoretical_loss": 4.365391336421589,
      "tokens_seen": 196345856
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004749949849548646,
      "loss": 3.4582,
      "theoretical_loss": 4.365208525525057,
      "tokens_seen": 196411392
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004749849548645938,
      "loss": 3.3546,
      "theoretical_loss": 4.36502579268925,
      "tokens_seen": 196476928
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047497492477432295,
      "loss": 3.1641,
      "theoretical_loss": 4.364843137854816,
      "tokens_seen": 196542464
    },
    {
      "epoch": 0.06,
      "objective/train/advantage_avg": 0.4935546815395355,
      "objective/train/docs_used": 121711,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.426180362701416,
      "objective/train/original_loss": 3.426180362701416,
      "objective/train/theoretical_loss": 4.364660560962464,
      "objective/train/tokens_used": 217068000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2436842918395996,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505940914154053,
      "objective/train/weighted_lm_loss": 3.599503517150879,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0425469875335693,
      "theoretical_loss": 4.364660560962464,
      "tokens_seen": 196608000
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004749648946840522,
      "loss": 3.3831,
      "theoretical_loss": 4.364660560962464,
      "tokens_seen": 196608000
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047495486459378137,
      "loss": 3.5212,
      "theoretical_loss": 4.364478061952973,
      "tokens_seen": 196673536
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047494483450351055,
      "loss": 3.405,
      "theoretical_loss": 4.364295640767184,
      "tokens_seen": 196739072
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047493480441323973,
      "loss": 3.2538,
      "theoretical_loss": 4.364113297346002,
      "tokens_seen": 196804608
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004749247743229689,
      "loss": 3.2134,
      "theoretical_loss": 4.363931031630398,
      "tokens_seen": 196870144
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004749147442326981,
      "loss": 3.5891,
      "theoretical_loss": 4.363748843561407,
      "tokens_seen": 196935680
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047490471414242733,
      "loss": 3.3788,
      "theoretical_loss": 4.363566733080129,
      "tokens_seen": 197001216
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047489468405215646,
      "loss": 3.5059,
      "theoretical_loss": 4.363384700127726,
      "tokens_seen": 197066752
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004748846539618857,
      "loss": 3.4662,
      "theoretical_loss": 4.363202744645427,
      "tokens_seen": 197132288
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004748746238716148,
      "loss": 3.1223,
      "theoretical_loss": 4.363020866574524,
      "tokens_seen": 197197824
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047486459378134405,
      "loss": 3.6354,
      "theoretical_loss": 4.362839065856372,
      "tokens_seen": 197263360
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047485456369107323,
      "loss": 3.3268,
      "theoretical_loss": 4.3626573424323905,
      "tokens_seen": 197328896
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004748445336008024,
      "loss": 3.2906,
      "theoretical_loss": 4.362475696244064,
      "tokens_seen": 197394432
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004748345035105316,
      "loss": 3.5836,
      "theoretical_loss": 4.362294127232939,
      "tokens_seen": 197459968
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047482447342026083,
      "loss": 3.4494,
      "theoretical_loss": 4.362112635340626,
      "tokens_seen": 197525504
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047481444332998996,
      "loss": 3.4225,
      "theoretical_loss": 4.3619312205088,
      "tokens_seen": 197591040
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004748044132397192,
      "loss": 3.612,
      "theoretical_loss": 4.361749882679199,
      "tokens_seen": 197656576
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004747943831494483,
      "loss": 3.5691,
      "theoretical_loss": 4.361568621793623,
      "tokens_seen": 197722112
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047478435305917756,
      "loss": 3.5773,
      "theoretical_loss": 4.361387437793937,
      "tokens_seen": 197787648
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047477432296890674,
      "loss": 3.3045,
      "theoretical_loss": 4.361206330622068,
      "tokens_seen": 197853184
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004747642928786359,
      "loss": 3.5541,
      "theoretical_loss": 4.361025300220008,
      "tokens_seen": 197918720
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004747542627883651,
      "loss": 3.2865,
      "theoretical_loss": 4.3608443465298095,
      "tokens_seen": 197984256
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004747442326980943,
      "loss": 3.116,
      "theoretical_loss": 4.360663469493589,
      "tokens_seen": 198049792
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047473420260782346,
      "loss": 3.6187,
      "theoretical_loss": 4.3604826690535265,
      "tokens_seen": 198115328
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004747241725175527,
      "loss": 3.3136,
      "theoretical_loss": 4.360301945151863,
      "tokens_seen": 198180864
    },
    {
      "epoch": 0.06,
      "objective/train/advantage_avg": 0.4937030076980591,
      "objective/train/docs_used": 122248,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.297278881072998,
      "objective/train/original_loss": 3.297278881072998,
      "objective/train/theoretical_loss": 4.360121297730904,
      "objective/train/tokens_used": 218706400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24380387365818024,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506097078323364,
      "objective/train/weighted_lm_loss": 3.4642958641052246,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0408108234405518,
      "theoretical_loss": 4.360121297730904,
      "tokens_seen": 198246400
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004747141424272818,
      "loss": 3.431,
      "theoretical_loss": 4.360121297730904,
      "tokens_seen": 198246400
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047470411233701106,
      "loss": 3.5432,
      "theoretical_loss": 4.359940726733017,
      "tokens_seen": 198311936
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004746940822467402,
      "loss": 3.2112,
      "theoretical_loss": 4.359760232100632,
      "tokens_seen": 198377472
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004746840521564694,
      "loss": 3.2775,
      "theoretical_loss": 4.359579813776239,
      "tokens_seen": 198443008
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004746740220661986,
      "loss": 3.519,
      "theoretical_loss": 4.359399471702394,
      "tokens_seen": 198508544
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004746639919759278,
      "loss": 3.5677,
      "theoretical_loss": 4.359219205821713,
      "tokens_seen": 198574080
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047465396188565696,
      "loss": 3.2063,
      "theoretical_loss": 4.359039016076876,
      "tokens_seen": 198639616
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004746439317953862,
      "loss": 3.165,
      "theoretical_loss": 4.358858902410621,
      "tokens_seen": 198705152
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004746339017051153,
      "loss": 3.2825,
      "theoretical_loss": 4.358678864765753,
      "tokens_seen": 198770688
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047462387161484456,
      "loss": 3.2146,
      "theoretical_loss": 4.358498903085135,
      "tokens_seen": 198836224
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004746138415245737,
      "loss": 3.2487,
      "theoretical_loss": 4.358319017311694,
      "tokens_seen": 198901760
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004746038114343029,
      "loss": 3.3362,
      "theoretical_loss": 4.358139207388417,
      "tokens_seen": 198967296
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004745937813440321,
      "loss": 3.4589,
      "theoretical_loss": 4.357959473258354,
      "tokens_seen": 199032832
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004745837512537613,
      "loss": 3.6343,
      "theoretical_loss": 4.3577798148646165,
      "tokens_seen": 199098368
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047457372116349047,
      "loss": 3.2463,
      "theoretical_loss": 4.357600232150376,
      "tokens_seen": 199163904
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047456369107321965,
      "loss": 3.501,
      "theoretical_loss": 4.357420725058867,
      "tokens_seen": 199229440
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047455366098294883,
      "loss": 3.3215,
      "theoretical_loss": 4.357241293533383,
      "tokens_seen": 199294976
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047454363089267807,
      "loss": 3.2844,
      "theoretical_loss": 4.3570619375172805,
      "tokens_seen": 199360512
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004745336008024072,
      "loss": 3.299,
      "theoretical_loss": 4.356882656953977,
      "tokens_seen": 199426048
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047452357071213643,
      "loss": 3.2776,
      "theoretical_loss": 4.3567034517869505,
      "tokens_seen": 199491584
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047451354062186555,
      "loss": 3.438,
      "theoretical_loss": 4.35652432195974,
      "tokens_seen": 199557120
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004745035105315948,
      "loss": 3.3397,
      "theoretical_loss": 4.3563452674159455,
      "tokens_seen": 199622656
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047449348044132397,
      "loss": 3.4031,
      "theoretical_loss": 4.356166288099227,
      "tokens_seen": 199688192
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047448345035105315,
      "loss": 3.3333,
      "theoretical_loss": 4.355987383953305,
      "tokens_seen": 199753728
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004744734202607824,
      "loss": 3.1021,
      "theoretical_loss": 4.355808554921964,
      "tokens_seen": 199819264
    },
    {
      "epoch": 0.06,
      "objective/train/advantage_avg": 0.4960399270057678,
      "objective/train/docs_used": 123553,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.2296133041381836,
      "objective/train/original_loss": 3.2296135425567627,
      "objective/train/theoretical_loss": 4.355629800949043,
      "objective/train/tokens_used": 220344800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24609293043613434,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050855040550232,
      "objective/train/weighted_lm_loss": 3.393812417984009,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0437887907028198,
      "theoretical_loss": 4.355629800949043,
      "tokens_seen": 199884800
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047446339017051157,
      "loss": 3.2686,
      "theoretical_loss": 4.355629800949043,
      "tokens_seen": 199884800
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047445336008024075,
      "loss": 3.355,
      "theoretical_loss": 4.3554511219784455,
      "tokens_seen": 199950336
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047444332998996993,
      "loss": 3.2805,
      "theoretical_loss": 4.355272517954134,
      "tokens_seen": 200015872
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004744332998996991,
      "loss": 3.3421,
      "theoretical_loss": 4.355093988820133,
      "tokens_seen": 200081408
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004744232698094283,
      "loss": 3.2293,
      "theoretical_loss": 4.354915534520525,
      "tokens_seen": 200146944
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047441323971915753,
      "loss": 3.54,
      "theoretical_loss": 4.354737154999452,
      "tokens_seen": 200212480
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047440320962888666,
      "loss": 3.1748,
      "theoretical_loss": 4.354558850201118,
      "tokens_seen": 200278016
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004743931795386159,
      "loss": 3.3855,
      "theoretical_loss": 4.3543806200697865,
      "tokens_seen": 200343552
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.000474383149448345,
      "loss": 3.1268,
      "theoretical_loss": 4.35420246454978,
      "tokens_seen": 200409088
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047437311935807425,
      "loss": 3.4996,
      "theoretical_loss": 4.354024383585481,
      "tokens_seen": 200474624
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047436308926780343,
      "loss": 3.2505,
      "theoretical_loss": 4.3538463771213305,
      "tokens_seen": 200540160
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004743530591775326,
      "loss": 3.3233,
      "theoretical_loss": 4.353668445101833,
      "tokens_seen": 200605696
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004743430290872618,
      "loss": 3.4902,
      "theoretical_loss": 4.353490587471546,
      "tokens_seen": 200671232
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047433299899699103,
      "loss": 3.6243,
      "theoretical_loss": 4.353312804175093,
      "tokens_seen": 200736768
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047432296890672016,
      "loss": 3.2862,
      "theoretical_loss": 4.353135095157152,
      "tokens_seen": 200802304
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004743129388164494,
      "loss": 3.4757,
      "theoretical_loss": 4.352957460362464,
      "tokens_seen": 200867840
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004743029087261785,
      "loss": 3.4707,
      "theoretical_loss": 4.352779899735825,
      "tokens_seen": 200933376
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047429287863590776,
      "loss": 3.4503,
      "theoretical_loss": 4.352602413222094,
      "tokens_seen": 200998912
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047428284854563694,
      "loss": 3.5404,
      "theoretical_loss": 4.3524250007661855,
      "tokens_seen": 201064448
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004742728184553661,
      "loss": 3.4568,
      "theoretical_loss": 4.352247662313075,
      "tokens_seen": 201129984
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004742627883650953,
      "loss": 3.7445,
      "theoretical_loss": 4.352070397807797,
      "tokens_seen": 201195520
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004742527582748245,
      "loss": 3.504,
      "theoretical_loss": 4.3518932071954435,
      "tokens_seen": 201261056
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047424272818455366,
      "loss": 3.4054,
      "theoretical_loss": 4.351716090421165,
      "tokens_seen": 201326592
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004742326980942829,
      "loss": 3.5227,
      "theoretical_loss": 4.351539047430173,
      "tokens_seen": 201392128
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.000474222668004012,
      "loss": 3.5383,
      "theoretical_loss": 4.351362078167735,
      "tokens_seen": 201457664
    },
    {
      "epoch": 0.06,
      "objective/train/advantage_avg": 0.4971599876880646,
      "objective/train/docs_used": 124253,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.02793025970459,
      "objective/train/original_loss": 3.027930498123169,
      "objective/train/theoretical_loss": 4.351185182579177,
      "objective/train/tokens_used": 221983200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24719446897506714,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0509727001190186,
      "objective/train/weighted_lm_loss": 3.1823017597198486,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0396549701690674,
      "theoretical_loss": 4.351185182579177,
      "tokens_seen": 201523200
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047421263791374126,
      "loss": 3.4516,
      "theoretical_loss": 4.351185182579177,
      "tokens_seen": 201523200
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004742026078234704,
      "loss": 3.5288,
      "theoretical_loss": 4.3510083606098835,
      "tokens_seen": 201588736
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004741925777331996,
      "loss": 3.4635,
      "theoretical_loss": 4.350831612205299,
      "tokens_seen": 201654272
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004741825476429288,
      "loss": 3.1951,
      "theoretical_loss": 4.350654937310924,
      "tokens_seen": 201719808
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.000474172517552658,
      "loss": 3.192,
      "theoretical_loss": 4.350478335872317,
      "tokens_seen": 201785344
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047416248746238716,
      "loss": 3.3349,
      "theoretical_loss": 4.350301807835097,
      "tokens_seen": 201850880
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004741524573721164,
      "loss": 3.3526,
      "theoretical_loss": 4.3501253531449375,
      "tokens_seen": 201916416
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004741424272818455,
      "loss": 3.3148,
      "theoretical_loss": 4.349948971747573,
      "tokens_seen": 201981952
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047413239719157476,
      "loss": 3.5045,
      "theoretical_loss": 4.3497726635887926,
      "tokens_seen": 202047488
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004741223671013039,
      "loss": 3.3787,
      "theoretical_loss": 4.349596428614447,
      "tokens_seen": 202113024
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004741123370110331,
      "loss": 3.4039,
      "theoretical_loss": 4.34942026677044,
      "tokens_seen": 202178560
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004741023069207623,
      "loss": 3.307,
      "theoretical_loss": 4.349244178002736,
      "tokens_seen": 202244096
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004740922768304915,
      "loss": 3.3663,
      "theoretical_loss": 4.349068162257357,
      "tokens_seen": 202309632
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047408224674022067,
      "loss": 3.2367,
      "theoretical_loss": 4.348892219480378,
      "tokens_seen": 202375168
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047407221664994985,
      "loss": 3.2597,
      "theoretical_loss": 4.348716349617939,
      "tokens_seen": 202440704
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047406218655967903,
      "loss": 3.3806,
      "theoretical_loss": 4.348540552616229,
      "tokens_seen": 202506240
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047405215646940827,
      "loss": 3.1722,
      "theoretical_loss": 4.3483648284214995,
      "tokens_seen": 202571776
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004740421263791374,
      "loss": 3.3297,
      "theoretical_loss": 4.348189176980057,
      "tokens_seen": 202637312
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047403209628886663,
      "loss": 3.3056,
      "theoretical_loss": 4.348013598238266,
      "tokens_seen": 202702848
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047402206619859575,
      "loss": 3.409,
      "theoretical_loss": 4.347838092142546,
      "tokens_seen": 202768384
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.000474012036108325,
      "loss": 3.2615,
      "theoretical_loss": 4.347662658639376,
      "tokens_seen": 202833920
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047400200601805417,
      "loss": 3.6166,
      "theoretical_loss": 4.347487297675288,
      "tokens_seen": 202899456
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047399197592778335,
      "loss": 3.5228,
      "theoretical_loss": 4.347312009196874,
      "tokens_seen": 202964992
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047398194583751253,
      "loss": 3.5579,
      "theoretical_loss": 4.347136793150781,
      "tokens_seen": 203030528
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047397191574724177,
      "loss": 3.3089,
      "theoretical_loss": 4.346961649483713,
      "tokens_seen": 203096064
    },
    {
      "epoch": 0.06,
      "objective/train/advantage_avg": 0.4948410391807556,
      "objective/train/docs_used": 125838,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.3435261249542236,
      "objective/train/original_loss": 3.343526601791382,
      "objective/train/theoretical_loss": 4.3467865781424315,
      "objective/train/tokens_used": 223621600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24492241442203522,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507292747497559,
      "objective/train/weighted_lm_loss": 3.5129811763763428,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0454468727111816,
      "theoretical_loss": 4.3467865781424315,
      "tokens_seen": 203161600
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004739618856569709,
      "loss": 3.2846,
      "theoretical_loss": 4.3467865781424315,
      "tokens_seen": 203161600
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047395185556670013,
      "loss": 3.5508,
      "theoretical_loss": 4.346611579073751,
      "tokens_seen": 203227136
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047394182547642926,
      "loss": 3.4722,
      "theoretical_loss": 4.346436652224545,
      "tokens_seen": 203292672
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004739317953861585,
      "loss": 3.4498,
      "theoretical_loss": 4.346261797541742,
      "tokens_seen": 203358208
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004739217652958877,
      "loss": 3.406,
      "theoretical_loss": 4.346087014972328,
      "tokens_seen": 203423744
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047391173520561686,
      "loss": 3.4674,
      "theoretical_loss": 4.345912304463344,
      "tokens_seen": 203489280
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047390170511534604,
      "loss": 3.3217,
      "theoretical_loss": 4.345737665961886,
      "tokens_seen": 203554816
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004738916750250752,
      "loss": 3.1988,
      "theoretical_loss": 4.3455630994151075,
      "tokens_seen": 203620352
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004738816449348044,
      "loss": 3.3691,
      "theoretical_loss": 4.345388604770218,
      "tokens_seen": 203685888
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047387161484453363,
      "loss": 3.3643,
      "theoretical_loss": 4.3452141819744785,
      "tokens_seen": 203751424
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047386158475426276,
      "loss": 3.298,
      "theoretical_loss": 4.345039830975213,
      "tokens_seen": 203816960
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.000473851554663992,
      "loss": 3.3145,
      "theoretical_loss": 4.344865551719795,
      "tokens_seen": 203882496
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004738415245737211,
      "loss": 3.4379,
      "theoretical_loss": 4.344691344155654,
      "tokens_seen": 203948032
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047383149448345036,
      "loss": 3.5176,
      "theoretical_loss": 4.344517208230279,
      "tokens_seen": 204013568
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047382146439317954,
      "loss": 3.4076,
      "theoretical_loss": 4.344343143891211,
      "tokens_seen": 204079104
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004738114343029087,
      "loss": 3.3364,
      "theoretical_loss": 4.344169151086046,
      "tokens_seen": 204144640
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004738014042126379,
      "loss": 3.3179,
      "theoretical_loss": 4.343995229762437,
      "tokens_seen": 204210176
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047379137412236714,
      "loss": 3.426,
      "theoretical_loss": 4.34382137986809,
      "tokens_seen": 204275712
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047378134403209626,
      "loss": 3.2512,
      "theoretical_loss": 4.343647601350769,
      "tokens_seen": 204341248
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004737713139418255,
      "loss": 3.2666,
      "theoretical_loss": 4.343473894158289,
      "tokens_seen": 204406784
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004737612838515546,
      "loss": 3.4137,
      "theoretical_loss": 4.343300258238523,
      "tokens_seen": 204472320
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047375125376128386,
      "loss": 3.2124,
      "theoretical_loss": 4.343126693539398,
      "tokens_seen": 204537856
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047374122367101304,
      "loss": 3.5229,
      "theoretical_loss": 4.342953200008895,
      "tokens_seen": 204603392
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004737311935807422,
      "loss": 3.204,
      "theoretical_loss": 4.342779777595051,
      "tokens_seen": 204668928
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047372116349047146,
      "loss": 3.1656,
      "theoretical_loss": 4.342606426245956,
      "tokens_seen": 204734464
    },
    {
      "epoch": 0.06,
      "objective/train/advantage_avg": 0.495534747838974,
      "objective/train/docs_used": 126453,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.8397738933563232,
      "objective/train/original_loss": 2.8397741317749023,
      "objective/train/theoretical_loss": 4.342433145909755,
      "objective/train/tokens_used": 225260000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24559704959392548,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0508019924163818,
      "objective/train/weighted_lm_loss": 2.9840543270111084,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 4.342433145909755,
      "tokens_seen": 204800000
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004737111334002006,
      "loss": 3.2944,
      "theoretical_loss": 4.342433145909755,
      "tokens_seen": 204800000
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004737011033099298,
      "loss": 3.1449,
      "theoretical_loss": 4.3422599365346475,
      "tokens_seen": 204865536
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.000473691073219659,
      "loss": 3.2656,
      "theoretical_loss": 4.342086798068888,
      "tokens_seen": 204931072
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004736810431293882,
      "loss": 3.6059,
      "theoretical_loss": 4.341913730460785,
      "tokens_seen": 204996608
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047367101303911736,
      "loss": 3.4713,
      "theoretical_loss": 4.341740733658701,
      "tokens_seen": 205062144
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004736609829488466,
      "loss": 3.405,
      "theoretical_loss": 4.34156780761105,
      "tokens_seen": 205127680
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004736509528585757,
      "loss": 3.3159,
      "theoretical_loss": 4.341394952266306,
      "tokens_seen": 205193216
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047364092276830496,
      "loss": 3.2535,
      "theoretical_loss": 4.341222167572992,
      "tokens_seen": 205258752
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004736308926780341,
      "loss": 3.4882,
      "theoretical_loss": 4.341049453479686,
      "tokens_seen": 205324288
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004736208625877633,
      "loss": 3.4878,
      "theoretical_loss": 4.340876809935021,
      "tokens_seen": 205389824
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004736108324974925,
      "loss": 3.3109,
      "theoretical_loss": 4.3407042368876825,
      "tokens_seen": 205455360
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004736008024072217,
      "loss": 3.3701,
      "theoretical_loss": 4.34053173428641,
      "tokens_seen": 205520896
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047359077231695087,
      "loss": 3.2176,
      "theoretical_loss": 4.340359302079998,
      "tokens_seen": 205586432
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047358074222668005,
      "loss": 3.6409,
      "theoretical_loss": 4.340186940217292,
      "tokens_seen": 205651968
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047357071213640923,
      "loss": 3.4794,
      "theoretical_loss": 4.340014648647193,
      "tokens_seen": 205717504
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047356068204613847,
      "loss": 3.0775,
      "theoretical_loss": 4.339842427318654,
      "tokens_seen": 205783040
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004735506519558676,
      "loss": 3.4465,
      "theoretical_loss": 4.339670276180684,
      "tokens_seen": 205848576
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047354062186559683,
      "loss": 3.5195,
      "theoretical_loss": 4.3394981951823395,
      "tokens_seen": 205914112
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047353059177532595,
      "loss": 3.5631,
      "theoretical_loss": 4.339326184272737,
      "tokens_seen": 205979648
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004735205616850552,
      "loss": 3.2034,
      "theoretical_loss": 4.339154243401042,
      "tokens_seen": 206045184
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047351053159478437,
      "loss": 3.3612,
      "theoretical_loss": 4.338982372516473,
      "tokens_seen": 206110720
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047350050150451355,
      "loss": 3.4322,
      "theoretical_loss": 4.338810571568304,
      "tokens_seen": 206176256
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047349047141424273,
      "loss": 3.1623,
      "theoretical_loss": 4.33863884050586,
      "tokens_seen": 206241792
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047348044132397197,
      "loss": 3.3721,
      "theoretical_loss": 4.338467179278519,
      "tokens_seen": 206307328
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004734704112337011,
      "loss": 3.0358,
      "theoretical_loss": 4.338295587835711,
      "tokens_seen": 206372864
    },
    {
      "epoch": 0.06,
      "objective/train/advantage_avg": 0.4944588243961334,
      "objective/train/docs_used": 127848,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.199066400527954,
      "objective/train/original_loss": 3.1990668773651123,
      "objective/train/theoretical_loss": 4.33812406612692,
      "objective/train/tokens_used": 226898400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24456678330898285,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506891012191772,
      "objective/train/weighted_lm_loss": 3.361234426498413,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0253151655197144,
      "theoretical_loss": 4.33812406612692,
      "tokens_seen": 206438400
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047346038114343033,
      "loss": 3.35,
      "theoretical_loss": 4.33812406612692,
      "tokens_seen": 206438400
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047345035105315946,
      "loss": 3.3375,
      "theoretical_loss": 4.337952614101683,
      "tokens_seen": 206503936
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004734403209628887,
      "loss": 3.6931,
      "theoretical_loss": 4.337781231709587,
      "tokens_seen": 206569472
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004734302908726179,
      "loss": 3.3256,
      "theoretical_loss": 4.337609918900275,
      "tokens_seen": 206635008
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047342026078234706,
      "loss": 3.4209,
      "theoretical_loss": 4.337438675623439,
      "tokens_seen": 206700544
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047341023069207624,
      "loss": 3.0433,
      "theoretical_loss": 4.3372675018288245,
      "tokens_seen": 206766080
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004734002006018054,
      "loss": 3.335,
      "theoretical_loss": 4.337096397466231,
      "tokens_seen": 206831616
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004733901705115346,
      "loss": 3.4254,
      "theoretical_loss": 4.336925362485507,
      "tokens_seen": 206897152
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047338014042126383,
      "loss": 3.4654,
      "theoretical_loss": 4.3367543968365565,
      "tokens_seen": 206962688
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047337011033099296,
      "loss": 3.3703,
      "theoretical_loss": 4.336583500469333,
      "tokens_seen": 207028224
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004733600802407222,
      "loss": 3.4909,
      "theoretical_loss": 4.336412673333841,
      "tokens_seen": 207093760
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004733500501504513,
      "loss": 3.4263,
      "theoretical_loss": 4.336241915380142,
      "tokens_seen": 207159296
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047334002006018056,
      "loss": 3.381,
      "theoretical_loss": 4.336071226558343,
      "tokens_seen": 207224832
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047332998996990974,
      "loss": 3.0569,
      "theoretical_loss": 4.3359006068186075,
      "tokens_seen": 207290368
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004733199598796389,
      "loss": 3.482,
      "theoretical_loss": 4.335730056111148,
      "tokens_seen": 207355904
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004733099297893681,
      "loss": 3.5339,
      "theoretical_loss": 4.335559574386228,
      "tokens_seen": 207421440
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047329989969909734,
      "loss": 3.4404,
      "theoretical_loss": 4.335389161594167,
      "tokens_seen": 207486976
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047328986960882646,
      "loss": 3.1195,
      "theoretical_loss": 4.33521881768533,
      "tokens_seen": 207552512
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004732798395185557,
      "loss": 3.2388,
      "theoretical_loss": 4.3350485426101395,
      "tokens_seen": 207618048
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004732698094282848,
      "loss": 3.4318,
      "theoretical_loss": 4.334878336319063,
      "tokens_seen": 207683584
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047325977933801406,
      "loss": 3.3613,
      "theoretical_loss": 4.334708198762624,
      "tokens_seen": 207749120
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047324974924774324,
      "loss": 3.222,
      "theoretical_loss": 4.334538129891396,
      "tokens_seen": 207814656
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004732397191574724,
      "loss": 3.2754,
      "theoretical_loss": 4.334368129656002,
      "tokens_seen": 207880192
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004732296890672016,
      "loss": 3.4618,
      "theoretical_loss": 4.334198198007117,
      "tokens_seen": 207945728
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004732196589769308,
      "loss": 3.129,
      "theoretical_loss": 4.33402833489547,
      "tokens_seen": 208011264
    },
    {
      "epoch": 0.06,
      "objective/train/advantage_avg": 0.49571681022644043,
      "objective/train/docs_used": 128338,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.321711540222168,
      "objective/train/original_loss": 3.321711540222168,
      "objective/train/theoretical_loss": 4.333858540271834,
      "objective/train/tokens_used": 228536800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24578484892845154,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.05082106590271,
      "objective/train/weighted_lm_loss": 3.490422010421753,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0425469875335693,
      "theoretical_loss": 4.333858540271834,
      "tokens_seen": 208076800
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047320962888665997,
      "loss": 3.4763,
      "theoretical_loss": 4.333858540271834,
      "tokens_seen": 208076800
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004731995987963892,
      "loss": 3.2729,
      "theoretical_loss": 4.33368881408704,
      "tokens_seen": 208142336
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047318956870611833,
      "loss": 3.504,
      "theoretical_loss": 4.333519156291965,
      "tokens_seen": 208207872
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047317953861584757,
      "loss": 3.4209,
      "theoretical_loss": 4.333349566837541,
      "tokens_seen": 208273408
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047316950852557675,
      "loss": 3.3426,
      "theoretical_loss": 4.3331800456747445,
      "tokens_seen": 208338944
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047315947843530593,
      "loss": 3.2281,
      "theoretical_loss": 4.333010592754609,
      "tokens_seen": 208404480
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004731494483450351,
      "loss": 3.1574,
      "theoretical_loss": 4.3328412080282135,
      "tokens_seen": 208470016
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004731394182547643,
      "loss": 3.302,
      "theoretical_loss": 4.332671891446692,
      "tokens_seen": 208535552
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047312938816449347,
      "loss": 3.2413,
      "theoretical_loss": 4.332502642961224,
      "tokens_seen": 208601088
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004731193580742227,
      "loss": 3.4709,
      "theoretical_loss": 4.332333462523044,
      "tokens_seen": 208666624
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047310932798395183,
      "loss": 3.1948,
      "theoretical_loss": 4.332164350083432,
      "tokens_seen": 208732160
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047309929789368107,
      "loss": 3.0033,
      "theoretical_loss": 4.331995305593723,
      "tokens_seen": 208797696
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004730892678034102,
      "loss": 3.5194,
      "theoretical_loss": 4.331826329005298,
      "tokens_seen": 208863232
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047307923771313943,
      "loss": 3.4711,
      "theoretical_loss": 4.3316574202695906,
      "tokens_seen": 208928768
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004730692076228686,
      "loss": 3.2807,
      "theoretical_loss": 4.3314885793380835,
      "tokens_seen": 208994304
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004730591775325978,
      "loss": 3.4371,
      "theoretical_loss": 4.331319806162309,
      "tokens_seen": 209059840
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.000473049147442327,
      "loss": 3.5063,
      "theoretical_loss": 4.33115110069385,
      "tokens_seen": 209125376
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047303911735205615,
      "loss": 3.4254,
      "theoretical_loss": 4.330982462884338,
      "tokens_seen": 209190912
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047302908726178534,
      "loss": 3.4323,
      "theoretical_loss": 4.3308138926854545,
      "tokens_seen": 209256448
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047301905717151457,
      "loss": 3.4492,
      "theoretical_loss": 4.330645390048932,
      "tokens_seen": 209321984
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004730090270812437,
      "loss": 3.5491,
      "theoretical_loss": 4.330476954926552,
      "tokens_seen": 209387520
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047299899699097293,
      "loss": 3.5176,
      "theoretical_loss": 4.330308587270143,
      "tokens_seen": 209453056
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004729889669007021,
      "loss": 3.1685,
      "theoretical_loss": 4.330140287031587,
      "tokens_seen": 209518592
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004729789368104313,
      "loss": 3.4069,
      "theoretical_loss": 4.329972054162813,
      "tokens_seen": 209584128
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047296890672016053,
      "loss": 3.2477,
      "theoretical_loss": 4.329803888615798,
      "tokens_seen": 209649664
    },
    {
      "epoch": 0.06,
      "objective/train/advantage_avg": 0.49206435680389404,
      "objective/train/docs_used": 129483,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.6365303993225098,
      "objective/train/original_loss": 3.6365301609039307,
      "objective/train/theoretical_loss": 4.3296357903425715,
      "objective/train/tokens_used": 230175200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2422230988740921,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050437569618225,
      "objective/train/weighted_lm_loss": 3.8198108673095703,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0408108234405518,
      "theoretical_loss": 4.3296357903425715,
      "tokens_seen": 209715200
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047295887662988966,
      "loss": 3.5988,
      "theoretical_loss": 4.3296357903425715,
      "tokens_seen": 209715200
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004729488465396189,
      "loss": 3.2148,
      "theoretical_loss": 4.329467759295211,
      "tokens_seen": 209780736
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004729388164493481,
      "loss": 3.3828,
      "theoretical_loss": 4.32929979542584,
      "tokens_seen": 209846272
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047292878635907726,
      "loss": 3.4815,
      "theoretical_loss": 4.3291318986866365,
      "tokens_seen": 209911808
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047291875626880644,
      "loss": 3.4334,
      "theoretical_loss": 4.328964069029822,
      "tokens_seen": 209977344
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004729087261785356,
      "loss": 3.5582,
      "theoretical_loss": 4.328796306407672,
      "tokens_seen": 210042880
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004728986960882648,
      "loss": 3.0508,
      "theoretical_loss": 4.3286286107725065,
      "tokens_seen": 210108416
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047288866599799403,
      "loss": 3.2715,
      "theoretical_loss": 4.328460982076697,
      "tokens_seen": 210173952
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047287863590772316,
      "loss": 3.2048,
      "theoretical_loss": 4.328293420272663,
      "tokens_seen": 210239488
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004728686058174524,
      "loss": 3.3365,
      "theoretical_loss": 4.328125925312872,
      "tokens_seen": 210305024
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004728585757271815,
      "loss": 3.3828,
      "theoretical_loss": 4.32795849714984,
      "tokens_seen": 210370560
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047284854563691076,
      "loss": 3.3091,
      "theoretical_loss": 4.327791135736133,
      "tokens_seen": 210436096
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047283851554663994,
      "loss": 3.3554,
      "theoretical_loss": 4.327623841024364,
      "tokens_seen": 210501632
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004728284854563691,
      "loss": 3.2658,
      "theoretical_loss": 4.327456612967195,
      "tokens_seen": 210567168
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004728184553660983,
      "loss": 3.3634,
      "theoretical_loss": 4.327289451517336,
      "tokens_seen": 210632704
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047280842527582754,
      "loss": 3.3476,
      "theoretical_loss": 4.327122356627546,
      "tokens_seen": 210698240
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047279839518555666,
      "loss": 3.4721,
      "theoretical_loss": 4.326955328250631,
      "tokens_seen": 210763776
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004727883650952859,
      "loss": 3.5147,
      "theoretical_loss": 4.326788366339445,
      "tokens_seen": 210829312
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.000472778335005015,
      "loss": 3.5053,
      "theoretical_loss": 4.326621470846892,
      "tokens_seen": 210894848
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047276830491474426,
      "loss": 3.5517,
      "theoretical_loss": 4.326454641725923,
      "tokens_seen": 210960384
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047275827482447344,
      "loss": 3.5328,
      "theoretical_loss": 4.326287878929536,
      "tokens_seen": 211025920
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004727482447342026,
      "loss": 3.0832,
      "theoretical_loss": 4.326121182410777,
      "tokens_seen": 211091456
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004727382146439318,
      "loss": 3.3283,
      "theoretical_loss": 4.325954552122742,
      "tokens_seen": 211156992
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.000472728184553661,
      "loss": 3.2901,
      "theoretical_loss": 4.325787988018572,
      "tokens_seen": 211222528
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047271815446339017,
      "loss": 3.449,
      "theoretical_loss": 4.325621490051457,
      "tokens_seen": 211288064
    },
    {
      "epoch": 0.06,
      "objective/train/advantage_avg": 0.49415093660354614,
      "objective/train/docs_used": 130263,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.5402567386627197,
      "objective/train/original_loss": 3.5402565002441406,
      "objective/train/theoretical_loss": 4.325455058174634,
      "objective/train/tokens_used": 231813600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2442353069782257,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506566762924194,
      "objective/train/weighted_lm_loss": 3.7195637226104736,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0454468727111816,
      "theoretical_loss": 4.325455058174634,
      "tokens_seen": 211353600
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004727081243731194,
      "loss": 3.6621,
      "theoretical_loss": 4.325455058174634,
      "tokens_seen": 211353600
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047269809428284853,
      "loss": 3.5465,
      "theoretical_loss": 4.325288692341388,
      "tokens_seen": 211419136
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047268806419257777,
      "loss": 3.4569,
      "theoretical_loss": 4.325122392505054,
      "tokens_seen": 211484672
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047267803410230695,
      "loss": 3.5498,
      "theoretical_loss": 4.324956158619008,
      "tokens_seen": 211550208
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047266800401203613,
      "loss": 3.4949,
      "theoretical_loss": 4.324789990636679,
      "tokens_seen": 211615744
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004726579739217653,
      "loss": 3.4349,
      "theoretical_loss": 4.3246238885115424,
      "tokens_seen": 211681280
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004726479438314945,
      "loss": 3.3454,
      "theoretical_loss": 4.324457852197119,
      "tokens_seen": 211746816
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047263791374122367,
      "loss": 3.4478,
      "theoretical_loss": 4.324291881646978,
      "tokens_seen": 211812352
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004726278836509529,
      "loss": 3.4195,
      "theoretical_loss": 4.324125976814734,
      "tokens_seen": 211877888
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047261785356068203,
      "loss": 3.1893,
      "theoretical_loss": 4.323960137654053,
      "tokens_seen": 211943424
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047260782347041127,
      "loss": 3.4192,
      "theoretical_loss": 4.323794364118644,
      "tokens_seen": 212008960
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004725977933801404,
      "loss": 3.3684,
      "theoretical_loss": 4.3236286561622626,
      "tokens_seen": 212074496
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047258776328986963,
      "loss": 3.2624,
      "theoretical_loss": 4.323463013738714,
      "tokens_seen": 212140032
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004725777331995988,
      "loss": 3.2673,
      "theoretical_loss": 4.323297436801848,
      "tokens_seen": 212205568
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.000472567703109328,
      "loss": 3.206,
      "theoretical_loss": 4.323131925305563,
      "tokens_seen": 212271104
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004725576730190572,
      "loss": 3.6303,
      "theoretical_loss": 4.322966479203803,
      "tokens_seen": 212336640
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047254764292878636,
      "loss": 3.3271,
      "theoretical_loss": 4.322801098450558,
      "tokens_seen": 212402176
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047253761283851554,
      "loss": 3.5194,
      "theoretical_loss": 4.322635782999866,
      "tokens_seen": 212467712
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047252758274824477,
      "loss": 3.6181,
      "theoretical_loss": 4.32247053280581,
      "tokens_seen": 212533248
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004725175526579739,
      "loss": 3.5683,
      "theoretical_loss": 4.322305347822521,
      "tokens_seen": 212598784
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047250752256770313,
      "loss": 3.2897,
      "theoretical_loss": 4.322140228004175,
      "tokens_seen": 212664320
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004724974924774323,
      "loss": 3.2841,
      "theoretical_loss": 4.321975173304995,
      "tokens_seen": 212729856
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004724874623871615,
      "loss": 3.5013,
      "theoretical_loss": 4.321810183679251,
      "tokens_seen": 212795392
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004724774322968907,
      "loss": 3.4187,
      "theoretical_loss": 4.321645259081256,
      "tokens_seen": 212860928
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047246740220661986,
      "loss": 3.5888,
      "theoretical_loss": 4.321480399465375,
      "tokens_seen": 212926464
    },
    {
      "epoch": 0.06,
      "objective/train/advantage_avg": 0.4957270622253418,
      "objective/train/docs_used": 131630,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.2086148262023926,
      "objective/train/original_loss": 2.2086148262023926,
      "objective/train/theoretical_loss": 4.321315604786012,
      "objective/train/tokens_used": 233452000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24581269919872284,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0508222579956055,
      "objective/train/weighted_lm_loss": 2.3208699226379395,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 4.321315604786012,
      "tokens_seen": 212992000
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047245737211634904,
      "loss": 3.0376,
      "theoretical_loss": 4.321315604786012,
      "tokens_seen": 212992000
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004724473420260783,
      "loss": 3.2508,
      "theoretical_loss": 4.321150874997622,
      "tokens_seen": 213057536
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004724373119358074,
      "loss": 3.3017,
      "theoretical_loss": 4.320986210054705,
      "tokens_seen": 213123072
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047242728184553664,
      "loss": 3.3286,
      "theoretical_loss": 4.320821609911805,
      "tokens_seen": 213188608
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047241725175526576,
      "loss": 3.274,
      "theoretical_loss": 4.320657074523515,
      "tokens_seen": 213254144
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.000472407221664995,
      "loss": 3.2877,
      "theoretical_loss": 4.32049260384447,
      "tokens_seen": 213319680
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004723971915747242,
      "loss": 2.884,
      "theoretical_loss": 4.320328197829354,
      "tokens_seen": 213385216
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047238716148445336,
      "loss": 3.4224,
      "theoretical_loss": 4.3201638564328935,
      "tokens_seen": 213450752
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047237713139418254,
      "loss": 3.3892,
      "theoretical_loss": 4.319999579609864,
      "tokens_seen": 213516288
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004723671013039117,
      "loss": 3.1934,
      "theoretical_loss": 4.319835367315085,
      "tokens_seen": 213581824
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004723570712136409,
      "loss": 3.4773,
      "theoretical_loss": 4.31967121950342,
      "tokens_seen": 213647360
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047234704112337014,
      "loss": 3.3083,
      "theoretical_loss": 4.31950713612978,
      "tokens_seen": 213712896
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047233701103309927,
      "loss": 3.2625,
      "theoretical_loss": 4.3193431171491214,
      "tokens_seen": 213778432
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004723269809428285,
      "loss": 3.2759,
      "theoretical_loss": 4.319179162516443,
      "tokens_seen": 213843968
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004723169508525577,
      "loss": 3.5301,
      "theoretical_loss": 4.3190152721867925,
      "tokens_seen": 213909504
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047230692076228686,
      "loss": 3.3998,
      "theoretical_loss": 4.318851446115261,
      "tokens_seen": 213975040
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047229689067201605,
      "loss": 3.2634,
      "theoretical_loss": 4.318687684256984,
      "tokens_seen": 214040576
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004722868605817452,
      "loss": 3.3461,
      "theoretical_loss": 4.318523986567144,
      "tokens_seen": 214106112
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.0004722768304914744,
      "loss": 3.4289,
      "theoretical_loss": 4.318360353000967,
      "tokens_seen": 214171648
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047226680040120364,
      "loss": 3.1528,
      "theoretical_loss": 4.318196783513724,
      "tokens_seen": 214237184
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047225677031093277,
      "loss": 3.3906,
      "theoretical_loss": 4.318033278060732,
      "tokens_seen": 214302720
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.000472246740220662,
      "loss": 3.4106,
      "theoretical_loss": 4.31786983659735,
      "tokens_seen": 214368256
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047223671013039113,
      "loss": 3.4642,
      "theoretical_loss": 4.317706459078985,
      "tokens_seen": 214433792
    },
    {
      "epoch": 0.06,
      "learning_rate": 0.00047222668004012037,
      "loss": 3.3755,
      "theoretical_loss": 4.317543145461089,
      "tokens_seen": 214499328
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004722166499498496,
      "loss": 3.363,
      "theoretical_loss": 4.317379895699155,
      "tokens_seen": 214564864
    },
    {
      "epoch": 0.07,
      "objective/train/advantage_avg": 0.4948301911354065,
      "objective/train/docs_used": 132228,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.3390228748321533,
      "objective/train/original_loss": 3.339022636413574,
      "objective/train/theoretical_loss": 4.317216709748722,
      "objective/train/tokens_used": 235090400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24493485689163208,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507280826568604,
      "objective/train/weighted_lm_loss": 3.5083279609680176,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0304546356201172,
      "theoretical_loss": 4.317216709748722,
      "tokens_seen": 214630400
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00047220661985957873,
      "loss": 3.3125,
      "theoretical_loss": 4.317216709748722,
      "tokens_seen": 214630400
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00047219658976930797,
      "loss": 3.01,
      "theoretical_loss": 4.317053587565377,
      "tokens_seen": 214695936
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00047218655967903715,
      "loss": 3.2437,
      "theoretical_loss": 4.3168905291047475,
      "tokens_seen": 214761472
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00047217652958876633,
      "loss": 3.4502,
      "theoretical_loss": 4.316727534322505,
      "tokens_seen": 214827008
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004721664994984955,
      "loss": 3.1968,
      "theoretical_loss": 4.316564603174369,
      "tokens_seen": 214892544
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004721564694082247,
      "loss": 3.3729,
      "theoretical_loss": 4.3164017356160995,
      "tokens_seen": 214958080
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00047214643931795387,
      "loss": 3.5623,
      "theoretical_loss": 4.316238931603504,
      "tokens_seen": 215023616
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004721364092276831,
      "loss": 3.3183,
      "theoretical_loss": 4.316076191092431,
      "tokens_seen": 215089152
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00047212637913741223,
      "loss": 3.3284,
      "theoretical_loss": 4.315913514038776,
      "tokens_seen": 215154688
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00047211634904714147,
      "loss": 3.3516,
      "theoretical_loss": 4.315750900398476,
      "tokens_seen": 215220224
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004721063189568706,
      "loss": 3.3021,
      "theoretical_loss": 4.3155883501275145,
      "tokens_seen": 215285760
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00047209628886659983,
      "loss": 3.3949,
      "theoretical_loss": 4.315425863181917,
      "tokens_seen": 215351296
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.000472086258776329,
      "loss": 3.3569,
      "theoretical_loss": 4.315263439517754,
      "tokens_seen": 215416832
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004720762286860582,
      "loss": 3.4861,
      "theoretical_loss": 4.315101079091139,
      "tokens_seen": 215482368
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004720661985957874,
      "loss": 3.5008,
      "theoretical_loss": 4.314938781858229,
      "tokens_seen": 215547904
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00047205616850551656,
      "loss": 3.4357,
      "theoretical_loss": 4.314776547775228,
      "tokens_seen": 215613440
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00047204613841524574,
      "loss": 3.3602,
      "theoretical_loss": 4.314614376798378,
      "tokens_seen": 215678976
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00047203610832497497,
      "loss": 3.0281,
      "theoretical_loss": 4.314452268883969,
      "tokens_seen": 215744512
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004720260782347041,
      "loss": 3.4055,
      "theoretical_loss": 4.314290223988333,
      "tokens_seen": 215810048
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00047201604814443333,
      "loss": 3.3821,
      "theoretical_loss": 4.314128242067846,
      "tokens_seen": 215875584
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004720060180541625,
      "loss": 3.3723,
      "theoretical_loss": 4.313966323078927,
      "tokens_seen": 215941120
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004719959879638917,
      "loss": 3.6314,
      "theoretical_loss": 4.313804466978039,
      "tokens_seen": 216006656
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004719859578736209,
      "loss": 3.3856,
      "theoretical_loss": 4.313642673721686,
      "tokens_seen": 216072192
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00047197592778335006,
      "loss": 3.1488,
      "theoretical_loss": 4.313480943266418,
      "tokens_seen": 216137728
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00047196589769307924,
      "loss": 3.4816,
      "theoretical_loss": 4.313319275568828,
      "tokens_seen": 216203264
    },
    {
      "epoch": 0.07,
      "objective/train/advantage_avg": 0.49443310499191284,
      "objective/train/docs_used": 133529,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.7198426723480225,
      "objective/train/original_loss": 3.7198429107666016,
      "objective/train/theoretical_loss": 4.313157670585552,
      "objective/train/tokens_used": 236728800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24456804990768433,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506864786148071,
      "objective/train/weighted_lm_loss": 3.9085710048675537,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0396549701690674,
      "theoretical_loss": 4.313157670585552,
      "tokens_seen": 216268800
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004719558676028085,
      "loss": 3.6163,
      "theoretical_loss": 4.313157670585552,
      "tokens_seen": 216268800
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004719458375125376,
      "loss": 3.2992,
      "theoretical_loss": 4.312996128273267,
      "tokens_seen": 216334336
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00047193580742226684,
      "loss": 3.4299,
      "theoretical_loss": 4.312834648588694,
      "tokens_seen": 216399872
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00047192577733199596,
      "loss": 3.0625,
      "theoretical_loss": 4.312673231488599,
      "tokens_seen": 216465408
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004719157472417252,
      "loss": 3.2843,
      "theoretical_loss": 4.312511876929789,
      "tokens_seen": 216530944
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004719057171514544,
      "loss": 3.194,
      "theoretical_loss": 4.312350584869114,
      "tokens_seen": 216596480
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00047189568706118356,
      "loss": 3.344,
      "theoretical_loss": 4.3121893552634685,
      "tokens_seen": 216662016
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00047188565697091274,
      "loss": 3.3888,
      "theoretical_loss": 4.312028188069787,
      "tokens_seen": 216727552
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004718756268806419,
      "loss": 3.6016,
      "theoretical_loss": 4.311867083245048,
      "tokens_seen": 216793088
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004718655967903711,
      "loss": 3.3809,
      "theoretical_loss": 4.311706040746273,
      "tokens_seen": 216858624
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00047185556670010034,
      "loss": 3.0825,
      "theoretical_loss": 4.311545060530526,
      "tokens_seen": 216924160
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00047184553660982947,
      "loss": 3.1923,
      "theoretical_loss": 4.311384142554914,
      "tokens_seen": 216989696
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004718355065195587,
      "loss": 3.587,
      "theoretical_loss": 4.311223286776586,
      "tokens_seen": 217055232
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004718254764292879,
      "loss": 3.1934,
      "theoretical_loss": 4.311062493152731,
      "tokens_seen": 217120768
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00047181544633901706,
      "loss": 3.476,
      "theoretical_loss": 4.310901761640586,
      "tokens_seen": 217186304
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00047180541624874625,
      "loss": 3.4244,
      "theoretical_loss": 4.310741092197424,
      "tokens_seen": 217251840
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004717953861584754,
      "loss": 3.4312,
      "theoretical_loss": 4.310580484780565,
      "tokens_seen": 217317376
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004717853560682046,
      "loss": 3.2727,
      "theoretical_loss": 4.310419939347369,
      "tokens_seen": 217382912
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00047177532597793384,
      "loss": 3.4913,
      "theoretical_loss": 4.310259455855238,
      "tokens_seen": 217448448
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00047176529588766297,
      "loss": 3.1087,
      "theoretical_loss": 4.3100990342616186,
      "tokens_seen": 217513984
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004717552657973922,
      "loss": 3.4969,
      "theoretical_loss": 4.309938674523996,
      "tokens_seen": 217579520
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00047174523570712133,
      "loss": 3.2789,
      "theoretical_loss": 4.3097783765998985,
      "tokens_seen": 217645056
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00047173520561685057,
      "loss": 3.3144,
      "theoretical_loss": 4.309618140446899,
      "tokens_seen": 217710592
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00047172517552657975,
      "loss": 3.5621,
      "theoretical_loss": 4.309457966022608,
      "tokens_seen": 217776128
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00047171514543630893,
      "loss": 3.2087,
      "theoretical_loss": 4.30929785328468,
      "tokens_seen": 217841664
    },
    {
      "epoch": 0.07,
      "objective/train/advantage_avg": 0.4954383373260498,
      "objective/train/docs_used": 134228,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.451570987701416,
      "objective/train/original_loss": 3.451570749282837,
      "objective/train/theoretical_loss": 4.309137802190812,
      "objective/train/tokens_used": 238367200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24553583562374115,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507919788360596,
      "objective/train/weighted_lm_loss": 3.626997709274292,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0363596677780151,
      "theoretical_loss": 4.309137802190812,
      "tokens_seen": 217907200
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004717051153460381,
      "loss": 3.2707,
      "theoretical_loss": 4.309137802190812,
      "tokens_seen": 217907200
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00047169508525576735,
      "loss": 3.4012,
      "theoretical_loss": 4.308977812698743,
      "tokens_seen": 217972736
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004716850551654965,
      "loss": 3.4991,
      "theoretical_loss": 4.30881788476625,
      "tokens_seen": 218038272
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004716750250752257,
      "loss": 3.2216,
      "theoretical_loss": 4.3086580183511565,
      "tokens_seen": 218103808
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00047166499498495484,
      "loss": 3.4859,
      "theoretical_loss": 4.308498213411323,
      "tokens_seen": 218169344
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00047165496489468407,
      "loss": 3.2536,
      "theoretical_loss": 4.3083384699046565,
      "tokens_seen": 218234880
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00047164493480441325,
      "loss": 3.367,
      "theoretical_loss": 4.308178787789101,
      "tokens_seen": 218300416
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00047163490471414243,
      "loss": 3.2301,
      "theoretical_loss": 4.3080191670226435,
      "tokens_seen": 218365952
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004716248746238716,
      "loss": 3.5891,
      "theoretical_loss": 4.307859607563313,
      "tokens_seen": 218431488
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004716148445336008,
      "loss": 3.2174,
      "theoretical_loss": 4.30770010936918,
      "tokens_seen": 218497024
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00047160481444333,
      "loss": 3.3146,
      "theoretical_loss": 4.307540672398354,
      "tokens_seen": 218562560
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004715947843530592,
      "loss": 3.5125,
      "theoretical_loss": 4.307381296608988,
      "tokens_seen": 218628096
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00047158475426278834,
      "loss": 3.2194,
      "theoretical_loss": 4.307221981959276,
      "tokens_seen": 218693632
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004715747241725176,
      "loss": 3.2854,
      "theoretical_loss": 4.307062728407451,
      "tokens_seen": 218759168
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004715646940822467,
      "loss": 3.4986,
      "theoretical_loss": 4.30690353591179,
      "tokens_seen": 218824704
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00047155466399197594,
      "loss": 3.2412,
      "theoretical_loss": 4.306744404430608,
      "tokens_seen": 218890240
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004715446339017051,
      "loss": 3.3608,
      "theoretical_loss": 4.306585333922263,
      "tokens_seen": 218955776
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004715346038114343,
      "loss": 3.3997,
      "theoretical_loss": 4.3064263243451535,
      "tokens_seen": 219021312
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004715245737211635,
      "loss": 3.3921,
      "theoretical_loss": 4.306267375657719,
      "tokens_seen": 219086848
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004715145436308927,
      "loss": 3.5181,
      "theoretical_loss": 4.306108487818438,
      "tokens_seen": 219152384
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00047150451354062184,
      "loss": 3.176,
      "theoretical_loss": 4.305949660785832,
      "tokens_seen": 219217920
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004714944834503511,
      "loss": 3.4354,
      "theoretical_loss": 4.305790894518463,
      "tokens_seen": 219283456
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004714844533600802,
      "loss": 3.1881,
      "theoretical_loss": 4.30563218897493,
      "tokens_seen": 219348992
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00047147442326980944,
      "loss": 3.3765,
      "theoretical_loss": 4.3054735441138785,
      "tokens_seen": 219414528
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004714643931795387,
      "loss": 3.4192,
      "theoretical_loss": 4.305314959893989,
      "tokens_seen": 219480064
    },
    {
      "epoch": 0.07,
      "objective/train/advantage_avg": 0.49497190117836,
      "objective/train/docs_used": 135676,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.721236228942871,
      "objective/train/original_loss": 3.72123646736145,
      "objective/train/theoretical_loss": 4.305156436273988,
      "objective/train/tokens_used": 240005600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24508699774742126,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507431030273438,
      "objective/train/weighted_lm_loss": 3.910102605819702,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0338950157165527,
      "theoretical_loss": 4.305156436273988,
      "tokens_seen": 219545600
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004714543630892678,
      "loss": 3.1863,
      "theoretical_loss": 4.305156436273988,
      "tokens_seen": 219545600
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00047144433299899704,
      "loss": 3.4613,
      "theoretical_loss": 4.304997973212636,
      "tokens_seen": 219611136
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00047143430290872616,
      "loss": 3.4833,
      "theoretical_loss": 4.304839570668738,
      "tokens_seen": 219676672
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004714242728184554,
      "loss": 3.3172,
      "theoretical_loss": 4.30468122860114,
      "tokens_seen": 219742208
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004714142427281846,
      "loss": 3.4566,
      "theoretical_loss": 4.304522946968725,
      "tokens_seen": 219807744
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00047140421263791376,
      "loss": 3.3406,
      "theoretical_loss": 4.3043647257304185,
      "tokens_seen": 219873280
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00047139418254764294,
      "loss": 3.2383,
      "theoretical_loss": 4.3042065648451855,
      "tokens_seen": 219938816
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004713841524573721,
      "loss": 3.3665,
      "theoretical_loss": 4.304048464272031,
      "tokens_seen": 220004352
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004713741223671013,
      "loss": 3.4381,
      "theoretical_loss": 4.30389042397,
      "tokens_seen": 220069888
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00047136409227683054,
      "loss": 3.2383,
      "theoretical_loss": 4.303732443898179,
      "tokens_seen": 220135424
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00047135406218655967,
      "loss": 3.4792,
      "theoretical_loss": 4.3035745240156915,
      "tokens_seen": 220200960
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004713440320962889,
      "loss": 3.4655,
      "theoretical_loss": 4.303416664281703,
      "tokens_seen": 220266496
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004713340020060181,
      "loss": 3.5672,
      "theoretical_loss": 4.303258864655419,
      "tokens_seen": 220332032
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00047132397191574726,
      "loss": 3.3774,
      "theoretical_loss": 4.303101125096083,
      "tokens_seen": 220397568
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00047131394182547645,
      "loss": 3.4515,
      "theoretical_loss": 4.302943445562979,
      "tokens_seen": 220463104
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00047130391173520563,
      "loss": 3.5172,
      "theoretical_loss": 4.302785826015434,
      "tokens_seen": 220528640
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004712938816449348,
      "loss": 3.6472,
      "theoretical_loss": 4.302628266412809,
      "tokens_seen": 220594176
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00047128385155466404,
      "loss": 3.3156,
      "theoretical_loss": 4.302470766714508,
      "tokens_seen": 220659712
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00047127382146439317,
      "loss": 3.3516,
      "theoretical_loss": 4.3023133268799745,
      "tokens_seen": 220725248
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004712637913741224,
      "loss": 3.3913,
      "theoretical_loss": 4.302155946868689,
      "tokens_seen": 220790784
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00047125376128385153,
      "loss": 3.2727,
      "theoretical_loss": 4.301998626640176,
      "tokens_seen": 220856320
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00047124373119358077,
      "loss": 3.3875,
      "theoretical_loss": 4.301841366153996,
      "tokens_seen": 220921856
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00047123370110330995,
      "loss": 3.4233,
      "theoretical_loss": 4.301684165369747,
      "tokens_seen": 220987392
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00047122367101303913,
      "loss": 3.5459,
      "theoretical_loss": 4.301527024247071,
      "tokens_seen": 221052928
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004712136409227683,
      "loss": 3.4701,
      "theoretical_loss": 4.301369942745647,
      "tokens_seen": 221118464
    },
    {
      "epoch": 0.07,
      "objective/train/advantage_avg": 0.4934816360473633,
      "objective/train/docs_used": 136371,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.691128969192505,
      "objective/train/original_loss": 2.691129207611084,
      "objective/train/theoretical_loss": 4.3012129208251935,
      "objective/train/tokens_used": 241644000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24362090229988098,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050586462020874,
      "objective/train/weighted_lm_loss": 2.827193260192871,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0304546356201172,
      "theoretical_loss": 4.3012129208251935,
      "tokens_seen": 221184000
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00047120361083249755,
      "loss": 3.2813,
      "theoretical_loss": 4.3012129208251935,
      "tokens_seen": 221184000
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004711935807422267,
      "loss": 3.4846,
      "theoretical_loss": 4.301055958445467,
      "tokens_seen": 221249536
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004711835506519559,
      "loss": 3.2889,
      "theoretical_loss": 4.300899055566264,
      "tokens_seen": 221315072
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00047117352056168504,
      "loss": 3.3029,
      "theoretical_loss": 4.300742212147419,
      "tokens_seen": 221380608
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00047116349047141427,
      "loss": 3.2739,
      "theoretical_loss": 4.300585428148809,
      "tokens_seen": 221446144
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00047115346038114345,
      "loss": 3.2689,
      "theoretical_loss": 4.300428703530346,
      "tokens_seen": 221511680
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00047114343029087263,
      "loss": 3.3378,
      "theoretical_loss": 4.300272038251983,
      "tokens_seen": 221577216
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004711334002006018,
      "loss": 3.3537,
      "theoretical_loss": 4.30011543227371,
      "tokens_seen": 221642752
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.000471123370110331,
      "loss": 3.2668,
      "theoretical_loss": 4.299958885555558,
      "tokens_seen": 221708288
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004711133400200602,
      "loss": 3.6132,
      "theoretical_loss": 4.2998023980575955,
      "tokens_seen": 221773824
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004711033099297894,
      "loss": 3.4317,
      "theoretical_loss": 4.299645969739929,
      "tokens_seen": 221839360
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00047109327983951854,
      "loss": 3.1034,
      "theoretical_loss": 4.299489600562707,
      "tokens_seen": 221904896
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004710832497492478,
      "loss": 3.4719,
      "theoretical_loss": 4.299333290486111,
      "tokens_seen": 221970432
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004710732196589769,
      "loss": 3.4251,
      "theoretical_loss": 4.299177039470367,
      "tokens_seen": 222035968
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00047106318956870614,
      "loss": 3.2384,
      "theoretical_loss": 4.2990208474757345,
      "tokens_seen": 222101504
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004710531594784353,
      "loss": 3.2894,
      "theoretical_loss": 4.298864714462514,
      "tokens_seen": 222167040
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004710431293881645,
      "loss": 3.3308,
      "theoretical_loss": 4.298708640391046,
      "tokens_seen": 222232576
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004710330992978937,
      "loss": 3.35,
      "theoretical_loss": 4.2985526252217054,
      "tokens_seen": 222298112
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004710230692076229,
      "loss": 3.2979,
      "theoretical_loss": 4.298396668914909,
      "tokens_seen": 222363648
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00047101303911735204,
      "loss": 3.2758,
      "theoretical_loss": 4.298240771431109,
      "tokens_seen": 222429184
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004710030090270813,
      "loss": 3.1999,
      "theoretical_loss": 4.298084932730797,
      "tokens_seen": 222494720
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004709929789368104,
      "loss": 3.1662,
      "theoretical_loss": 4.297929152774504,
      "tokens_seen": 222560256
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00047098294884653964,
      "loss": 3.5133,
      "theoretical_loss": 4.297773431522797,
      "tokens_seen": 222625792
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004709729187562688,
      "loss": 3.3515,
      "theoretical_loss": 4.297617768936283,
      "tokens_seen": 222691328
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.000470962888665998,
      "loss": 3.3431,
      "theoretical_loss": 4.297462164975606,
      "tokens_seen": 222756864
    },
    {
      "epoch": 0.07,
      "objective/train/advantage_avg": 0.49245473742485046,
      "objective/train/docs_used": 137650,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.8358867168426514,
      "objective/train/original_loss": 2.835886240005493,
      "objective/train/theoretical_loss": 4.297306619601446,
      "objective/train/tokens_used": 243282400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24259643256664276,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0504786968231201,
      "objective/train/weighted_lm_loss": 2.978990077972412,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0338950157165527,
      "theoretical_loss": 4.297306619601446,
      "tokens_seen": 222822400
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004709528585757272,
      "loss": 3.1249,
      "theoretical_loss": 4.297306619601446,
      "tokens_seen": 222822400
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00047094282848545636,
      "loss": 3.2169,
      "theoretical_loss": 4.297151132774527,
      "tokens_seen": 222887936
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00047093279839518555,
      "loss": 3.361,
      "theoretical_loss": 4.296995704455602,
      "tokens_seen": 222953472
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004709227683049148,
      "loss": 3.6015,
      "theoretical_loss": 4.29684033460547,
      "tokens_seen": 223019008
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004709127382146439,
      "loss": 3.3718,
      "theoretical_loss": 4.296685023184963,
      "tokens_seen": 223084544
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00047090270812437314,
      "loss": 3.4504,
      "theoretical_loss": 4.296529770154954,
      "tokens_seen": 223150080
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00047089267803410227,
      "loss": 3.0993,
      "theoretical_loss": 4.296374575476349,
      "tokens_seen": 223215616
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004708826479438315,
      "loss": 3.4547,
      "theoretical_loss": 4.2962194391100965,
      "tokens_seen": 223281152
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004708726178535607,
      "loss": 3.6027,
      "theoretical_loss": 4.296064361017181,
      "tokens_seen": 223346688
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00047086258776328987,
      "loss": 3.4645,
      "theoretical_loss": 4.295909341158622,
      "tokens_seen": 223412224
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00047085255767301905,
      "loss": 3.2452,
      "theoretical_loss": 4.295754379495481,
      "tokens_seen": 223477760
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004708425275827483,
      "loss": 3.2008,
      "theoretical_loss": 4.295599475988853,
      "tokens_seen": 223543296
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004708324974924774,
      "loss": 3.4218,
      "theoretical_loss": 4.295444630599873,
      "tokens_seen": 223608832
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00047082246740220665,
      "loss": 3.2976,
      "theoretical_loss": 4.29528984328971,
      "tokens_seen": 223674368
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004708124373119358,
      "loss": 3.3103,
      "theoretical_loss": 4.295135114019576,
      "tokens_seen": 223739904
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.000470802407221665,
      "loss": 3.3018,
      "theoretical_loss": 4.294980442750715,
      "tokens_seen": 223805440
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004707923771313942,
      "loss": 3.3891,
      "theoretical_loss": 4.294825829444411,
      "tokens_seen": 223870976
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00047078234704112337,
      "loss": 3.5242,
      "theoretical_loss": 4.294671274061983,
      "tokens_seen": 223936512
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00047077231695085255,
      "loss": 3.3832,
      "theoretical_loss": 4.29451677656479,
      "tokens_seen": 224002048
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00047076228686058173,
      "loss": 3.2483,
      "theoretical_loss": 4.294362336914225,
      "tokens_seen": 224067584
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004707522567703109,
      "loss": 3.2883,
      "theoretical_loss": 4.2942079550717205,
      "tokens_seen": 224133120
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00047074222668004015,
      "loss": 3.3585,
      "theoretical_loss": 4.294053630998745,
      "tokens_seen": 224198656
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004707321965897693,
      "loss": 3.3587,
      "theoretical_loss": 4.293899364656803,
      "tokens_seen": 224264192
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004707221664994985,
      "loss": 3.2745,
      "theoretical_loss": 4.293745156007438,
      "tokens_seen": 224329728
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00047071213640922775,
      "loss": 3.4523,
      "theoretical_loss": 4.293591005012228,
      "tokens_seen": 224395264
    },
    {
      "epoch": 0.07,
      "objective/train/advantage_avg": 0.49481630325317383,
      "objective/train/docs_used": 138458,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.2242019176483154,
      "objective/train/original_loss": 3.2242014408111572,
      "objective/train/theoretical_loss": 4.29343691163279,
      "objective/train/tokens_used": 244920800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2449270486831665,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507267713546753,
      "objective/train/weighted_lm_loss": 3.3876562118530273,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0396549701690674,
      "theoretical_loss": 4.29343691163279,
      "tokens_seen": 224460800
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004707021063189569,
      "loss": 3.4619,
      "theoretical_loss": 4.29343691163279,
      "tokens_seen": 224460800
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004706920762286861,
      "loss": 3.1875,
      "theoretical_loss": 4.293282875830776,
      "tokens_seen": 224526336
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00047068204613841524,
      "loss": 3.3541,
      "theoretical_loss": 4.293128897567877,
      "tokens_seen": 224591872
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00047067201604814447,
      "loss": 3.2303,
      "theoretical_loss": 4.292974976805818,
      "tokens_seen": 224657408
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00047066198595787365,
      "loss": 3.132,
      "theoretical_loss": 4.29282111350636,
      "tokens_seen": 224722944
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00047065195586760283,
      "loss": 3.4509,
      "theoretical_loss": 4.292667307631305,
      "tokens_seen": 224788480
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.000470641925777332,
      "loss": 3.2558,
      "theoretical_loss": 4.292513559142487,
      "tokens_seen": 224854016
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004706318956870612,
      "loss": 3.2995,
      "theoretical_loss": 4.292359868001779,
      "tokens_seen": 224919552
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004706218655967904,
      "loss": 3.5248,
      "theoretical_loss": 4.29220623417109,
      "tokens_seen": 224985088
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004706118355065196,
      "loss": 3.3535,
      "theoretical_loss": 4.2920526576123645,
      "tokens_seen": 225050624
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00047060180541624874,
      "loss": 3.4932,
      "theoretical_loss": 4.291899138287585,
      "tokens_seen": 225116160
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.000470591775325978,
      "loss": 3.5047,
      "theoretical_loss": 4.291745676158767,
      "tokens_seen": 225181696
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004705817452357071,
      "loss": 3.4136,
      "theoretical_loss": 4.291592271187967,
      "tokens_seen": 225247232
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00047057171514543634,
      "loss": 3.5736,
      "theoretical_loss": 4.291438923337274,
      "tokens_seen": 225312768
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004705616850551655,
      "loss": 3.3302,
      "theoretical_loss": 4.291285632568814,
      "tokens_seen": 225378304
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004705516549648947,
      "loss": 3.289,
      "theoretical_loss": 4.291132398844749,
      "tokens_seen": 225443840
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004705416248746239,
      "loss": 3.3868,
      "theoretical_loss": 4.29097922212728,
      "tokens_seen": 225509376
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004705315947843531,
      "loss": 3.1854,
      "theoretical_loss": 4.290826102378639,
      "tokens_seen": 225574912
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00047052156469408224,
      "loss": 3.3155,
      "theoretical_loss": 4.290673039561098,
      "tokens_seen": 225640448
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004705115346038115,
      "loss": 3.4667,
      "theoretical_loss": 4.290520033636963,
      "tokens_seen": 225705984
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004705015045135406,
      "loss": 3.2479,
      "theoretical_loss": 4.290367084568576,
      "tokens_seen": 225771520
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00047049147442326984,
      "loss": 3.3415,
      "theoretical_loss": 4.290214192318317,
      "tokens_seen": 225837056
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.000470481444332999,
      "loss": 3.5547,
      "theoretical_loss": 4.290061356848597,
      "tokens_seen": 225902592
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004704714142427282,
      "loss": 2.9668,
      "theoretical_loss": 4.289908578121868,
      "tokens_seen": 225968128
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004704613841524574,
      "loss": 3.4885,
      "theoretical_loss": 4.289755856100615,
      "tokens_seen": 226033664
    },
    {
      "epoch": 0.07,
      "objective/train/advantage_avg": 0.4939148426055908,
      "objective/train/docs_used": 139641,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.300494432449341,
      "objective/train/original_loss": 3.300494432449341,
      "objective/train/theoretical_loss": 4.289603190747359,
      "objective/train/tokens_used": 246559200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24403415620326996,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050632119178772,
      "objective/train/weighted_lm_loss": 3.467542886734009,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0417574644088745,
      "theoretical_loss": 4.289603190747359,
      "tokens_seen": 226099200
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00047045135406218656,
      "loss": 3.2944,
      "theoretical_loss": 4.289603190747359,
      "tokens_seen": 226099200
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00047044132397191575,
      "loss": 2.9991,
      "theoretical_loss": 4.2894505820246565,
      "tokens_seen": 226164736
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.000470431293881645,
      "loss": 3.2306,
      "theoretical_loss": 4.2892980298951,
      "tokens_seen": 226230272
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004704212637913741,
      "loss": 3.2607,
      "theoretical_loss": 4.289145534321319,
      "tokens_seen": 226295808
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00047041123370110334,
      "loss": 3.5875,
      "theoretical_loss": 4.288993095265975,
      "tokens_seen": 226361344
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00047040120361083247,
      "loss": 3.2208,
      "theoretical_loss": 4.288840712691767,
      "tokens_seen": 226426880
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004703911735205617,
      "loss": 3.2337,
      "theoretical_loss": 4.2886883865614305,
      "tokens_seen": 226492416
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004703811434302909,
      "loss": 3.4325,
      "theoretical_loss": 4.288536116837733,
      "tokens_seen": 226557952
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00047037111334002007,
      "loss": 3.3103,
      "theoretical_loss": 4.288383903483481,
      "tokens_seen": 226623488
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00047036108324974925,
      "loss": 3.398,
      "theoretical_loss": 4.288231746461515,
      "tokens_seen": 226689024
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004703510531594785,
      "loss": 3.1812,
      "theoretical_loss": 4.28807964573471,
      "tokens_seen": 226754560
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004703410230692076,
      "loss": 3.5536,
      "theoretical_loss": 4.287927601265976,
      "tokens_seen": 226820096
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00047033099297893685,
      "loss": 3.3938,
      "theoretical_loss": 4.28777561301826,
      "tokens_seen": 226885632
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.000470320962888666,
      "loss": 3.5688,
      "theoretical_loss": 4.287623680954542,
      "tokens_seen": 226951168
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004703109327983952,
      "loss": 3.4441,
      "theoretical_loss": 4.287471805037837,
      "tokens_seen": 227016704
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004703009027081244,
      "loss": 3.4443,
      "theoretical_loss": 4.287319985231198,
      "tokens_seen": 227082240
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00047029087261785357,
      "loss": 3.5409,
      "theoretical_loss": 4.287168221497709,
      "tokens_seen": 227147776
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00047028084252758275,
      "loss": 3.416,
      "theoretical_loss": 4.287016513800492,
      "tokens_seen": 227213312
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00047027081243731193,
      "loss": 3.4304,
      "theoretical_loss": 4.2868648621027035,
      "tokens_seen": 227278848
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004702607823470411,
      "loss": 3.4447,
      "theoretical_loss": 4.286713266367532,
      "tokens_seen": 227344384
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00047025075225677035,
      "loss": 3.2678,
      "theoretical_loss": 4.2865617265582046,
      "tokens_seen": 227409920
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004702407221664995,
      "loss": 3.2462,
      "theoretical_loss": 4.28641024263798,
      "tokens_seen": 227475456
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004702306920762287,
      "loss": 3.4649,
      "theoretical_loss": 4.286258814570154,
      "tokens_seen": 227540992
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00047022066198595784,
      "loss": 3.3256,
      "theoretical_loss": 4.286107442318055,
      "tokens_seen": 227606528
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004702106318956871,
      "loss": 3.3261,
      "theoretical_loss": 4.28595612584505,
      "tokens_seen": 227672064
    },
    {
      "epoch": 0.07,
      "objective/train/advantage_avg": 0.4902028441429138,
      "objective/train/docs_used": 140376,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.3041958808898926,
      "objective/train/original_loss": 3.3041951656341553,
      "objective/train/theoretical_loss": 4.2858048651145335,
      "objective/train/tokens_used": 248197600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2405281811952591,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0502429008483887,
      "objective/train/weighted_lm_loss": 3.4697442054748535,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0408108234405518,
      "theoretical_loss": 4.2858048651145335,
      "tokens_seen": 227737600
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00047020060180541626,
      "loss": 3.2194,
      "theoretical_loss": 4.2858048651145335,
      "tokens_seen": 227737600
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00047019057171514544,
      "loss": 3.5497,
      "theoretical_loss": 4.285653660089942,
      "tokens_seen": 227803136
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004701805416248746,
      "loss": 3.2976,
      "theoretical_loss": 4.285502510734742,
      "tokens_seen": 227868672
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00047017051153460385,
      "loss": 3.4235,
      "theoretical_loss": 4.285351417012436,
      "tokens_seen": 227934208
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.000470160481444333,
      "loss": 3.3335,
      "theoretical_loss": 4.285200378886559,
      "tokens_seen": 227999744
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004701504513540622,
      "loss": 3.3827,
      "theoretical_loss": 4.2850493963206855,
      "tokens_seen": 228065280
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00047014042126379134,
      "loss": 3.5547,
      "theoretical_loss": 4.284898469278417,
      "tokens_seen": 228130816
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004701303911735206,
      "loss": 3.3879,
      "theoretical_loss": 4.2847475977233955,
      "tokens_seen": 228196352
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00047012036108324976,
      "loss": 3.2711,
      "theoretical_loss": 4.284596781619294,
      "tokens_seen": 228261888
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00047011033099297894,
      "loss": 3.4474,
      "theoretical_loss": 4.284446020929821,
      "tokens_seen": 228327424
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004701003009027081,
      "loss": 3.2504,
      "theoretical_loss": 4.284295315618719,
      "tokens_seen": 228392960
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004700902708124373,
      "loss": 3.4807,
      "theoretical_loss": 4.284144665649762,
      "tokens_seen": 228458496
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004700802407221665,
      "loss": 3.2243,
      "theoretical_loss": 4.283994070986764,
      "tokens_seen": 228524032
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004700702106318957,
      "loss": 3.4175,
      "theoretical_loss": 4.283843531593567,
      "tokens_seen": 228589568
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00047006018054162484,
      "loss": 3.2813,
      "theoretical_loss": 4.283693047434049,
      "tokens_seen": 228655104
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004700501504513541,
      "loss": 3.3399,
      "theoretical_loss": 4.283542618472125,
      "tokens_seen": 228720640
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00047004012036108326,
      "loss": 3.2043,
      "theoretical_loss": 4.283392244671739,
      "tokens_seen": 228786176
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00047003009027081244,
      "loss": 3.2642,
      "theoretical_loss": 4.2832419259968715,
      "tokens_seen": 228851712
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004700200601805416,
      "loss": 3.4377,
      "theoretical_loss": 4.283091662411537,
      "tokens_seen": 228917248
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004700100300902708,
      "loss": 3.4675,
      "theoretical_loss": 4.282941453879784,
      "tokens_seen": 228982784
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00047,
      "loss": 3.3341,
      "theoretical_loss": 4.282791300365692,
      "tokens_seen": 229048320
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004699899699097292,
      "loss": 3.3406,
      "theoretical_loss": 4.2826412018333775,
      "tokens_seen": 229113856
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00046997993981945835,
      "loss": 3.1815,
      "theoretical_loss": 4.282491158246989,
      "tokens_seen": 229179392
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004699699097291876,
      "loss": 3.5106,
      "theoretical_loss": 4.28234116957071,
      "tokens_seen": 229244928
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00046995987963891676,
      "loss": 3.2365,
      "theoretical_loss": 4.282191235768756,
      "tokens_seen": 229310464
    },
    {
      "epoch": 0.07,
      "objective/train/advantage_avg": 0.49467551708221436,
      "objective/train/docs_used": 141693,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.2104620933532715,
      "objective/train/original_loss": 3.2104623317718506,
      "objective/train/theoretical_loss": 4.282041356805376,
      "objective/train/tokens_used": 249836000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24477003514766693,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050711750984192,
      "objective/train/weighted_lm_loss": 3.3732564449310303,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0338950157165527,
      "theoretical_loss": 4.282041356805376,
      "tokens_seen": 229376000
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00046994984954864595,
      "loss": 3.4146,
      "theoretical_loss": 4.282041356805376,
      "tokens_seen": 229376000
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004699398194583752,
      "loss": 3.4452,
      "theoretical_loss": 4.2818915326448534,
      "tokens_seen": 229441536
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004699297893681043,
      "loss": 3.3656,
      "theoretical_loss": 4.281741763251505,
      "tokens_seen": 229507072
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00046991975927783354,
      "loss": 3.4781,
      "theoretical_loss": 4.281592048589681,
      "tokens_seen": 229572608
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00046990972918756267,
      "loss": 3.1911,
      "theoretical_loss": 4.281442388623764,
      "tokens_seen": 229638144
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004698996990972919,
      "loss": 3.5733,
      "theoretical_loss": 4.281292783318172,
      "tokens_seen": 229703680
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004698896690070211,
      "loss": 3.3795,
      "theoretical_loss": 4.281143232637353,
      "tokens_seen": 229769216
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00046987963891675027,
      "loss": 3.7265,
      "theoretical_loss": 4.280993736545792,
      "tokens_seen": 229834752
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00046986960882647945,
      "loss": 3.2181,
      "theoretical_loss": 4.280844295008004,
      "tokens_seen": 229900288
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004698595787362087,
      "loss": 3.1506,
      "theoretical_loss": 4.280694907988541,
      "tokens_seen": 229965824
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004698495486459378,
      "loss": 3.6936,
      "theoretical_loss": 4.280545575451983,
      "tokens_seen": 230031360
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00046983951855566705,
      "loss": 3.1975,
      "theoretical_loss": 4.280396297362948,
      "tokens_seen": 230096896
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004698294884653962,
      "loss": 3.1777,
      "theoretical_loss": 4.280247073686083,
      "tokens_seen": 230162432
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004698194583751254,
      "loss": 3.4917,
      "theoretical_loss": 4.280097904386071,
      "tokens_seen": 230227968
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004698094282848546,
      "loss": 3.5011,
      "theoretical_loss": 4.279948789427628,
      "tokens_seen": 230293504
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00046979939819458377,
      "loss": 3.3982,
      "theoretical_loss": 4.2797997287755,
      "tokens_seen": 230359040
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00046978936810431295,
      "loss": 3.1558,
      "theoretical_loss": 4.279650722394468,
      "tokens_seen": 230424576
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00046977933801404213,
      "loss": 3.1821,
      "theoretical_loss": 4.279501770249346,
      "tokens_seen": 230490112
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004697693079237713,
      "loss": 3.1235,
      "theoretical_loss": 4.27935287230498,
      "tokens_seen": 230555648
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00046975927783350055,
      "loss": 3.4949,
      "theoretical_loss": 4.279204028526249,
      "tokens_seen": 230621184
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004697492477432297,
      "loss": 3.7009,
      "theoretical_loss": 4.279055238878065,
      "tokens_seen": 230686720
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004697392176529589,
      "loss": 3.2573,
      "theoretical_loss": 4.2789065033253735,
      "tokens_seen": 230752256
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00046972918756268804,
      "loss": 3.5087,
      "theoretical_loss": 4.278757821833151,
      "tokens_seen": 230817792
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004697191574724173,
      "loss": 3.2496,
      "theoretical_loss": 4.278609194366406,
      "tokens_seen": 230883328
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00046970912738214646,
      "loss": 3.2387,
      "theoretical_loss": 4.278460620890183,
      "tokens_seen": 230948864
    },
    {
      "epoch": 0.07,
      "objective/train/advantage_avg": 0.4933275878429413,
      "objective/train/docs_used": 142407,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.8157331943511963,
      "objective/train/original_loss": 2.8157334327697754,
      "objective/train/theoretical_loss": 4.278312101369555,
      "objective/train/tokens_used": 251474400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24345888197422028,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505702495574951,
      "objective/train/weighted_lm_loss": 2.9582557678222656,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 4.278312101369555,
      "tokens_seen": 231014400
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00046969909729187564,
      "loss": 3.1723,
      "theoretical_loss": 4.278312101369555,
      "tokens_seen": 231014400
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004696890672016048,
      "loss": 3.1882,
      "theoretical_loss": 4.278163635769631,
      "tokens_seen": 231079936
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00046967903711133405,
      "loss": 3.1959,
      "theoretical_loss": 4.27801522405555,
      "tokens_seen": 231145472
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004696690070210632,
      "loss": 3.2735,
      "theoretical_loss": 4.277866866192484,
      "tokens_seen": 231211008
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004696589769307924,
      "loss": 3.0868,
      "theoretical_loss": 4.277718562145639,
      "tokens_seen": 231276544
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00046964894684052154,
      "loss": 3.0481,
      "theoretical_loss": 4.27757031188025,
      "tokens_seen": 231342080
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004696389167502508,
      "loss": 3.3166,
      "theoretical_loss": 4.277422115361588,
      "tokens_seen": 231407616
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00046962888665997996,
      "loss": 3.252,
      "theoretical_loss": 4.277273972554954,
      "tokens_seen": 231473152
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00046961885656970914,
      "loss": 3.1399,
      "theoretical_loss": 4.277125883425682,
      "tokens_seen": 231538688
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004696088264794383,
      "loss": 3.1629,
      "theoretical_loss": 4.276977847939136,
      "tokens_seen": 231604224
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004695987963891675,
      "loss": 3.2517,
      "theoretical_loss": 4.276829866060717,
      "tokens_seen": 231669760
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004695887662988967,
      "loss": 3.2012,
      "theoretical_loss": 4.276681937755853,
      "tokens_seen": 231735296
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004695787362086259,
      "loss": 3.203,
      "theoretical_loss": 4.276534062990007,
      "tokens_seen": 231800832
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00046956870611835505,
      "loss": 3.2261,
      "theoretical_loss": 4.276386241728673,
      "tokens_seen": 231866368
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004695586760280843,
      "loss": 3.3395,
      "theoretical_loss": 4.2762384739373775,
      "tokens_seen": 231931904
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00046954864593781346,
      "loss": 3.2537,
      "theoretical_loss": 4.276090759581678,
      "tokens_seen": 231997440
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00046953861584754264,
      "loss": 3.0909,
      "theoretical_loss": 4.275943098627166,
      "tokens_seen": 232062976
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004695285857572718,
      "loss": 3.4288,
      "theoretical_loss": 4.275795491039462,
      "tokens_seen": 232128512
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.000469518555667001,
      "loss": 3.4288,
      "theoretical_loss": 4.27564793678422,
      "tokens_seen": 232194048
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004695085255767302,
      "loss": 3.3699,
      "theoretical_loss": 4.2755004358271265,
      "tokens_seen": 232259584
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004694984954864594,
      "loss": 2.9048,
      "theoretical_loss": 4.275352988133898,
      "tokens_seen": 232325120
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00046948846539618855,
      "loss": 3.3693,
      "theoretical_loss": 4.275205593670284,
      "tokens_seen": 232390656
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004694784353059178,
      "loss": 3.4159,
      "theoretical_loss": 4.275058252402065,
      "tokens_seen": 232456192
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004694684052156469,
      "loss": 3.0428,
      "theoretical_loss": 4.274910964295053,
      "tokens_seen": 232521728
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00046945837512537615,
      "loss": 3.2135,
      "theoretical_loss": 4.274763729315092,
      "tokens_seen": 232587264
    },
    {
      "epoch": 0.07,
      "objective/train/advantage_avg": 0.49441030621528625,
      "objective/train/docs_used": 143093,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.16507625579834,
      "objective/train/original_loss": 3.165076732635498,
      "objective/train/theoretical_loss": 4.274616547428058,
      "objective/train/tokens_used": 253112800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24450406432151794,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506839752197266,
      "objective/train/weighted_lm_loss": 3.325345993041992,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0363596677780151,
      "theoretical_loss": 4.274616547428058,
      "tokens_seen": 232652800
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004694483450351053,
      "loss": 3.4222,
      "theoretical_loss": 4.274616547428058,
      "tokens_seen": 232652800
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004694383149448345,
      "loss": 3.3846,
      "theoretical_loss": 4.274469418599858,
      "tokens_seen": 232718336
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004694282848545637,
      "loss": 3.2363,
      "theoretical_loss": 4.274322342796429,
      "tokens_seen": 232783872
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00046941825476429287,
      "loss": 3.3444,
      "theoretical_loss": 4.274175319983744,
      "tokens_seen": 232849408
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00046940822467402205,
      "loss": 3.0549,
      "theoretical_loss": 4.274028350127802,
      "tokens_seen": 232914944
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004693981945837513,
      "loss": 3.4953,
      "theoretical_loss": 4.273881433194635,
      "tokens_seen": 232980480
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004693881644934804,
      "loss": 3.001,
      "theoretical_loss": 4.27373456915031,
      "tokens_seen": 233046016
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00046937813440320965,
      "loss": 3.3988,
      "theoretical_loss": 4.27358775796092,
      "tokens_seen": 233111552
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00046936810431293883,
      "loss": 3.0585,
      "theoretical_loss": 4.273440999592593,
      "tokens_seen": 233177088
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.000469358074222668,
      "loss": 3.2517,
      "theoretical_loss": 4.273294294011485,
      "tokens_seen": 233242624
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004693480441323972,
      "loss": 3.2327,
      "theoretical_loss": 4.273147641183788,
      "tokens_seen": 233308160
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004693380140421264,
      "loss": 3.4285,
      "theoretical_loss": 4.2730010410757195,
      "tokens_seen": 233373696
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00046932798395185555,
      "loss": 3.3685,
      "theoretical_loss": 4.272854493653532,
      "tokens_seen": 233439232
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004693179538615848,
      "loss": 3.3908,
      "theoretical_loss": 4.272707998883508,
      "tokens_seen": 233504768
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004693079237713139,
      "loss": 3.0362,
      "theoretical_loss": 4.27256155673196,
      "tokens_seen": 233570304
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00046929789368104315,
      "loss": 3.077,
      "theoretical_loss": 4.272415167165232,
      "tokens_seen": 233635840
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004692878635907723,
      "loss": 3.3826,
      "theoretical_loss": 4.272268830149702,
      "tokens_seen": 233701376
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004692778335005015,
      "loss": 3.3459,
      "theoretical_loss": 4.272122545651774,
      "tokens_seen": 233766912
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004692678034102307,
      "loss": 3.5646,
      "theoretical_loss": 4.271976313637885,
      "tokens_seen": 233832448
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004692577733199599,
      "loss": 3.3322,
      "theoretical_loss": 4.271830134074504,
      "tokens_seen": 233897984
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00046924774322968906,
      "loss": 3.2407,
      "theoretical_loss": 4.27168400692813,
      "tokens_seen": 233963520
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00046923771313941824,
      "loss": 3.3557,
      "theoretical_loss": 4.271537932165291,
      "tokens_seen": 234029056
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004692276830491474,
      "loss": 3.1518,
      "theoretical_loss": 4.27139190975255,
      "tokens_seen": 234094592
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00046921765295887666,
      "loss": 3.4586,
      "theoretical_loss": 4.271245939656495,
      "tokens_seen": 234160128
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00046920762286860584,
      "loss": 3.6006,
      "theoretical_loss": 4.2711000218437505,
      "tokens_seen": 234225664
    },
    {
      "epoch": 0.07,
      "objective/train/advantage_avg": 0.49478384852409363,
      "objective/train/docs_used": 144323,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.224773645401001,
      "objective/train/original_loss": 3.224773645401001,
      "objective/train/theoretical_loss": 4.2709541562809665,
      "objective/train/tokens_used": 254751200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2448880821466446,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507233142852783,
      "objective/train/weighted_lm_loss": 3.3881731033325195,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0363596677780151,
      "theoretical_loss": 4.2709541562809665,
      "tokens_seen": 234291200
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.000469197592778335,
      "loss": 3.4748,
      "theoretical_loss": 4.2709541562809665,
      "tokens_seen": 234291200
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00046918756268806425,
      "loss": 3.4187,
      "theoretical_loss": 4.270808342934827,
      "tokens_seen": 234356736
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004691775325977934,
      "loss": 3.3383,
      "theoretical_loss": 4.270662581772045,
      "tokens_seen": 234422272
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004691675025075226,
      "loss": 3.3838,
      "theoretical_loss": 4.270516872759366,
      "tokens_seen": 234487808
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00046915747241725174,
      "loss": 3.7077,
      "theoretical_loss": 4.2703712158635625,
      "tokens_seen": 234553344
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.000469147442326981,
      "loss": 3.4286,
      "theoretical_loss": 4.2702256110514405,
      "tokens_seen": 234618880
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00046913741223671016,
      "loss": 3.2508,
      "theoretical_loss": 4.270080058289835,
      "tokens_seen": 234684416
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00046912738214643934,
      "loss": 3.3192,
      "theoretical_loss": 4.269934557545612,
      "tokens_seen": 234749952
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004691173520561685,
      "loss": 3.3326,
      "theoretical_loss": 4.269789108785667,
      "tokens_seen": 234815488
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004691073219658977,
      "loss": 3.4892,
      "theoretical_loss": 4.269643711976926,
      "tokens_seen": 234881024
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004690972918756269,
      "loss": 3.2419,
      "theoretical_loss": 4.269498367086347,
      "tokens_seen": 234946560
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004690872617853561,
      "loss": 3.2865,
      "theoretical_loss": 4.269353074080915,
      "tokens_seen": 235012096
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00046907723169508525,
      "loss": 3.3891,
      "theoretical_loss": 4.269207832927649,
      "tokens_seen": 235077632
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004690672016048145,
      "loss": 3.1921,
      "theoretical_loss": 4.269062643593594,
      "tokens_seen": 235143168
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00046905717151454366,
      "loss": 3.21,
      "theoretical_loss": 4.268917506045829,
      "tokens_seen": 235208704
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00046904714142427284,
      "loss": 3.2401,
      "theoretical_loss": 4.268772420251461,
      "tokens_seen": 235274240
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.000469037111334002,
      "loss": 3.3975,
      "theoretical_loss": 4.268627386177627,
      "tokens_seen": 235339776
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004690270812437312,
      "loss": 3.4467,
      "theoretical_loss": 4.268482403791495,
      "tokens_seen": 235405312
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004690170511534604,
      "loss": 3.3419,
      "theoretical_loss": 4.268337473060262,
      "tokens_seen": 235470848
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004690070210631896,
      "loss": 3.4152,
      "theoretical_loss": 4.268192593951154,
      "tokens_seen": 235536384
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00046899699097291875,
      "loss": 3.3525,
      "theoretical_loss": 4.2680477664314305,
      "tokens_seen": 235601920
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.000468986960882648,
      "loss": 3.158,
      "theoretical_loss": 4.267902990468377,
      "tokens_seen": 235667456
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004689769307923771,
      "loss": 3.1524,
      "theoretical_loss": 4.267758266029312,
      "tokens_seen": 235732992
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00046896690070210635,
      "loss": 3.2339,
      "theoretical_loss": 4.267613593081581,
      "tokens_seen": 235798528
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00046895687061183553,
      "loss": 3.4241,
      "theoretical_loss": 4.26746897159256,
      "tokens_seen": 235864064
    },
    {
      "epoch": 0.07,
      "objective/train/advantage_avg": 0.49462002515792847,
      "objective/train/docs_used": 145084,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.2993507385253906,
      "objective/train/original_loss": 3.2993507385253906,
      "objective/train/theoretical_loss": 4.267324401529657,
      "objective/train/tokens_used": 256389600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24468524754047394,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507057905197144,
      "objective/train/weighted_lm_loss": 3.4666106700897217,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0457526445388794,
      "theoretical_loss": 4.267324401529657,
      "tokens_seen": 235929600
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004689468405215647,
      "loss": 3.5284,
      "theoretical_loss": 4.267324401529657,
      "tokens_seen": 235929600
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004689368104312939,
      "loss": 3.3279,
      "theoretical_loss": 4.267179882860306,
      "tokens_seen": 235995136
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00046892678034102307,
      "loss": 3.0509,
      "theoretical_loss": 4.267035415551975,
      "tokens_seen": 236060672
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00046891675025075225,
      "loss": 3.1341,
      "theoretical_loss": 4.266890999572157,
      "tokens_seen": 236126208
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004689067201604815,
      "loss": 3.3216,
      "theoretical_loss": 4.266746634888377,
      "tokens_seen": 236191744
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004688966900702106,
      "loss": 3.2114,
      "theoretical_loss": 4.26660232146819,
      "tokens_seen": 236257280
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00046888665997993985,
      "loss": 3.4073,
      "theoretical_loss": 4.26645805927918,
      "tokens_seen": 236322816
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00046887662988966903,
      "loss": 3.4481,
      "theoretical_loss": 4.266313848288959,
      "tokens_seen": 236388352
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004688665997993982,
      "loss": 3.1932,
      "theoretical_loss": 4.266169688465172,
      "tokens_seen": 236453888
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004688565697091274,
      "loss": 3.1142,
      "theoretical_loss": 4.266025579775489,
      "tokens_seen": 236519424
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004688465396188566,
      "loss": 3.3174,
      "theoretical_loss": 4.265881522187613,
      "tokens_seen": 236584960
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00046883650952858575,
      "loss": 3.2869,
      "theoretical_loss": 4.265737515669274,
      "tokens_seen": 236650496
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.000468826479438315,
      "loss": 3.1362,
      "theoretical_loss": 4.2655935601882335,
      "tokens_seen": 236716032
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004688164493480441,
      "loss": 3.0741,
      "theoretical_loss": 4.265449655712279,
      "tokens_seen": 236781568
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00046880641925777335,
      "loss": 3.2866,
      "theoretical_loss": 4.265305802209231,
      "tokens_seen": 236847104
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004687963891675025,
      "loss": 3.4095,
      "theoretical_loss": 4.265161999646936,
      "tokens_seen": 236912640
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004687863590772317,
      "loss": 2.9752,
      "theoretical_loss": 4.265018247993272,
      "tokens_seen": 236978176
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004687763289869609,
      "loss": 2.974,
      "theoretical_loss": 4.264874547216145,
      "tokens_seen": 237043712
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004687662988966901,
      "loss": 3.2319,
      "theoretical_loss": 4.26473089728349,
      "tokens_seen": 237109248
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00046875626880641926,
      "loss": 3.2241,
      "theoretical_loss": 4.264587298163272,
      "tokens_seen": 237174784
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00046874623871614844,
      "loss": 3.1381,
      "theoretical_loss": 4.264443749823484,
      "tokens_seen": 237240320
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004687362086258776,
      "loss": 3.1906,
      "theoretical_loss": 4.2643002522321485,
      "tokens_seen": 237305856
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00046872617853560686,
      "loss": 3.0523,
      "theoretical_loss": 4.264156805357317,
      "tokens_seen": 237371392
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.000468716148445336,
      "loss": 3.2512,
      "theoretical_loss": 4.264013409167069,
      "tokens_seen": 237436928
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004687061183550652,
      "loss": 3.0678,
      "theoretical_loss": 4.263870063629515,
      "tokens_seen": 237502464
    },
    {
      "epoch": 0.07,
      "objective/train/advantage_avg": 0.4954288899898529,
      "objective/train/docs_used": 146518,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.3830654621124268,
      "objective/train/original_loss": 3.3830654621124268,
      "objective/train/theoretical_loss": 4.263726768712791,
      "objective/train/tokens_used": 258028000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24553626775741577,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507910251617432,
      "objective/train/weighted_lm_loss": 3.554736852645874,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0253151655197144,
      "theoretical_loss": 4.263726768712791,
      "tokens_seen": 237568000
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004686960882647944,
      "loss": 3.2357,
      "theoretical_loss": 4.263726768712791,
      "tokens_seen": 237568000
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004686860581745236,
      "loss": 3.3957,
      "theoretical_loss": 4.263583524385067,
      "tokens_seen": 237633536
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00046867602808425276,
      "loss": 3.4975,
      "theoretical_loss": 4.263440330614536,
      "tokens_seen": 237699072
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00046866599799398194,
      "loss": 2.9859,
      "theoretical_loss": 4.263297187369423,
      "tokens_seen": 237764608
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004686559679037111,
      "loss": 3.3981,
      "theoretical_loss": 4.263154094617981,
      "tokens_seen": 237830144
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00046864593781344036,
      "loss": 3.3527,
      "theoretical_loss": 4.263011052328492,
      "tokens_seen": 237895680
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004686359077231695,
      "loss": 3.3028,
      "theoretical_loss": 4.262868060469268,
      "tokens_seen": 237961216
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004686258776328987,
      "loss": 3.2816,
      "theoretical_loss": 4.262725119008646,
      "tokens_seen": 238026752
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00046861584754262785,
      "loss": 3.382,
      "theoretical_loss": 4.262582227914994,
      "tokens_seen": 238092288
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004686058174523571,
      "loss": 3.3408,
      "theoretical_loss": 4.262439387156709,
      "tokens_seen": 238157824
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00046859578736208626,
      "loss": 3.428,
      "theoretical_loss": 4.262296596702216,
      "tokens_seen": 238223360
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00046858575727181545,
      "loss": 3.2878,
      "theoretical_loss": 4.2621538565199675,
      "tokens_seen": 238288896
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004685757271815446,
      "loss": 3.1557,
      "theoretical_loss": 4.262011166578445,
      "tokens_seen": 238354432
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004685656970912738,
      "loss": 3.4059,
      "theoretical_loss": 4.261868526846159,
      "tokens_seen": 238419968
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.000468555667001003,
      "loss": 3.0069,
      "theoretical_loss": 4.261725937291648,
      "tokens_seen": 238485504
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004685456369107322,
      "loss": 3.5388,
      "theoretical_loss": 4.261583397883479,
      "tokens_seen": 238551040
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00046853560682046135,
      "loss": 3.2951,
      "theoretical_loss": 4.261440908590247,
      "tokens_seen": 238616576
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004685255767301906,
      "loss": 3.0463,
      "theoretical_loss": 4.261298469380575,
      "tokens_seen": 238682112
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00046851554663991977,
      "loss": 3.4061,
      "theoretical_loss": 4.261156080223116,
      "tokens_seen": 238747648
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00046850551654964895,
      "loss": 3.4402,
      "theoretical_loss": 4.261013741086549,
      "tokens_seen": 238813184
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00046849548645937813,
      "loss": 3.3689,
      "theoretical_loss": 4.260871451939582,
      "tokens_seen": 238878720
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004684854563691073,
      "loss": 3.1239,
      "theoretical_loss": 4.2607292127509515,
      "tokens_seen": 238944256
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004684754262788365,
      "loss": 3.3653,
      "theoretical_loss": 4.260587023489422,
      "tokens_seen": 239009792
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00046846539618856573,
      "loss": 3.2862,
      "theoretical_loss": 4.260444884123785,
      "tokens_seen": 239075328
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004684553660982949,
      "loss": 3.4851,
      "theoretical_loss": 4.260302794622864,
      "tokens_seen": 239140864
    },
    {
      "epoch": 0.07,
      "objective/train/advantage_avg": 0.4935317039489746,
      "objective/train/docs_used": 147039,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.8737471103668213,
      "objective/train/original_loss": 2.8737471103668213,
      "objective/train/theoretical_loss": 4.260160754955504,
      "objective/train/tokens_used": 259666400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24365365505218506,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505915880203247,
      "objective/train/weighted_lm_loss": 3.019080877304077,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0396549701690674,
      "theoretical_loss": 4.260160754955504,
      "tokens_seen": 239206400
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004684453360080241,
      "loss": 3.2565,
      "theoretical_loss": 4.260160754955504,
      "tokens_seen": 239206400
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00046843530591775327,
      "loss": 3.3962,
      "theoretical_loss": 4.260018765090583,
      "tokens_seen": 239271936
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00046842527582748245,
      "loss": 3.4718,
      "theoretical_loss": 4.259876824997007,
      "tokens_seen": 239337472
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004684152457372117,
      "loss": 3.2368,
      "theoretical_loss": 4.259734934643706,
      "tokens_seen": 239403008
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004684052156469408,
      "loss": 3.4146,
      "theoretical_loss": 4.259593093999641,
      "tokens_seen": 239468544
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00046839518555667005,
      "loss": 3.2613,
      "theoretical_loss": 4.259451303033801,
      "tokens_seen": 239534080
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00046838515546639923,
      "loss": 3.3513,
      "theoretical_loss": 4.259309561715201,
      "tokens_seen": 239599616
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004683751253761284,
      "loss": 3.2175,
      "theoretical_loss": 4.259167870012885,
      "tokens_seen": 239665152
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004683650952858576,
      "loss": 3.2401,
      "theoretical_loss": 4.259026227895926,
      "tokens_seen": 239730688
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004683550651955868,
      "loss": 3.4099,
      "theoretical_loss": 4.258884635333421,
      "tokens_seen": 239796224
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00046834503510531595,
      "loss": 3.3729,
      "theoretical_loss": 4.258743092294498,
      "tokens_seen": 239861760
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004683350050150452,
      "loss": 3.5174,
      "theoretical_loss": 4.258601598748312,
      "tokens_seen": 239927296
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004683249749247743,
      "loss": 3.3796,
      "theoretical_loss": 4.2584601546640455,
      "tokens_seen": 239992832
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00046831494483450355,
      "loss": 3.058,
      "theoretical_loss": 4.258318760010908,
      "tokens_seen": 240058368
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004683049147442327,
      "loss": 3.2449,
      "theoretical_loss": 4.258177414758135,
      "tokens_seen": 240123904
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004682948846539619,
      "loss": 3.5812,
      "theoretical_loss": 4.258036118874995,
      "tokens_seen": 240189440
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004682848545636911,
      "loss": 3.4178,
      "theoretical_loss": 4.257894872330778,
      "tokens_seen": 240254976
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004682748244734203,
      "loss": 3.2164,
      "theoretical_loss": 4.257753675094804,
      "tokens_seen": 240320512
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00046826479438314946,
      "loss": 3.3075,
      "theoretical_loss": 4.2576125271364225,
      "tokens_seen": 240386048
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00046825476429287864,
      "loss": 3.2016,
      "theoretical_loss": 4.257471428425006,
      "tokens_seen": 240451584
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004682447342026078,
      "loss": 3.4854,
      "theoretical_loss": 4.257330378929957,
      "tokens_seen": 240517120
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00046823470411233706,
      "loss": 3.4063,
      "theoretical_loss": 4.257189378620707,
      "tokens_seen": 240582656
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004682246740220662,
      "loss": 3.3915,
      "theoretical_loss": 4.257048427466709,
      "tokens_seen": 240648192
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004682146439317954,
      "loss": 3.3459,
      "theoretical_loss": 4.256907525437451,
      "tokens_seen": 240713728
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004682046138415246,
      "loss": 3.3984,
      "theoretical_loss": 4.256766672502442,
      "tokens_seen": 240779264
    },
    {
      "epoch": 0.07,
      "objective/train/advantage_avg": 0.49348604679107666,
      "objective/train/docs_used": 148118,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.2646536827087402,
      "objective/train/original_loss": 3.2646536827087402,
      "objective/train/theoretical_loss": 4.256625868631222,
      "objective/train/tokens_used": 261304800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2436017543077469,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505868196487427,
      "objective/train/weighted_lm_loss": 3.4298274517059326,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0363596677780151,
      "theoretical_loss": 4.256625868631222,
      "tokens_seen": 240844800
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004681945837512538,
      "loss": 3.1939,
      "theoretical_loss": 4.256625868631222,
      "tokens_seen": 240844800
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00046818455366098296,
      "loss": 3.3981,
      "theoretical_loss": 4.256485113793355,
      "tokens_seen": 240910336
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00046817452357071214,
      "loss": 3.2461,
      "theoretical_loss": 4.2563444079584345,
      "tokens_seen": 240975872
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004681644934804413,
      "loss": 3.2583,
      "theoretical_loss": 4.2562037510960815,
      "tokens_seen": 241041408
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00046815446339017056,
      "loss": 3.3781,
      "theoretical_loss": 4.2560631431759415,
      "tokens_seen": 241106944
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004681444332998997,
      "loss": 3.5868,
      "theoretical_loss": 4.25592258416769,
      "tokens_seen": 241172480
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004681344032096289,
      "loss": 3.4752,
      "theoretical_loss": 4.255782074041028,
      "tokens_seen": 241238016
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00046812437311935805,
      "loss": 3.3679,
      "theoretical_loss": 4.255641612765683,
      "tokens_seen": 241303552
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004681143430290873,
      "loss": 3.4867,
      "theoretical_loss": 4.255501200311411,
      "tokens_seen": 241369088
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00046810431293881646,
      "loss": 3.4168,
      "theoretical_loss": 4.255360836647993,
      "tokens_seen": 241434624
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00046809428284854565,
      "loss": 3.2242,
      "theoretical_loss": 4.255220521745239,
      "tokens_seen": 241500160
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004680842527582748,
      "loss": 3.4282,
      "theoretical_loss": 4.255080255572984,
      "tokens_seen": 241565696
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.000468074222668004,
      "loss": 3.5517,
      "theoretical_loss": 4.254940038101091,
      "tokens_seen": 241631232
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004680641925777332,
      "loss": 3.3082,
      "theoretical_loss": 4.254799869299449,
      "tokens_seen": 241696768
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004680541624874624,
      "loss": 3.3261,
      "theoretical_loss": 4.254659749137975,
      "tokens_seen": 241762304
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00046804413239719155,
      "loss": 3.3206,
      "theoretical_loss": 4.254519677586611,
      "tokens_seen": 241827840
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004680341023069208,
      "loss": 3.4753,
      "theoretical_loss": 4.254379654615327,
      "tokens_seen": 241893376
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00046802407221664997,
      "loss": 3.6213,
      "theoretical_loss": 4.254239680194121,
      "tokens_seen": 241958912
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00046801404212637915,
      "loss": 3.2561,
      "theoretical_loss": 4.254099754293014,
      "tokens_seen": 242024448
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00046800401203610833,
      "loss": 3.3247,
      "theoretical_loss": 4.253959876882057,
      "tokens_seen": 242089984
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004679939819458375,
      "loss": 3.5448,
      "theoretical_loss": 4.253820047931324,
      "tokens_seen": 242155520
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004679839518555667,
      "loss": 3.1996,
      "theoretical_loss": 4.253680267410921,
      "tokens_seen": 242221056
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00046797392176529593,
      "loss": 3.3556,
      "theoretical_loss": 4.253540535290975,
      "tokens_seen": 242286592
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00046796389167502505,
      "loss": 3.3143,
      "theoretical_loss": 4.253400851541643,
      "tokens_seen": 242352128
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004679538615847543,
      "loss": 3.2621,
      "theoretical_loss": 4.253261216133106,
      "tokens_seen": 242417664
    },
    {
      "epoch": 0.07,
      "objective/train/advantage_avg": 0.4953969419002533,
      "objective/train/docs_used": 148826,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.2161221504211426,
      "objective/train/original_loss": 3.2161221504211426,
      "objective/train/theoretical_loss": 4.253121629035574,
      "objective/train/tokens_used": 262943200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24547702074050903,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507876873016357,
      "objective/train/weighted_lm_loss": 3.379395008087158,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0425469875335693,
      "theoretical_loss": 4.253121629035574,
      "tokens_seen": 242483200
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004679438314944834,
      "loss": 3.5222,
      "theoretical_loss": 4.253121629035574,
      "tokens_seen": 242483200
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00046793380140421265,
      "loss": 2.9445,
      "theoretical_loss": 4.252982090219281,
      "tokens_seen": 242548736
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00046792377131394183,
      "loss": 3.4088,
      "theoretical_loss": 4.25284259965449,
      "tokens_seen": 242614272
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.000467913741223671,
      "loss": 3.3383,
      "theoretical_loss": 4.252703157311487,
      "tokens_seen": 242679808
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004679037111334002,
      "loss": 3.3348,
      "theoretical_loss": 4.2525637631605875,
      "tokens_seen": 242745344
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00046789368104312943,
      "loss": 3.3775,
      "theoretical_loss": 4.252424417172132,
      "tokens_seen": 242810880
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00046788365095285856,
      "loss": 3.2562,
      "theoretical_loss": 4.252285119316486,
      "tokens_seen": 242876416
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004678736208625878,
      "loss": 3.3588,
      "theoretical_loss": 4.252145869564044,
      "tokens_seen": 242941952
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004678635907723169,
      "loss": 3.3468,
      "theoretical_loss": 4.252006667885224,
      "tokens_seen": 243007488
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00046785356068204616,
      "loss": 3.3116,
      "theoretical_loss": 4.251867514250472,
      "tokens_seen": 243073024
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00046784353059177534,
      "loss": 3.2502,
      "theoretical_loss": 4.2517284086302585,
      "tokens_seen": 243138560
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004678335005015045,
      "loss": 3.1445,
      "theoretical_loss": 4.251589350995081,
      "tokens_seen": 243204096
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004678234704112337,
      "loss": 3.2868,
      "theoretical_loss": 4.251450341315464,
      "tokens_seen": 243269632
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004678134403209629,
      "loss": 3.2312,
      "theoretical_loss": 4.251311379561956,
      "tokens_seen": 243335168
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00046780341023069206,
      "loss": 3.1203,
      "theoretical_loss": 4.251172465705134,
      "tokens_seen": 243400704
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004677933801404213,
      "loss": 3.2085,
      "theoretical_loss": 4.2510335997155995,
      "tokens_seen": 243466240
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004677833500501504,
      "loss": 3.056,
      "theoretical_loss": 4.250894781563979,
      "tokens_seen": 243531776
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00046777331995987966,
      "loss": 3.1149,
      "theoretical_loss": 4.2507560112209255,
      "tokens_seen": 243597312
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004677632898696088,
      "loss": 3.1976,
      "theoretical_loss": 4.25061728865712,
      "tokens_seen": 243662848
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.000467753259779338,
      "loss": 3.2688,
      "theoretical_loss": 4.250478613843266,
      "tokens_seen": 243728384
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004677432296890672,
      "loss": 3.152,
      "theoretical_loss": 4.250339986750097,
      "tokens_seen": 243793920
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004677331995987964,
      "loss": 3.4187,
      "theoretical_loss": 4.250201407348368,
      "tokens_seen": 243859456
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00046772316950852556,
      "loss": 3.3239,
      "theoretical_loss": 4.250062875608862,
      "tokens_seen": 243924992
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004677131394182548,
      "loss": 3.214,
      "theoretical_loss": 4.249924391502387,
      "tokens_seen": 243990528
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.000467703109327984,
      "loss": 3.225,
      "theoretical_loss": 4.249785954999778,
      "tokens_seen": 244056064
    },
    {
      "epoch": 0.07,
      "objective/train/advantage_avg": 0.4923168122768402,
      "objective/train/docs_used": 150000,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.970130443572998,
      "objective/train/original_loss": 2.970130443572998,
      "objective/train/theoretical_loss": 4.249647566071895,
      "objective/train/tokens_used": 264581600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24254094064235687,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0504645109176636,
      "objective/train/weighted_lm_loss": 3.1196744441986084,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0396549701690674,
      "theoretical_loss": 4.249647566071895,
      "tokens_seen": 244121600
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00046769307923771316,
      "loss": 3.1565,
      "theoretical_loss": 4.249647566071895,
      "tokens_seen": 244121600
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00046768304914744234,
      "loss": 3.3615,
      "theoretical_loss": 4.249509224689621,
      "tokens_seen": 244187136
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004676730190571715,
      "loss": 3.4456,
      "theoretical_loss": 4.24937093082387,
      "tokens_seen": 244252672
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00046766298896690076,
      "loss": 3.1848,
      "theoretical_loss": 4.249232684445579,
      "tokens_seen": 244318208
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004676529588766299,
      "loss": 3.0099,
      "theoretical_loss": 4.249094485525706,
      "tokens_seen": 244383744
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004676429287863591,
      "loss": 3.373,
      "theoretical_loss": 4.248956334035244,
      "tokens_seen": 244449280
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00046763289869608825,
      "loss": 3.286,
      "theoretical_loss": 4.248818229945202,
      "tokens_seen": 244514816
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004676228686058175,
      "loss": 3.2895,
      "theoretical_loss": 4.248680173226621,
      "tokens_seen": 244580352
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00046761283851554666,
      "loss": 3.2611,
      "theoretical_loss": 4.248542163850566,
      "tokens_seen": 244645888
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00046760280842527585,
      "loss": 3.2227,
      "theoretical_loss": 4.248404201788124,
      "tokens_seen": 244711424
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.000467592778335005,
      "loss": 3.441,
      "theoretical_loss": 4.248266287010411,
      "tokens_seen": 244776960
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004675827482447342,
      "loss": 3.0761,
      "theoretical_loss": 4.248128419488569,
      "tokens_seen": 244842496
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004675727181544634,
      "loss": 3.4979,
      "theoretical_loss": 4.247990599193761,
      "tokens_seen": 244908032
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004675626880641926,
      "loss": 3.2665,
      "theoretical_loss": 4.247852826097181,
      "tokens_seen": 244973568
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00046755265797392175,
      "loss": 3.3228,
      "theoretical_loss": 4.247715100170042,
      "tokens_seen": 245039104
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.000467542627883651,
      "loss": 3.4194,
      "theoretical_loss": 4.247577421383589,
      "tokens_seen": 245104640
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00046753259779338017,
      "loss": 3.3108,
      "theoretical_loss": 4.247439789709086,
      "tokens_seen": 245170176
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00046752256770310935,
      "loss": 3.3401,
      "theoretical_loss": 4.247302205117826,
      "tokens_seen": 245235712
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00046751253761283853,
      "loss": 3.132,
      "theoretical_loss": 4.247164667581126,
      "tokens_seen": 245301248
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004675025075225677,
      "loss": 3.3102,
      "theoretical_loss": 4.247027177070329,
      "tokens_seen": 245366784
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004674924774322969,
      "loss": 3.2028,
      "theoretical_loss": 4.2468897335568006,
      "tokens_seen": 245432320
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00046748244734202613,
      "loss": 3.0451,
      "theoretical_loss": 4.246752337011935,
      "tokens_seen": 245497856
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00046747241725175525,
      "loss": 3.3525,
      "theoretical_loss": 4.246614987407148,
      "tokens_seen": 245563392
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004674623871614845,
      "loss": 3.4957,
      "theoretical_loss": 4.246477684713883,
      "tokens_seen": 245628928
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004674523570712136,
      "loss": 3.1549,
      "theoretical_loss": 4.246340428903608,
      "tokens_seen": 245694464
    },
    {
      "epoch": 0.07,
      "objective/train/advantage_avg": 0.4957495629787445,
      "objective/train/docs_used": 150657,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.423431634902954,
      "objective/train/original_loss": 3.423431873321533,
      "objective/train/theoretical_loss": 4.246203219947814,
      "objective/train/tokens_used": 266220000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24582162499427795,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050824522972107,
      "objective/train/weighted_lm_loss": 3.5973122119903564,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0304546356201172,
      "theoretical_loss": 4.246203219947814,
      "tokens_seen": 245760000
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00046744232698094285,
      "loss": 3.491,
      "theoretical_loss": 4.246203219947814,
      "tokens_seen": 245760000
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00046743229689067203,
      "loss": 3.4411,
      "theoretical_loss": 4.24606605781802,
      "tokens_seen": 245825536
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004674222668004012,
      "loss": 3.4744,
      "theoretical_loss": 4.245928942485766,
      "tokens_seen": 245891072
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004674122367101304,
      "loss": 3.0959,
      "theoretical_loss": 4.245791873922621,
      "tokens_seen": 245956608
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00046740220661985963,
      "loss": 3.0578,
      "theoretical_loss": 4.245654852100177,
      "tokens_seen": 246022144
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00046739217652958876,
      "loss": 3.313,
      "theoretical_loss": 4.245517876990049,
      "tokens_seen": 246087680
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.000467382146439318,
      "loss": 3.3342,
      "theoretical_loss": 4.245380948563881,
      "tokens_seen": 246153216
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004673721163490471,
      "loss": 3.198,
      "theoretical_loss": 4.245244066793337,
      "tokens_seen": 246218752
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00046736208625877636,
      "loss": 3.1686,
      "theoretical_loss": 4.24510723165011,
      "tokens_seen": 246284288
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00046735205616850554,
      "loss": 3.229,
      "theoretical_loss": 4.244970443105913,
      "tokens_seen": 246349824
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004673420260782347,
      "loss": 3.3874,
      "theoretical_loss": 4.24483370113249,
      "tokens_seen": 246415360
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004673319959879639,
      "loss": 3.2629,
      "theoretical_loss": 4.244697005701604,
      "tokens_seen": 246480896
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004673219658976931,
      "loss": 3.3455,
      "theoretical_loss": 4.244560356785045,
      "tokens_seen": 246546432
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00046731193580742226,
      "loss": 3.3782,
      "theoretical_loss": 4.244423754354627,
      "tokens_seen": 246611968
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004673019057171515,
      "loss": 3.4181,
      "theoretical_loss": 4.24428719838219,
      "tokens_seen": 246677504
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004672918756268806,
      "loss": 3.3785,
      "theoretical_loss": 4.2441506888395955,
      "tokens_seen": 246743040
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00046728184553660986,
      "loss": 3.0819,
      "theoretical_loss": 4.2440142256987325,
      "tokens_seen": 246808576
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.000467271815446339,
      "loss": 3.5088,
      "theoretical_loss": 4.243877808931515,
      "tokens_seen": 246874112
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004672617853560682,
      "loss": 3.2667,
      "theoretical_loss": 4.243741438509876,
      "tokens_seen": 246939648
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004672517552657974,
      "loss": 3.2659,
      "theoretical_loss": 4.243605114405781,
      "tokens_seen": 247005184
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004672417251755266,
      "loss": 3.1656,
      "theoretical_loss": 4.243468836591212,
      "tokens_seen": 247070720
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00046723169508525576,
      "loss": 3.2326,
      "theoretical_loss": 4.243332605038182,
      "tokens_seen": 247136256
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.000467221664994985,
      "loss": 3.0944,
      "theoretical_loss": 4.243196419718723,
      "tokens_seen": 247201792
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004672116349047141,
      "loss": 3.4673,
      "theoretical_loss": 4.243060280604896,
      "tokens_seen": 247267328
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.00046720160481444336,
      "loss": 3.3794,
      "theoretical_loss": 4.242924187668781,
      "tokens_seen": 247332864
    },
    {
      "epoch": 0.07,
      "objective/train/advantage_avg": 0.49477818608283997,
      "objective/train/docs_used": 151862,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.304267168045044,
      "objective/train/original_loss": 3.304267644882202,
      "objective/train/theoretical_loss": 4.242788140882488,
      "objective/train/tokens_used": 267858400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2448626011610031,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050722599029541,
      "objective/train/weighted_lm_loss": 3.4719183444976807,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0417574644088745,
      "theoretical_loss": 4.242788140882488,
      "tokens_seen": 247398400
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004671915747241725,
      "loss": 3.2108,
      "theoretical_loss": 4.242788140882488,
      "tokens_seen": 247398400
    },
    {
      "epoch": 0.07,
      "learning_rate": 0.0004671815446339017,
      "loss": 3.2124,
      "theoretical_loss": 4.242652140218147,
      "tokens_seen": 247463936
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004671715145436309,
      "loss": 3.0349,
      "theoretical_loss": 4.242516185647914,
      "tokens_seen": 247529472
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004671614844533601,
      "loss": 3.4644,
      "theoretical_loss": 4.242380277143969,
      "tokens_seen": 247595008
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046715145436308927,
      "loss": 3.3872,
      "theoretical_loss": 4.242244414678515,
      "tokens_seen": 247660544
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046714142427281845,
      "loss": 3.3603,
      "theoretical_loss": 4.242108598223782,
      "tokens_seen": 247726080
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046713139418254763,
      "loss": 3.2707,
      "theoretical_loss": 4.24197282775202,
      "tokens_seen": 247791616
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046712136409227686,
      "loss": 3.2769,
      "theoretical_loss": 4.2418371032355076,
      "tokens_seen": 247857152
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.000467111334002006,
      "loss": 3.5404,
      "theoretical_loss": 4.241701424646544,
      "tokens_seen": 247922688
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004671013039117352,
      "loss": 3.2446,
      "theoretical_loss": 4.241565791957452,
      "tokens_seen": 247988224
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046709127382146435,
      "loss": 3.2777,
      "theoretical_loss": 4.241430205140583,
      "tokens_seen": 248053760
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004670812437311936,
      "loss": 3.3769,
      "theoretical_loss": 4.241294664168306,
      "tokens_seen": 248119296
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046707121364092277,
      "loss": 3.3163,
      "theoretical_loss": 4.24115916901302,
      "tokens_seen": 248184832
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046706118355065195,
      "loss": 3.4071,
      "theoretical_loss": 4.241023719647143,
      "tokens_seen": 248250368
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046705115346038113,
      "loss": 3.2341,
      "theoretical_loss": 4.2408883160431206,
      "tokens_seen": 248315904
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046704112337011037,
      "loss": 3.1442,
      "theoretical_loss": 4.240752958173419,
      "tokens_seen": 248381440
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004670310932798395,
      "loss": 3.3236,
      "theoretical_loss": 4.240617646010532,
      "tokens_seen": 248446976
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046702106318956873,
      "loss": 3.2022,
      "theoretical_loss": 4.240482379526973,
      "tokens_seen": 248512512
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046701103309929786,
      "loss": 3.1022,
      "theoretical_loss": 4.240347158695281,
      "tokens_seen": 248578048
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004670010030090271,
      "loss": 3.1834,
      "theoretical_loss": 4.240211983488021,
      "tokens_seen": 248643584
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004669909729187563,
      "loss": 3.6121,
      "theoretical_loss": 4.240076853877778,
      "tokens_seen": 248709120
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046698094282848545,
      "loss": 3.3056,
      "theoretical_loss": 4.239941769837165,
      "tokens_seen": 248774656
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046697091273821464,
      "loss": 3.346,
      "theoretical_loss": 4.2398067313388115,
      "tokens_seen": 248840192
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004669608826479438,
      "loss": 3.3562,
      "theoretical_loss": 4.239671738355379,
      "tokens_seen": 248905728
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046695085255767305,
      "loss": 3.1564,
      "theoretical_loss": 4.239536790859547,
      "tokens_seen": 248971264
    },
    {
      "epoch": 0.08,
      "objective/train/advantage_avg": 0.49605026841163635,
      "objective/train/docs_used": 152583,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.070378541946411,
      "objective/train/original_loss": 3.070378541946411,
      "objective/train/theoretical_loss": 4.2394018888240215,
      "objective/train/tokens_used": 269496800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2461431324481964,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050856351852417,
      "objective/train/weighted_lm_loss": 3.2264676094055176,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0338950157165527,
      "theoretical_loss": 4.2394018888240215,
      "tokens_seen": 249036800
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046694082246740223,
      "loss": 3.2606,
      "theoretical_loss": 4.2394018888240215,
      "tokens_seen": 249036800
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004669307923771314,
      "loss": 3.3154,
      "theoretical_loss": 4.239267032221529,
      "tokens_seen": 249102336
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004669207622868606,
      "loss": 3.2671,
      "theoretical_loss": 4.239132221024823,
      "tokens_seen": 249167872
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046691073219658983,
      "loss": 3.3395,
      "theoretical_loss": 4.238997455206679,
      "tokens_seen": 249233408
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046690070210631896,
      "loss": 3.1609,
      "theoretical_loss": 4.238862734739896,
      "tokens_seen": 249298944
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004668906720160482,
      "loss": 3.2598,
      "theoretical_loss": 4.238728059597294,
      "tokens_seen": 249364480
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004668806419257773,
      "loss": 3.4713,
      "theoretical_loss": 4.238593429751722,
      "tokens_seen": 249430016
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046687061183550656,
      "loss": 3.0671,
      "theoretical_loss": 4.238458845176048,
      "tokens_seen": 249495552
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046686058174523574,
      "loss": 3.2294,
      "theoretical_loss": 4.2383243058431646,
      "tokens_seen": 249561088
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004668505516549649,
      "loss": 3.5044,
      "theoretical_loss": 4.238189811725987,
      "tokens_seen": 249626624
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004668405215646941,
      "loss": 3.2489,
      "theoretical_loss": 4.238055362797455,
      "tokens_seen": 249692160
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004668304914744233,
      "loss": 3.3178,
      "theoretical_loss": 4.2379209590305305,
      "tokens_seen": 249757696
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046682046138415246,
      "loss": 3.6179,
      "theoretical_loss": 4.2377866003982,
      "tokens_seen": 249823232
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004668104312938817,
      "loss": 3.6083,
      "theoretical_loss": 4.237652286873473,
      "tokens_seen": 249888768
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004668004012036108,
      "loss": 3.4591,
      "theoretical_loss": 4.237518018429381,
      "tokens_seen": 249954304
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046679037111334006,
      "loss": 3.0555,
      "theoretical_loss": 4.23738379503898,
      "tokens_seen": 250019840
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004667803410230692,
      "loss": 3.3391,
      "theoretical_loss": 4.2372496166753475,
      "tokens_seen": 250085376
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004667703109327984,
      "loss": 3.1135,
      "theoretical_loss": 4.237115483311586,
      "tokens_seen": 250150912
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004667602808425276,
      "loss": 3.3868,
      "theoretical_loss": 4.23698139492082,
      "tokens_seen": 250216448
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004667502507522568,
      "loss": 3.1747,
      "theoretical_loss": 4.2368473514761975,
      "tokens_seen": 250281984
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046674022066198596,
      "loss": 3.1481,
      "theoretical_loss": 4.23671335295089,
      "tokens_seen": 250347520
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004667301905717152,
      "loss": 3.5506,
      "theoretical_loss": 4.2365793993180905,
      "tokens_seen": 250413056
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004667201604814443,
      "loss": 3.0331,
      "theoretical_loss": 4.2364454905510165,
      "tokens_seen": 250478592
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046671013039117356,
      "loss": 3.3739,
      "theoretical_loss": 4.236311626622909,
      "tokens_seen": 250544128
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004667001003009027,
      "loss": 3.2916,
      "theoretical_loss": 4.23617780750703,
      "tokens_seen": 250609664
    },
    {
      "epoch": 0.08,
      "objective/train/advantage_avg": 0.49638235569000244,
      "objective/train/docs_used": 153242,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.0931203365325928,
      "objective/train/original_loss": 3.0931203365325928,
      "objective/train/theoretical_loss": 4.236044033176665,
      "objective/train/tokens_used": 271135200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2464200109243393,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0508910417556763,
      "objective/train/weighted_lm_loss": 3.25056791305542,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0470744371414185,
      "theoretical_loss": 4.236044033176665,
      "tokens_seen": 250675200
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004666900702106319,
      "loss": 3.2975,
      "theoretical_loss": 4.236044033176665,
      "tokens_seen": 250675200
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004666800401203611,
      "loss": 3.1068,
      "theoretical_loss": 4.235910303605123,
      "tokens_seen": 250740736
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004666700100300903,
      "loss": 3.2529,
      "theoretical_loss": 4.235776618765736,
      "tokens_seen": 250806272
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046665997993981947,
      "loss": 3.2307,
      "theoretical_loss": 4.235642978631859,
      "tokens_seen": 250871808
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046664994984954865,
      "loss": 3.0933,
      "theoretical_loss": 4.2355093831768675,
      "tokens_seen": 250937344
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046663991975927783,
      "loss": 3.42,
      "theoretical_loss": 4.235375832374164,
      "tokens_seen": 251002880
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046662988966900706,
      "loss": 3.3386,
      "theoretical_loss": 4.23524232619717,
      "tokens_seen": 251068416
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004666198595787362,
      "loss": 3.564,
      "theoretical_loss": 4.235108864619331,
      "tokens_seen": 251133952
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046660982948846543,
      "loss": 3.1923,
      "theoretical_loss": 4.234975447614117,
      "tokens_seen": 251199488
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046659979939819455,
      "loss": 3.0921,
      "theoretical_loss": 4.234842075155017,
      "tokens_seen": 251265024
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004665897693079238,
      "loss": 3.2516,
      "theoretical_loss": 4.234708747215546,
      "tokens_seen": 251330560
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046657973921765297,
      "loss": 3.1257,
      "theoretical_loss": 4.23457546376924,
      "tokens_seen": 251396096
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046656970912738215,
      "loss": 3.4249,
      "theoretical_loss": 4.2344422247896585,
      "tokens_seen": 251461632
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046655967903711133,
      "loss": 3.2552,
      "theoretical_loss": 4.234309030250383,
      "tokens_seen": 251527168
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046654964894684057,
      "loss": 3.4067,
      "theoretical_loss": 4.234175880125018,
      "tokens_seen": 251592704
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004665396188565697,
      "loss": 3.8582,
      "theoretical_loss": 4.23404277438719,
      "tokens_seen": 251658240
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046652958876629893,
      "loss": 3.3056,
      "theoretical_loss": 4.233909713010547,
      "tokens_seen": 251723776
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046651955867602806,
      "loss": 3.56,
      "theoretical_loss": 4.233776695968762,
      "tokens_seen": 251789312
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004665095285857573,
      "loss": 3.5756,
      "theoretical_loss": 4.23364372323553,
      "tokens_seen": 251854848
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004664994984954865,
      "loss": 3.1614,
      "theoretical_loss": 4.233510794784567,
      "tokens_seen": 251920384
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046648946840521565,
      "loss": 3.2836,
      "theoretical_loss": 4.233377910589612,
      "tokens_seen": 251985920
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046647943831494484,
      "loss": 3.1099,
      "theoretical_loss": 4.233245070624426,
      "tokens_seen": 252051456
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.000466469408224674,
      "loss": 3.376,
      "theoretical_loss": 4.233112274862793,
      "tokens_seen": 252116992
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004664593781344032,
      "loss": 3.5401,
      "theoretical_loss": 4.232979523278521,
      "tokens_seen": 252182528
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046644934804413243,
      "loss": 3.4294,
      "theoretical_loss": 4.232846815845436,
      "tokens_seen": 252248064
    },
    {
      "epoch": 0.08,
      "objective/train/advantage_avg": 0.4929395318031311,
      "objective/train/docs_used": 154565,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.1743080615997314,
      "objective/train/original_loss": 3.1743085384368896,
      "objective/train/theoretical_loss": 4.232714152537391,
      "objective/train/tokens_used": 272773600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24311085045337677,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505297183990479,
      "objective/train/weighted_lm_loss": 3.3346147537231445,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0304546356201172,
      "theoretical_loss": 4.232714152537391,
      "tokens_seen": 252313600
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046643931795386156,
      "loss": 3.6117,
      "theoretical_loss": 4.232714152537391,
      "tokens_seen": 252313600
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004664292878635908,
      "loss": 3.6231,
      "theoretical_loss": 4.232581533328259,
      "tokens_seen": 252379136
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004664192577733199,
      "loss": 3.2631,
      "theoretical_loss": 4.2324489581919345,
      "tokens_seen": 252444672
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046640922768304916,
      "loss": 3.4514,
      "theoretical_loss": 4.232316427102335,
      "tokens_seen": 252510208
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046639919759277834,
      "loss": 3.1263,
      "theoretical_loss": 4.2321839400334005,
      "tokens_seen": 252575744
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004663891675025075,
      "loss": 3.5181,
      "theoretical_loss": 4.232051496959093,
      "tokens_seen": 252641280
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004663791374122367,
      "loss": 3.3134,
      "theoretical_loss": 4.231919097853398,
      "tokens_seen": 252706816
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046636910732196594,
      "loss": 3.2923,
      "theoretical_loss": 4.2317867426903195,
      "tokens_seen": 252772352
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046635907723169506,
      "loss": 3.4187,
      "theoretical_loss": 4.2316544314438875,
      "tokens_seen": 252837888
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004663490471414243,
      "loss": 3.2353,
      "theoretical_loss": 4.231522164088151,
      "tokens_seen": 252903424
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004663390170511534,
      "loss": 3.2269,
      "theoretical_loss": 4.231389940597185,
      "tokens_seen": 252968960
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046632898696088266,
      "loss": 3.3668,
      "theoretical_loss": 4.231257760945081,
      "tokens_seen": 253034496
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046631895687061184,
      "loss": 3.3583,
      "theoretical_loss": 4.231125625105959,
      "tokens_seen": 253100032
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.000466308926780341,
      "loss": 3.1874,
      "theoretical_loss": 4.230993533053954,
      "tokens_seen": 253165568
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004662988966900702,
      "loss": 3.3424,
      "theoretical_loss": 4.230861484763229,
      "tokens_seen": 253231104
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004662888665997994,
      "loss": 3.3016,
      "theoretical_loss": 4.230729480207964,
      "tokens_seen": 253296640
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046627883650952857,
      "loss": 3.1638,
      "theoretical_loss": 4.230597519362366,
      "tokens_seen": 253362176
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004662688064192578,
      "loss": 3.3478,
      "theoretical_loss": 4.23046560220066,
      "tokens_seen": 253427712
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046625877632898693,
      "loss": 3.1478,
      "theoretical_loss": 4.230333728697093,
      "tokens_seen": 253493248
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046624874623871616,
      "loss": 3.1896,
      "theoretical_loss": 4.2302018988259364,
      "tokens_seen": 253558784
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046623871614844535,
      "loss": 3.2314,
      "theoretical_loss": 4.2300701125614815,
      "tokens_seen": 253624320
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004662286860581745,
      "loss": 3.3436,
      "theoretical_loss": 4.229938369878042,
      "tokens_seen": 253689856
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004662186559679037,
      "loss": 3.2172,
      "theoretical_loss": 4.2298066707499515,
      "tokens_seen": 253755392
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004662086258776329,
      "loss": 3.483,
      "theoretical_loss": 4.229675015151569,
      "tokens_seen": 253820928
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004661985957873621,
      "loss": 3.3471,
      "theoretical_loss": 4.2295434030572725,
      "tokens_seen": 253886464
    },
    {
      "epoch": 0.08,
      "objective/train/advantage_avg": 0.49285799264907837,
      "objective/train/docs_used": 156106,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.3105342388153076,
      "objective/train/original_loss": 3.3105344772338867,
      "objective/train/theoretical_loss": 4.229411834441462,
      "objective/train/tokens_used": 274412000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24298949539661407,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050520896911621,
      "objective/train/weighted_lm_loss": 3.47755765914917,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0432155132293701,
      "theoretical_loss": 4.229411834441462,
      "tokens_seen": 253952000
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004661885656970913,
      "loss": 3.502,
      "theoretical_loss": 4.229411834441462,
      "tokens_seen": 253952000
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004661785356068205,
      "loss": 3.3068,
      "theoretical_loss": 4.229280309278561,
      "tokens_seen": 254017536
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046616850551654967,
      "loss": 3.2905,
      "theoretical_loss": 4.2291488275430105,
      "tokens_seen": 254083072
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046615847542627885,
      "loss": 3.4752,
      "theoretical_loss": 4.229017389209279,
      "tokens_seen": 254148608
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046614844533600803,
      "loss": 3.3739,
      "theoretical_loss": 4.228885994251851,
      "tokens_seen": 254214144
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046613841524573727,
      "loss": 3.5478,
      "theoretical_loss": 4.228754642645235,
      "tokens_seen": 254279680
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004661283851554664,
      "loss": 3.3063,
      "theoretical_loss": 4.228623334363962,
      "tokens_seen": 254345216
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046611835506519563,
      "loss": 3.264,
      "theoretical_loss": 4.228492069382583,
      "tokens_seen": 254410752
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046610832497492475,
      "loss": 3.0904,
      "theoretical_loss": 4.228360847675672,
      "tokens_seen": 254476288
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.000466098294884654,
      "loss": 3.328,
      "theoretical_loss": 4.228229669217823,
      "tokens_seen": 254541824
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046608826479438317,
      "loss": 3.4318,
      "theoretical_loss": 4.228098533983651,
      "tokens_seen": 254607360
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046607823470411235,
      "loss": 3.5538,
      "theoretical_loss": 4.227967441947795,
      "tokens_seen": 254672896
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046606820461384153,
      "loss": 3.2614,
      "theoretical_loss": 4.227836393084912,
      "tokens_seen": 254738432
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046605817452357077,
      "loss": 3.2129,
      "theoretical_loss": 4.227705387369683,
      "tokens_seen": 254803968
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004660481444332999,
      "loss": 3.3059,
      "theoretical_loss": 4.227574424776811,
      "tokens_seen": 254869504
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046603811434302913,
      "loss": 3.3784,
      "theoretical_loss": 4.227443505281016,
      "tokens_seen": 254935040
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046602808425275826,
      "loss": 3.3157,
      "theoretical_loss": 4.227312628857045,
      "tokens_seen": 255000576
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004660180541624875,
      "loss": 3.4744,
      "theoretical_loss": 4.2271817954796616,
      "tokens_seen": 255066112
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004660080240722167,
      "loss": 3.3024,
      "theoretical_loss": 4.227051005123653,
      "tokens_seen": 255131648
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046599799398194586,
      "loss": 3.312,
      "theoretical_loss": 4.226920257763827,
      "tokens_seen": 255197184
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046598796389167504,
      "loss": 3.2539,
      "theoretical_loss": 4.226789553375013,
      "tokens_seen": 255262720
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004659779338014042,
      "loss": 3.2197,
      "theoretical_loss": 4.226658891932061,
      "tokens_seen": 255328256
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004659679037111334,
      "loss": 3.3147,
      "theoretical_loss": 4.226528273409844,
      "tokens_seen": 255393792
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046595787362086263,
      "loss": 3.2953,
      "theoretical_loss": 4.226397697783252,
      "tokens_seen": 255459328
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046594784353059176,
      "loss": 3.1665,
      "theoretical_loss": 4.226267165027202,
      "tokens_seen": 255524864
    },
    {
      "epoch": 0.08,
      "objective/train/advantage_avg": 0.49567723274230957,
      "objective/train/docs_used": 156630,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.560075283050537,
      "objective/train/original_loss": 2.560075283050537,
      "objective/train/theoretical_loss": 4.226136675116626,
      "objective/train/tokens_used": 276050400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24573896825313568,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0508168935775757,
      "objective/train/weighted_lm_loss": 2.690164804458618,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0400675535202026,
      "theoretical_loss": 4.226136675116626,
      "tokens_seen": 255590400
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.000465937813440321,
      "loss": 3.1507,
      "theoretical_loss": 4.226136675116626,
      "tokens_seen": 255590400
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004659277833500501,
      "loss": 3.2344,
      "theoretical_loss": 4.226006228026482,
      "tokens_seen": 255655936
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046591775325977936,
      "loss": 3.3705,
      "theoretical_loss": 4.225875823731746,
      "tokens_seen": 255721472
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046590772316950854,
      "loss": 3.145,
      "theoretical_loss": 4.225745462207417,
      "tokens_seen": 255787008
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004658976930792377,
      "loss": 3.541,
      "theoretical_loss": 4.225615143428513,
      "tokens_seen": 255852544
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004658876629889669,
      "loss": 3.5806,
      "theoretical_loss": 4.225484867370074,
      "tokens_seen": 255918080
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046587763289869614,
      "loss": 3.254,
      "theoretical_loss": 4.225354634007163,
      "tokens_seen": 255983616
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046586760280842526,
      "loss": 3.0685,
      "theoretical_loss": 4.225224443314859,
      "tokens_seen": 256049152
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004658575727181545,
      "loss": 3.4369,
      "theoretical_loss": 4.225094295268267,
      "tokens_seen": 256114688
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004658475426278836,
      "loss": 3.2607,
      "theoretical_loss": 4.22496418984251,
      "tokens_seen": 256180224
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046583751253761286,
      "loss": 3.3356,
      "theoretical_loss": 4.224834127012732,
      "tokens_seen": 256245760
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046582748244734204,
      "loss": 3.062,
      "theoretical_loss": 4.224704106754101,
      "tokens_seen": 256311296
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004658174523570712,
      "loss": 3.0773,
      "theoretical_loss": 4.224574129041801,
      "tokens_seen": 256376832
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004658074222668004,
      "loss": 2.9838,
      "theoretical_loss": 4.224444193851039,
      "tokens_seen": 256442368
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004657973921765296,
      "loss": 3.212,
      "theoretical_loss": 4.224314301157044,
      "tokens_seen": 256507904
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046578736208625877,
      "loss": 3.011,
      "theoretical_loss": 4.224184450935065,
      "tokens_seen": 256573440
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.000465777331995988,
      "loss": 3.29,
      "theoretical_loss": 4.22405464316037,
      "tokens_seen": 256638976
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046576730190571713,
      "loss": 3.3248,
      "theoretical_loss": 4.223924877808249,
      "tokens_seen": 256704512
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046575727181544636,
      "loss": 3.3307,
      "theoretical_loss": 4.223795154854015,
      "tokens_seen": 256770048
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046574724172517555,
      "loss": 3.4802,
      "theoretical_loss": 4.223665474272996,
      "tokens_seen": 256835584
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004657372116349047,
      "loss": 3.4953,
      "theoretical_loss": 4.223535836040548,
      "tokens_seen": 256901120
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004657271815446339,
      "loss": 3.4075,
      "theoretical_loss": 4.223406240132041,
      "tokens_seen": 256966656
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004657171514543631,
      "loss": 3.2489,
      "theoretical_loss": 4.223276686522869,
      "tokens_seen": 257032192
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046570712136409227,
      "loss": 3.0372,
      "theoretical_loss": 4.223147175188446,
      "tokens_seen": 257097728
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004656970912738215,
      "loss": 3.2713,
      "theoretical_loss": 4.223017706104207,
      "tokens_seen": 257163264
    },
    {
      "epoch": 0.08,
      "objective/train/advantage_avg": 0.49346402287483215,
      "objective/train/docs_used": 157258,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.4031970500946045,
      "objective/train/original_loss": 3.4031975269317627,
      "objective/train/theoretical_loss": 4.2228882792456055,
      "objective/train/tokens_used": 277688800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2435728758573532,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505845546722412,
      "objective/train/weighted_lm_loss": 3.5753979682922363,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 4.2228882792456055,
      "tokens_seen": 257228800
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046568706118355063,
      "loss": 3.1164,
      "theoretical_loss": 4.2228882792456055,
      "tokens_seen": 257228800
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046567703109327987,
      "loss": 3.5461,
      "theoretical_loss": 4.222758894588118,
      "tokens_seen": 257294336
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.000465667001003009,
      "loss": 3.3769,
      "theoretical_loss": 4.22262955210724,
      "tokens_seen": 257359872
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046565697091273823,
      "loss": 3.3781,
      "theoretical_loss": 4.222500251778489,
      "tokens_seen": 257425408
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004656469408224674,
      "loss": 3.2854,
      "theoretical_loss": 4.2223709935774005,
      "tokens_seen": 257490944
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004656369107321966,
      "loss": 3.3594,
      "theoretical_loss": 4.222241777479533,
      "tokens_seen": 257556480
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004656268806419258,
      "loss": 3.0136,
      "theoretical_loss": 4.222112603460462,
      "tokens_seen": 257622016
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046561685055165495,
      "loss": 3.2569,
      "theoretical_loss": 4.221983471495789,
      "tokens_seen": 257687552
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046560682046138414,
      "loss": 3.2193,
      "theoretical_loss": 4.22185438156113,
      "tokens_seen": 257753088
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046559679037111337,
      "loss": 3.0531,
      "theoretical_loss": 4.221725333632123,
      "tokens_seen": 257818624
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004655867602808425,
      "loss": 3.2622,
      "theoretical_loss": 4.221596327684429,
      "tokens_seen": 257884160
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046557673019057173,
      "loss": 2.9811,
      "theoretical_loss": 4.221467363693727,
      "tokens_seen": 257949696
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004655667001003009,
      "loss": 3.275,
      "theoretical_loss": 4.221338441635716,
      "tokens_seen": 258015232
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004655566700100301,
      "loss": 3.323,
      "theoretical_loss": 4.221209561486116,
      "tokens_seen": 258080768
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004655466399197593,
      "loss": 3.1546,
      "theoretical_loss": 4.221080723220668,
      "tokens_seen": 258146304
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046553660982948846,
      "loss": 3.3392,
      "theoretical_loss": 4.22095192681513,
      "tokens_seen": 258211840
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046552657973921764,
      "loss": 3.3936,
      "theoretical_loss": 4.220823172245284,
      "tokens_seen": 258277376
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004655165496489469,
      "loss": 3.1002,
      "theoretical_loss": 4.220694459486931,
      "tokens_seen": 258342912
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.000465506519558676,
      "loss": 3.3248,
      "theoretical_loss": 4.220565788515891,
      "tokens_seen": 258408448
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046549648946840524,
      "loss": 3.2462,
      "theoretical_loss": 4.220437159308005,
      "tokens_seen": 258473984
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046548645937813436,
      "loss": 3.2011,
      "theoretical_loss": 4.2203085718391335,
      "tokens_seen": 258539520
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004654764292878636,
      "loss": 3.3226,
      "theoretical_loss": 4.220180026085158,
      "tokens_seen": 258605056
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004654663991975928,
      "loss": 3.4375,
      "theoretical_loss": 4.220051522021979,
      "tokens_seen": 258670592
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046545636910732196,
      "loss": 3.2813,
      "theoretical_loss": 4.219923059625518,
      "tokens_seen": 258736128
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004654463390170512,
      "loss": 3.2475,
      "theoretical_loss": 4.219794638871717,
      "tokens_seen": 258801664
    },
    {
      "epoch": 0.08,
      "objective/train/advantage_avg": 0.49539265036582947,
      "objective/train/docs_used": 158510,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.6545803546905518,
      "objective/train/original_loss": 2.654580593109131,
      "objective/train/theoretical_loss": 4.219666259736535,
      "objective/train/tokens_used": 279327200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2454570084810257,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507869720458984,
      "objective/train/weighted_lm_loss": 2.7892425060272217,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0468999147415161,
      "theoretical_loss": 4.219666259736535,
      "tokens_seen": 258867200
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004654363089267803,
      "loss": 3.3639,
      "theoretical_loss": 4.219666259736535,
      "tokens_seen": 258867200
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046542627883650956,
      "loss": 3.3392,
      "theoretical_loss": 4.219537922195954,
      "tokens_seen": 258932736
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046541624874623874,
      "loss": 3.1022,
      "theoretical_loss": 4.219409626225975,
      "tokens_seen": 258998272
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004654062186559679,
      "loss": 3.3921,
      "theoretical_loss": 4.219281371802619,
      "tokens_seen": 259063808
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004653961885656971,
      "loss": 3.3632,
      "theoretical_loss": 4.219153158901927,
      "tokens_seen": 259129344
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046538615847542634,
      "loss": 3.289,
      "theoretical_loss": 4.219024987499958,
      "tokens_seen": 259194880
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046537612838515546,
      "loss": 3.222,
      "theoretical_loss": 4.218896857572794,
      "tokens_seen": 259260416
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004653660982948847,
      "loss": 3.4285,
      "theoretical_loss": 4.218768769096535,
      "tokens_seen": 259325952
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004653560682046138,
      "loss": 3.1656,
      "theoretical_loss": 4.218640722047301,
      "tokens_seen": 259391488
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046534603811434306,
      "loss": 3.3189,
      "theoretical_loss": 4.218512716401232,
      "tokens_seen": 259457024
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046533600802407224,
      "loss": 3.296,
      "theoretical_loss": 4.218384752134487,
      "tokens_seen": 259522560
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004653259779338014,
      "loss": 3.4392,
      "theoretical_loss": 4.218256829223247,
      "tokens_seen": 259588096
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004653159478435306,
      "loss": 3.3784,
      "theoretical_loss": 4.2181289476437085,
      "tokens_seen": 259653632
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004653059177532598,
      "loss": 3.4066,
      "theoretical_loss": 4.218001107372092,
      "tokens_seen": 259719168
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046529588766298897,
      "loss": 3.1782,
      "theoretical_loss": 4.217873308384636,
      "tokens_seen": 259784704
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004652858575727182,
      "loss": 3.1685,
      "theoretical_loss": 4.2177455506575985,
      "tokens_seen": 259850240
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046527582748244733,
      "loss": 3.41,
      "theoretical_loss": 4.217617834167258,
      "tokens_seen": 259915776
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046526579739217656,
      "loss": 3.1554,
      "theoretical_loss": 4.217490158889911,
      "tokens_seen": 259981312
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046525576730190575,
      "loss": 3.3115,
      "theoretical_loss": 4.217362524801874,
      "tokens_seen": 260046848
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004652457372116349,
      "loss": 3.1584,
      "theoretical_loss": 4.217234931879483,
      "tokens_seen": 260112384
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004652357071213641,
      "loss": 3.3943,
      "theoretical_loss": 4.2171073800990975,
      "tokens_seen": 260177920
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004652256770310933,
      "loss": 3.3007,
      "theoretical_loss": 4.21697986943709,
      "tokens_seen": 260243456
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046521564694082247,
      "loss": 3.0405,
      "theoretical_loss": 4.216852399869857,
      "tokens_seen": 260308992
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004652056168505517,
      "loss": 3.2575,
      "theoretical_loss": 4.216724971373813,
      "tokens_seen": 260374528
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046519558676028083,
      "loss": 3.1222,
      "theoretical_loss": 4.216597583925391,
      "tokens_seen": 260440064
    },
    {
      "epoch": 0.08,
      "objective/train/advantage_avg": 0.49481797218322754,
      "objective/train/docs_used": 159098,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.189685583114624,
      "objective/train/original_loss": 3.1896862983703613,
      "objective/train/theoretical_loss": 4.216470237501046,
      "objective/train/tokens_used": 280965600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24491101503372192,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507268905639648,
      "objective/train/weighted_lm_loss": 3.35136342048645,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0363596677780151,
      "theoretical_loss": 4.216470237501046,
      "tokens_seen": 260505600
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046518555667001007,
      "loss": 3.1348,
      "theoretical_loss": 4.216470237501046,
      "tokens_seen": 260505600
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004651755265797392,
      "loss": 3.2959,
      "theoretical_loss": 4.21634293207725,
      "tokens_seen": 260571136
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046516549648946843,
      "loss": 3.3158,
      "theoretical_loss": 4.216215667630497,
      "tokens_seen": 260636672
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004651554663991976,
      "loss": 3.2388,
      "theoretical_loss": 4.216088444137297,
      "tokens_seen": 260702208
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004651454363089268,
      "loss": 3.2042,
      "theoretical_loss": 4.215961261574181,
      "tokens_seen": 260767744
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.000465135406218656,
      "loss": 3.3281,
      "theoretical_loss": 4.215834119917702,
      "tokens_seen": 260833280
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046512537612838515,
      "loss": 3.2763,
      "theoretical_loss": 4.215707019144427,
      "tokens_seen": 260898816
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046511534603811434,
      "loss": 3.2023,
      "theoretical_loss": 4.215579959230946,
      "tokens_seen": 260964352
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046510531594784357,
      "loss": 3.1073,
      "theoretical_loss": 4.2154529401538685,
      "tokens_seen": 261029888
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004650952858575727,
      "loss": 3.5426,
      "theoretical_loss": 4.215325961889821,
      "tokens_seen": 261095424
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046508525576730193,
      "loss": 3.448,
      "theoretical_loss": 4.215199024415451,
      "tokens_seen": 261160960
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004650752256770311,
      "loss": 3.4698,
      "theoretical_loss": 4.215072127707424,
      "tokens_seen": 261226496
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004650651955867603,
      "loss": 3.4116,
      "theoretical_loss": 4.214945271742426,
      "tokens_seen": 261292032
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004650551654964895,
      "loss": 2.996,
      "theoretical_loss": 4.214818456497161,
      "tokens_seen": 261357568
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046504513540621866,
      "loss": 3.1026,
      "theoretical_loss": 4.214691681948352,
      "tokens_seen": 261423104
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046503510531594784,
      "loss": 3.4243,
      "theoretical_loss": 4.214564948072743,
      "tokens_seen": 261488640
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004650250752256771,
      "loss": 3.265,
      "theoretical_loss": 4.214438254847096,
      "tokens_seen": 261554176
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004650150451354062,
      "loss": 3.3359,
      "theoretical_loss": 4.214311602248191,
      "tokens_seen": 261619712
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046500501504513544,
      "loss": 3.1675,
      "theoretical_loss": 4.2141849902528286,
      "tokens_seen": 261685248
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046499498495486456,
      "loss": 3.2194,
      "theoretical_loss": 4.214058418837828,
      "tokens_seen": 261750784
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004649849548645938,
      "loss": 3.3328,
      "theoretical_loss": 4.213931887980027,
      "tokens_seen": 261816320
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.000464974924774323,
      "loss": 3.1016,
      "theoretical_loss": 4.213805397656283,
      "tokens_seen": 261881856
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046496489468405216,
      "loss": 3.4963,
      "theoretical_loss": 4.213678947843472,
      "tokens_seen": 261947392
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046495486459378134,
      "loss": 3.3557,
      "theoretical_loss": 4.213552538518489,
      "tokens_seen": 262012928
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004649448345035105,
      "loss": 3.4938,
      "theoretical_loss": 4.2134261696582485,
      "tokens_seen": 262078464
    },
    {
      "epoch": 0.08,
      "objective/train/advantage_avg": 0.49375057220458984,
      "objective/train/docs_used": 160106,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.024266481399536,
      "objective/train/original_loss": 3.0242667198181152,
      "objective/train/theoretical_loss": 4.213299841239684,
      "objective/train/tokens_used": 282604000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24386641383171082,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506147146224976,
      "objective/train/weighted_lm_loss": 3.177248001098633,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0304546356201172,
      "theoretical_loss": 4.213299841239684,
      "tokens_seen": 262144000
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004649348044132397,
      "loss": 3.3574,
      "theoretical_loss": 4.213299841239684,
      "tokens_seen": 262144000
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046492477432296894,
      "loss": 3.2579,
      "theoretical_loss": 4.213173553239747,
      "tokens_seen": 262209536
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046491474423269807,
      "loss": 3.4172,
      "theoretical_loss": 4.213047305635407,
      "tokens_seen": 262275072
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004649047141424273,
      "loss": 3.4652,
      "theoretical_loss": 4.212921098403656,
      "tokens_seen": 262340608
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004648946840521565,
      "loss": 3.4522,
      "theoretical_loss": 4.2127949315215005,
      "tokens_seen": 262406144
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046488465396188566,
      "loss": 3.3904,
      "theoretical_loss": 4.212668804965968,
      "tokens_seen": 262471680
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046487462387161485,
      "loss": 3.2328,
      "theoretical_loss": 4.212542718714107,
      "tokens_seen": 262537216
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.000464864593781344,
      "loss": 3.0579,
      "theoretical_loss": 4.212416672742981,
      "tokens_seen": 262602752
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004648545636910732,
      "loss": 3.4744,
      "theoretical_loss": 4.212290667029672,
      "tokens_seen": 262668288
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046484453360080244,
      "loss": 3.3507,
      "theoretical_loss": 4.212164701551285,
      "tokens_seen": 262733824
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046483450351053157,
      "loss": 3.4861,
      "theoretical_loss": 4.212038776284941,
      "tokens_seen": 262799360
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004648244734202608,
      "loss": 3.2037,
      "theoretical_loss": 4.211912891207779,
      "tokens_seen": 262864896
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046481444332998993,
      "loss": 3.1012,
      "theoretical_loss": 4.2117870462969575,
      "tokens_seen": 262930432
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046480441323971917,
      "loss": 3.2158,
      "theoretical_loss": 4.211661241529654,
      "tokens_seen": 262995968
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046479438314944835,
      "loss": 3.0504,
      "theoretical_loss": 4.211535476883066,
      "tokens_seen": 263061504
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046478435305917753,
      "loss": 3.3602,
      "theoretical_loss": 4.211409752334407,
      "tokens_seen": 263127040
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004647743229689067,
      "loss": 3.1195,
      "theoretical_loss": 4.211284067860909,
      "tokens_seen": 263192576
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046476429287863595,
      "loss": 3.2458,
      "theoretical_loss": 4.211158423439825,
      "tokens_seen": 263258112
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004647542627883651,
      "loss": 3.2472,
      "theoretical_loss": 4.211032819048425,
      "tokens_seen": 263323648
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004647442326980943,
      "loss": 3.2167,
      "theoretical_loss": 4.210907254663999,
      "tokens_seen": 263389184
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046473420260782344,
      "loss": 3.2455,
      "theoretical_loss": 4.210781730263853,
      "tokens_seen": 263454720
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046472417251755267,
      "loss": 3.3262,
      "theoretical_loss": 4.210656245825313,
      "tokens_seen": 263520256
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004647141424272819,
      "loss": 3.2587,
      "theoretical_loss": 4.210530801325724,
      "tokens_seen": 263585792
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046470411233701103,
      "loss": 3.1157,
      "theoretical_loss": 4.21040539674245,
      "tokens_seen": 263651328
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046469408224674027,
      "loss": 3.3647,
      "theoretical_loss": 4.21028003205287,
      "tokens_seen": 263716864
    },
    {
      "epoch": 0.08,
      "objective/train/advantage_avg": 0.49442407488822937,
      "objective/train/docs_used": 160649,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.9327964782714844,
      "objective/train/original_loss": 2.9327964782714844,
      "objective/train/theoretical_loss": 4.210154707234386,
      "objective/train/tokens_used": 284242400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24454809725284576,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506855249404907,
      "objective/train/weighted_lm_loss": 3.0815911293029785,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0396549701690674,
      "theoretical_loss": 4.210154707234386,
      "tokens_seen": 263782400
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004646840521564694,
      "loss": 3.1386,
      "theoretical_loss": 4.210154707234386,
      "tokens_seen": 263782400
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046467402206619863,
      "loss": 3.2863,
      "theoretical_loss": 4.210029422264414,
      "tokens_seen": 263847936
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004646639919759278,
      "loss": 3.1865,
      "theoretical_loss": 4.209904177120393,
      "tokens_seen": 263913472
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.000464653961885657,
      "loss": 3.1673,
      "theoretical_loss": 4.209778971779777,
      "tokens_seen": 263979008
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004646439317953862,
      "loss": 3.4052,
      "theoretical_loss": 4.209653806220039,
      "tokens_seen": 264044544
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046463390170511535,
      "loss": 3.1652,
      "theoretical_loss": 4.2095286804186705,
      "tokens_seen": 264110080
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046462387161484454,
      "loss": 3.2642,
      "theoretical_loss": 4.209403594353183,
      "tokens_seen": 264175616
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046461384152457377,
      "loss": 3.0833,
      "theoretical_loss": 4.209278548001103,
      "tokens_seen": 264241152
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004646038114343029,
      "loss": 3.2357,
      "theoretical_loss": 4.209153541339977,
      "tokens_seen": 264306688
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046459378134403213,
      "loss": 3.1683,
      "theoretical_loss": 4.209028574347371,
      "tokens_seen": 264372224
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004645837512537613,
      "loss": 3.2556,
      "theoretical_loss": 4.208903647000868,
      "tokens_seen": 264437760
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004645737211634905,
      "loss": 3.2828,
      "theoretical_loss": 4.20877875927807,
      "tokens_seen": 264503296
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004645636910732197,
      "loss": 3.4592,
      "theoretical_loss": 4.208653911156595,
      "tokens_seen": 264568832
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046455366098294886,
      "loss": 3.1622,
      "theoretical_loss": 4.20852910261408,
      "tokens_seen": 264634368
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046454363089267804,
      "loss": 3.2535,
      "theoretical_loss": 4.208404333628184,
      "tokens_seen": 264699904
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004645336008024073,
      "loss": 2.9357,
      "theoretical_loss": 4.208279604176577,
      "tokens_seen": 264765440
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004645235707121364,
      "loss": 3.4079,
      "theoretical_loss": 4.208154914236954,
      "tokens_seen": 264830976
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046451354062186564,
      "loss": 3.0383,
      "theoretical_loss": 4.208030263787025,
      "tokens_seen": 264896512
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046450351053159476,
      "loss": 3.2044,
      "theoretical_loss": 4.2079056528045164,
      "tokens_seen": 264962048
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.000464493480441324,
      "loss": 3.3491,
      "theoretical_loss": 4.207781081267177,
      "tokens_seen": 265027584
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004644834503510532,
      "loss": 3.0343,
      "theoretical_loss": 4.207656549152769,
      "tokens_seen": 265093120
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046447342026078236,
      "loss": 3.444,
      "theoretical_loss": 4.207532056439075,
      "tokens_seen": 265158656
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046446339017051154,
      "loss": 3.302,
      "theoretical_loss": 4.207407603103897,
      "tokens_seen": 265224192
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004644533600802407,
      "loss": 3.5367,
      "theoretical_loss": 4.207283189125054,
      "tokens_seen": 265289728
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004644433299899699,
      "loss": 3.2831,
      "theoretical_loss": 4.20715881448038,
      "tokens_seen": 265355264
    },
    {
      "epoch": 0.08,
      "objective/train/advantage_avg": 0.4958003759384155,
      "objective/train/docs_used": 161970,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.8476176261901855,
      "objective/train/original_loss": 2.8476171493530273,
      "objective/train/theoretical_loss": 4.20703447914773,
      "objective/train/tokens_used": 285880800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2458822876214981,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0508298873901367,
      "objective/train/weighted_lm_loss": 2.992149829864502,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0338950157165527,
      "theoretical_loss": 4.20703447914773,
      "tokens_seen": 265420800
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046443329989969914,
      "loss": 3.056,
      "theoretical_loss": 4.20703447914773,
      "tokens_seen": 265420800
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046442326980942827,
      "loss": 3.1875,
      "theoretical_loss": 4.206910183104977,
      "tokens_seen": 265486336
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004644132397191575,
      "loss": 3.5416,
      "theoretical_loss": 4.20678592633001,
      "tokens_seen": 265551872
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004644032096288867,
      "loss": 3.1435,
      "theoretical_loss": 4.206661708800739,
      "tokens_seen": 265617408
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046439317953861586,
      "loss": 3.2714,
      "theoretical_loss": 4.206537530495089,
      "tokens_seen": 265682944
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046438314944834505,
      "loss": 3.2405,
      "theoretical_loss": 4.2064133913910045,
      "tokens_seen": 265748480
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004643731193580742,
      "loss": 3.1429,
      "theoretical_loss": 4.206289291466445,
      "tokens_seen": 265814016
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004643630892678034,
      "loss": 3.5762,
      "theoretical_loss": 4.206165230699392,
      "tokens_seen": 265879552
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046435305917753264,
      "loss": 3.0256,
      "theoretical_loss": 4.206041209067842,
      "tokens_seen": 265945088
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046434302908726177,
      "loss": 3.0938,
      "theoretical_loss": 4.20591722654981,
      "tokens_seen": 266010624
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.000464332998996991,
      "loss": 3.0884,
      "theoretical_loss": 4.20579328312333,
      "tokens_seen": 266076160
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046432296890672013,
      "loss": 3.5019,
      "theoretical_loss": 4.2056693787664505,
      "tokens_seen": 266141696
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046431293881644937,
      "loss": 3.1497,
      "theoretical_loss": 4.205545513457242,
      "tokens_seen": 266207232
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046430290872617855,
      "loss": 3.0552,
      "theoretical_loss": 4.205421687173789,
      "tokens_seen": 266272768
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046429287863590773,
      "loss": 3.2826,
      "theoretical_loss": 4.2052978998941954,
      "tokens_seen": 266338304
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004642828485456369,
      "loss": 3.0964,
      "theoretical_loss": 4.205174151596583,
      "tokens_seen": 266403840
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046427281845536615,
      "loss": 3.169,
      "theoretical_loss": 4.205050442259091,
      "tokens_seen": 266469376
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004642627883650953,
      "loss": 3.3483,
      "theoretical_loss": 4.204926771859874,
      "tokens_seen": 266534912
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004642527582748245,
      "loss": 3.3515,
      "theoretical_loss": 4.204803140377109,
      "tokens_seen": 266600448
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046424272818455364,
      "loss": 3.337,
      "theoretical_loss": 4.204679547788987,
      "tokens_seen": 266665984
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046423269809428287,
      "loss": 3.2694,
      "theoretical_loss": 4.204555994073717,
      "tokens_seen": 266731520
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046422266800401205,
      "loss": 2.8969,
      "theoretical_loss": 4.204432479209526,
      "tokens_seen": 266797056
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046421263791374123,
      "loss": 3.2333,
      "theoretical_loss": 4.204309003174659,
      "tokens_seen": 266862592
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004642026078234704,
      "loss": 3.0853,
      "theoretical_loss": 4.204185565947378,
      "tokens_seen": 266928128
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004641925777331996,
      "loss": 3.2278,
      "theoretical_loss": 4.2040621675059615,
      "tokens_seen": 266993664
    },
    {
      "epoch": 0.08,
      "objective/train/advantage_avg": 0.4954144358634949,
      "objective/train/docs_used": 162606,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.9374938011169434,
      "objective/train/original_loss": 2.9374940395355225,
      "objective/train/theoretical_loss": 4.203938807828708,
      "objective/train/tokens_used": 287519200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2454727739095688,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507893562316895,
      "objective/train/weighted_lm_loss": 3.08664870262146,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0432155132293701,
      "theoretical_loss": 4.203938807828708,
      "tokens_seen": 267059200
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004641825476429288,
      "loss": 2.9838,
      "theoretical_loss": 4.203938807828708,
      "tokens_seen": 267059200
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.000464172517552658,
      "loss": 3.068,
      "theoretical_loss": 4.20381548689393,
      "tokens_seen": 267124736
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046416248746238714,
      "loss": 3.114,
      "theoretical_loss": 4.203692204679962,
      "tokens_seen": 267190272
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004641524573721164,
      "loss": 3.2576,
      "theoretical_loss": 4.203568961165152,
      "tokens_seen": 267255808
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004641424272818455,
      "loss": 3.0404,
      "theoretical_loss": 4.203445756327866,
      "tokens_seen": 267321344
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046413239719157474,
      "loss": 3.4665,
      "theoretical_loss": 4.203322590146491,
      "tokens_seen": 267386880
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004641223671013039,
      "loss": 3.2001,
      "theoretical_loss": 4.203199462599425,
      "tokens_seen": 267452416
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004641123370110331,
      "loss": 3.2129,
      "theoretical_loss": 4.203076373665089,
      "tokens_seen": 267517952
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004641023069207623,
      "loss": 3.1346,
      "theoretical_loss": 4.202953323321919,
      "tokens_seen": 267583488
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004640922768304915,
      "loss": 3.0699,
      "theoretical_loss": 4.202830311548369,
      "tokens_seen": 267649024
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046408224674022064,
      "loss": 3.2311,
      "theoretical_loss": 4.2027073383229085,
      "tokens_seen": 267714560
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004640722166499499,
      "loss": 3.267,
      "theoretical_loss": 4.2025844036240265,
      "tokens_seen": 267780096
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.000464062186559679,
      "loss": 3.3832,
      "theoretical_loss": 4.202461507430228,
      "tokens_seen": 267845632
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046405215646940824,
      "loss": 3.1238,
      "theoretical_loss": 4.202338649720037,
      "tokens_seen": 267911168
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004640421263791374,
      "loss": 3.1114,
      "theoretical_loss": 4.202215830471991,
      "tokens_seen": 267976704
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004640320962888666,
      "loss": 3.5002,
      "theoretical_loss": 4.20209304966465,
      "tokens_seen": 268042240
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004640220661985958,
      "loss": 3.3462,
      "theoretical_loss": 4.201970307276586,
      "tokens_seen": 268107776
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046401203610832496,
      "loss": 3.2773,
      "theoretical_loss": 4.201847603286391,
      "tokens_seen": 268173312
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046400200601805414,
      "loss": 3.1897,
      "theoretical_loss": 4.201724937672674,
      "tokens_seen": 268238848
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004639919759277834,
      "loss": 3.1589,
      "theoretical_loss": 4.201602310414062,
      "tokens_seen": 268304384
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004639819458375125,
      "loss": 3.2254,
      "theoretical_loss": 4.2014797214891955,
      "tokens_seen": 268369920
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046397191574724174,
      "loss": 3.1519,
      "theoretical_loss": 4.2013571708767365,
      "tokens_seen": 268435456
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004639618856569709,
      "loss": 3.2255,
      "theoretical_loss": 4.20123465855536,
      "tokens_seen": 268500992
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004639518555667001,
      "loss": 2.9796,
      "theoretical_loss": 4.201112184503762,
      "tokens_seen": 268566528
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046394182547642934,
      "loss": 2.9009,
      "theoretical_loss": 4.200989748700653,
      "tokens_seen": 268632064
    },
    {
      "epoch": 0.08,
      "objective/train/advantage_avg": 0.49540141224861145,
      "objective/train/docs_used": 163909,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.3215978145599365,
      "objective/train/original_loss": 3.3215978145599365,
      "objective/train/theoretical_loss": 4.200867351124762,
      "objective/train/tokens_used": 289157600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24549348652362823,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050788164138794,
      "objective/train/weighted_lm_loss": 3.490323543548584,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0417574644088745,
      "theoretical_loss": 4.200867351124762,
      "tokens_seen": 268697600
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046393179538615847,
      "loss": 3.2784,
      "theoretical_loss": 4.200867351124762,
      "tokens_seen": 268697600
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004639217652958877,
      "loss": 3.4025,
      "theoretical_loss": 4.200744991754833,
      "tokens_seen": 268763136
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004639117352056169,
      "loss": 3.249,
      "theoretical_loss": 4.200622670569628,
      "tokens_seen": 268828672
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046390170511534606,
      "loss": 3.4896,
      "theoretical_loss": 4.200500387547929,
      "tokens_seen": 268894208
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046389167502507525,
      "loss": 3.036,
      "theoretical_loss": 4.200378142668528,
      "tokens_seen": 268959744
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004638816449348044,
      "loss": 3.2981,
      "theoretical_loss": 4.200255935910242,
      "tokens_seen": 269025280
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004638716148445336,
      "loss": 3.1599,
      "theoretical_loss": 4.2001337672518995,
      "tokens_seen": 269090816
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046386158475426284,
      "loss": 3.2382,
      "theoretical_loss": 4.200011636672347,
      "tokens_seen": 269156352
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046385155466399197,
      "loss": 3.25,
      "theoretical_loss": 4.199889544150449,
      "tokens_seen": 269221888
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004638415245737212,
      "loss": 3.128,
      "theoretical_loss": 4.199767489665086,
      "tokens_seen": 269287424
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046383149448345033,
      "loss": 3.254,
      "theoretical_loss": 4.199645473195155,
      "tokens_seen": 269352960
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046382146439317957,
      "loss": 3.3306,
      "theoretical_loss": 4.199523494719571,
      "tokens_seen": 269418496
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046381143430290875,
      "loss": 3.4049,
      "theoretical_loss": 4.199401554217266,
      "tokens_seen": 269484032
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046380140421263793,
      "loss": 3.1715,
      "theoretical_loss": 4.1992796516671875,
      "tokens_seen": 269549568
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004637913741223671,
      "loss": 3.1583,
      "theoretical_loss": 4.1991577870483,
      "tokens_seen": 269615104
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046378134403209635,
      "loss": 3.4835,
      "theoretical_loss": 4.199035960339585,
      "tokens_seen": 269680640
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004637713139418255,
      "loss": 3.1024,
      "theoretical_loss": 4.198914171520041,
      "tokens_seen": 269746176
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004637612838515547,
      "loss": 3.2155,
      "theoretical_loss": 4.198792420568684,
      "tokens_seen": 269811712
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046375125376128384,
      "loss": 3.2799,
      "theoretical_loss": 4.198670707464545,
      "tokens_seen": 269877248
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046374122367101307,
      "loss": 3.434,
      "theoretical_loss": 4.198549032186672,
      "tokens_seen": 269942784
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046373119358074225,
      "loss": 3.1065,
      "theoretical_loss": 4.19842739471413,
      "tokens_seen": 270008320
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046372116349047143,
      "loss": 3.0138,
      "theoretical_loss": 4.198305795026004,
      "tokens_seen": 270073856
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004637111334002006,
      "loss": 3.2409,
      "theoretical_loss": 4.198184233101388,
      "tokens_seen": 270139392
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004637011033099298,
      "loss": 3.3798,
      "theoretical_loss": 4.1980627089194,
      "tokens_seen": 270204928
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.000463691073219659,
      "loss": 3.4042,
      "theoretical_loss": 4.197941222459171,
      "tokens_seen": 270270464
    },
    {
      "epoch": 0.08,
      "objective/train/advantage_avg": 0.4941849410533905,
      "objective/train/docs_used": 164570,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.5010392665863037,
      "objective/train/original_loss": 3.501039505004883,
      "objective/train/theoretical_loss": 4.197819773699849,
      "objective/train/tokens_used": 290796000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24428284168243408,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050660252571106,
      "objective/train/weighted_lm_loss": 3.6784470081329346,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0396549701690674,
      "theoretical_loss": 4.197819773699849,
      "tokens_seen": 270336000
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004636810431293882,
      "loss": 3.2191,
      "theoretical_loss": 4.197819773699849,
      "tokens_seen": 270336000
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046367101303911734,
      "loss": 3.3022,
      "theoretical_loss": 4.197698362620599,
      "tokens_seen": 270401536
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004636609829488466,
      "loss": 3.2937,
      "theoretical_loss": 4.197576989200604,
      "tokens_seen": 270467072
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004636509528585757,
      "loss": 3.3595,
      "theoretical_loss": 4.19745565341906,
      "tokens_seen": 270532608
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046364092276830494,
      "loss": 3.3163,
      "theoretical_loss": 4.197334355255182,
      "tokens_seen": 270598144
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004636308926780341,
      "loss": 3.325,
      "theoretical_loss": 4.197213094688202,
      "tokens_seen": 270663680
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004636208625877633,
      "loss": 3.1882,
      "theoretical_loss": 4.197091871697366,
      "tokens_seen": 270729216
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004636108324974925,
      "loss": 3.3581,
      "theoretical_loss": 4.196970686261939,
      "tokens_seen": 270794752
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004636008024072217,
      "loss": 3.2856,
      "theoretical_loss": 4.196849538361201,
      "tokens_seen": 270860288
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046359077231695084,
      "loss": 3.1961,
      "theoretical_loss": 4.196728427974451,
      "tokens_seen": 270925824
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004635807422266801,
      "loss": 3.4672,
      "theoretical_loss": 4.196607355080999,
      "tokens_seen": 270991360
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004635707121364092,
      "loss": 3.4041,
      "theoretical_loss": 4.196486319660177,
      "tokens_seen": 271056896
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046356068204613844,
      "loss": 3.2307,
      "theoretical_loss": 4.19636532169133,
      "tokens_seen": 271122432
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004635506519558676,
      "loss": 3.1557,
      "theoretical_loss": 4.196244361153821,
      "tokens_seen": 271187968
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004635406218655968,
      "loss": 3.4619,
      "theoretical_loss": 4.196123438027029,
      "tokens_seen": 271253504
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.000463530591775326,
      "loss": 3.4068,
      "theoretical_loss": 4.1960025522903495,
      "tokens_seen": 271319040
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046352056168505516,
      "loss": 3.2668,
      "theoretical_loss": 4.1958817039231935,
      "tokens_seen": 271384576
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046351053159478434,
      "loss": 3.2197,
      "theoretical_loss": 4.195760892904988,
      "tokens_seen": 271450112
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004635005015045136,
      "loss": 3.1693,
      "theoretical_loss": 4.195640119215179,
      "tokens_seen": 271515648
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004634904714142427,
      "loss": 3.3117,
      "theoretical_loss": 4.195519382833226,
      "tokens_seen": 271581184
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046348044132397194,
      "loss": 3.4852,
      "theoretical_loss": 4.195398683738605,
      "tokens_seen": 271646720
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046347041123370107,
      "loss": 3.2239,
      "theoretical_loss": 4.19527802191081,
      "tokens_seen": 271712256
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004634603811434303,
      "loss": 3.0711,
      "theoretical_loss": 4.195157397329349,
      "tokens_seen": 271777792
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004634503510531595,
      "loss": 3.1142,
      "theoretical_loss": 4.1950368099737485,
      "tokens_seen": 271843328
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046344032096288867,
      "loss": 3.2495,
      "theoretical_loss": 4.1949162598235485,
      "tokens_seen": 271908864
    },
    {
      "epoch": 0.08,
      "objective/train/advantage_avg": 0.49496516585350037,
      "objective/train/docs_used": 165961,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.3817739486694336,
      "objective/train/original_loss": 3.3817739486694336,
      "objective/train/theoretical_loss": 4.194795746858309,
      "objective/train/tokens_used": 292434400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24505899846553802,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050742268562317,
      "objective/train/weighted_lm_loss": 3.553297519683838,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0253151655197144,
      "theoretical_loss": 4.194795746858309,
      "tokens_seen": 271974400
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046343029087261785,
      "loss": 3.2224,
      "theoretical_loss": 4.194795746858309,
      "tokens_seen": 271974400
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004634202607823471,
      "loss": 3.3045,
      "theoretical_loss": 4.194675271057601,
      "tokens_seen": 272039936
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004634102306920762,
      "loss": 3.4555,
      "theoretical_loss": 4.194554832401017,
      "tokens_seen": 272105472
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046340020060180545,
      "loss": 3.3794,
      "theoretical_loss": 4.194434430868161,
      "tokens_seen": 272171008
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046339017051153457,
      "loss": 3.4541,
      "theoretical_loss": 4.194314066438656,
      "tokens_seen": 272236544
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004633801404212638,
      "loss": 3.2282,
      "theoretical_loss": 4.194193739092141,
      "tokens_seen": 272302080
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.000463370110330993,
      "loss": 3.1169,
      "theoretical_loss": 4.194073448808269,
      "tokens_seen": 272367616
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046336008024072217,
      "loss": 3.2801,
      "theoretical_loss": 4.193953195566712,
      "tokens_seen": 272433152
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046335005015045135,
      "loss": 3.008,
      "theoretical_loss": 4.193832979347155,
      "tokens_seen": 272498688
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046334002006018053,
      "loss": 3.1138,
      "theoretical_loss": 4.193712800129301,
      "tokens_seen": 272564224
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004633299899699097,
      "loss": 3.2033,
      "theoretical_loss": 4.193592657892869,
      "tokens_seen": 272629760
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046331995987963895,
      "loss": 3.3578,
      "theoretical_loss": 4.193472552617592,
      "tokens_seen": 272695296
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004633099297893681,
      "loss": 3.2379,
      "theoretical_loss": 4.193352484283222,
      "tokens_seen": 272760832
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004632998996990973,
      "loss": 3.2911,
      "theoretical_loss": 4.193232452869525,
      "tokens_seen": 272826368
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046328986960882644,
      "loss": 3.2163,
      "theoretical_loss": 4.193112458356283,
      "tokens_seen": 272891904
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004632798395185557,
      "loss": 3.5531,
      "theoretical_loss": 4.192992500723294,
      "tokens_seen": 272957440
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046326980942828485,
      "loss": 3.0282,
      "theoretical_loss": 4.192872579950373,
      "tokens_seen": 273022976
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046325977933801404,
      "loss": 3.2411,
      "theoretical_loss": 4.192752696017349,
      "tokens_seen": 273088512
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004632497492477432,
      "loss": 3.2473,
      "theoretical_loss": 4.1926328489040685,
      "tokens_seen": 273154048
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046323971915747245,
      "loss": 3.1736,
      "theoretical_loss": 4.192513038590393,
      "tokens_seen": 273219584
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004632296890672016,
      "loss": 3.2463,
      "theoretical_loss": 4.192393265056201,
      "tokens_seen": 273285120
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004632196589769308,
      "loss": 3.2967,
      "theoretical_loss": 4.192273528281385,
      "tokens_seen": 273350656
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046320962888666,
      "loss": 3.0729,
      "theoretical_loss": 4.1921538282458535,
      "tokens_seen": 273416192
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004631995987963892,
      "loss": 3.2298,
      "theoretical_loss": 4.192034164929534,
      "tokens_seen": 273481728
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004631895687061184,
      "loss": 3.3324,
      "theoretical_loss": 4.191914538312365,
      "tokens_seen": 273547264
    },
    {
      "epoch": 0.08,
      "objective/train/advantage_avg": 0.4931058883666992,
      "objective/train/docs_used": 166474,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.4296326637268066,
      "objective/train/original_loss": 3.4296329021453857,
      "objective/train/theoretical_loss": 4.191794948374304,
      "objective/train/tokens_used": 294072800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24323640763759613,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505471229553223,
      "objective/train/weighted_lm_loss": 3.6028192043304443,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0417574644088745,
      "theoretical_loss": 4.191794948374304,
      "tokens_seen": 273612800
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046317953861584754,
      "loss": 3.2264,
      "theoretical_loss": 4.191794948374304,
      "tokens_seen": 273612800
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004631695085255768,
      "loss": 3.1726,
      "theoretical_loss": 4.191675395095324,
      "tokens_seen": 273678336
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004631594784353059,
      "loss": 3.1795,
      "theoretical_loss": 4.191555878455411,
      "tokens_seen": 273743872
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046314944834503514,
      "loss": 3.2632,
      "theoretical_loss": 4.191436398434571,
      "tokens_seen": 273809408
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004631394182547643,
      "loss": 3.3704,
      "theoretical_loss": 4.191316955012822,
      "tokens_seen": 273874944
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004631293881644935,
      "loss": 3.3504,
      "theoretical_loss": 4.1911975481702,
      "tokens_seen": 273940480
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004631193580742227,
      "loss": 3.1622,
      "theoretical_loss": 4.191078177886755,
      "tokens_seen": 274006016
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004631093279839519,
      "loss": 3.278,
      "theoretical_loss": 4.190958844142555,
      "tokens_seen": 274071552
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046309929789368104,
      "loss": 3.2279,
      "theoretical_loss": 4.1908395469176805,
      "tokens_seen": 274137088
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004630892678034103,
      "loss": 3.27,
      "theoretical_loss": 4.19072028619223,
      "tokens_seen": 274202624
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004630792377131394,
      "loss": 3.1868,
      "theoretical_loss": 4.190601061946316,
      "tokens_seen": 274268160
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046306920762286864,
      "loss": 3.2785,
      "theoretical_loss": 4.190481874160068,
      "tokens_seen": 274333696
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004630591775325978,
      "loss": 3.4865,
      "theoretical_loss": 4.19036272281363,
      "tokens_seen": 274399232
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.000463049147442327,
      "loss": 3.2173,
      "theoretical_loss": 4.190243607887163,
      "tokens_seen": 274464768
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004630391173520562,
      "loss": 3.3815,
      "theoretical_loss": 4.190124529360842,
      "tokens_seen": 274530304
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046302908726178536,
      "loss": 3.3365,
      "theoretical_loss": 4.190005487214857,
      "tokens_seen": 274595840
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046301905717151455,
      "loss": 3.3612,
      "theoretical_loss": 4.189886481429417,
      "tokens_seen": 274661376
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004630090270812438,
      "loss": 3.3354,
      "theoretical_loss": 4.189767511984741,
      "tokens_seen": 274726912
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004629989969909729,
      "loss": 3.1589,
      "theoretical_loss": 4.189648578861069,
      "tokens_seen": 274792448
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046298896690070214,
      "loss": 3.4048,
      "theoretical_loss": 4.1895296820386525,
      "tokens_seen": 274857984
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046297893681043127,
      "loss": 3.0599,
      "theoretical_loss": 4.189410821497761,
      "tokens_seen": 274923520
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004629689067201605,
      "loss": 3.2,
      "theoretical_loss": 4.189291997218677,
      "tokens_seen": 274989056
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004629588766298897,
      "loss": 3.5654,
      "theoretical_loss": 4.1891732091817016,
      "tokens_seen": 275054592
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046294884653961887,
      "loss": 3.3246,
      "theoretical_loss": 4.189054457367148,
      "tokens_seen": 275120128
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046293881644934805,
      "loss": 3.2228,
      "theoretical_loss": 4.188935741755347,
      "tokens_seen": 275185664
    },
    {
      "epoch": 0.08,
      "objective/train/advantage_avg": 0.4936336576938629,
      "objective/train/docs_used": 167916,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.483511447906494,
      "objective/train/original_loss": 3.483511209487915,
      "objective/train/theoretical_loss": 4.188817062326644,
      "objective/train/tokens_used": 295711200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24374255537986755,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506024360656738,
      "objective/train/weighted_lm_loss": 3.6597180366516113,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0363596677780151,
      "theoretical_loss": 4.188817062326644,
      "tokens_seen": 275251200
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004629287863590773,
      "loss": 3.434,
      "theoretical_loss": 4.188817062326644,
      "tokens_seen": 275251200
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004629187562688064,
      "loss": 3.3444,
      "theoretical_loss": 4.188698419061398,
      "tokens_seen": 275316736
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046290872617853565,
      "loss": 3.3962,
      "theoretical_loss": 4.188579811939988,
      "tokens_seen": 275382272
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046289869608826477,
      "loss": 3.3484,
      "theoretical_loss": 4.188461240942803,
      "tokens_seen": 275447808
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.000462888665997994,
      "loss": 3.1914,
      "theoretical_loss": 4.18834270605025,
      "tokens_seen": 275513344
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004628786359077232,
      "loss": 3.1575,
      "theoretical_loss": 4.188224207242753,
      "tokens_seen": 275578880
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046286860581745237,
      "loss": 3.0535,
      "theoretical_loss": 4.188105744500747,
      "tokens_seen": 275644416
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046285857572718155,
      "loss": 3.2265,
      "theoretical_loss": 4.187987317804685,
      "tokens_seen": 275709952
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046284854563691073,
      "loss": 3.4046,
      "theoretical_loss": 4.187868927135035,
      "tokens_seen": 275775488
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004628385155466399,
      "loss": 3.4597,
      "theoretical_loss": 4.1877505724722806,
      "tokens_seen": 275841024
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046282848545636915,
      "loss": 3.283,
      "theoretical_loss": 4.187632253796918,
      "tokens_seen": 275906560
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004628184553660983,
      "loss": 3.3828,
      "theoretical_loss": 4.187513971089463,
      "tokens_seen": 275972096
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004628084252758275,
      "loss": 3.3772,
      "theoretical_loss": 4.187395724330443,
      "tokens_seen": 276037632
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046279839518555664,
      "loss": 3.2425,
      "theoretical_loss": 4.187277513500401,
      "tokens_seen": 276103168
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004627883650952859,
      "loss": 3.3421,
      "theoretical_loss": 4.187159338579897,
      "tokens_seen": 276168704
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046277833500501505,
      "loss": 3.2867,
      "theoretical_loss": 4.187041199549505,
      "tokens_seen": 276234240
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046276830491474424,
      "loss": 3.0777,
      "theoretical_loss": 4.186923096389812,
      "tokens_seen": 276299776
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004627582748244734,
      "loss": 3.3568,
      "theoretical_loss": 4.186805029081425,
      "tokens_seen": 276365312
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046274824473420265,
      "loss": 3.2269,
      "theoretical_loss": 4.186686997604962,
      "tokens_seen": 276430848
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004627382146439318,
      "loss": 3.2297,
      "theoretical_loss": 4.186569001941057,
      "tokens_seen": 276496384
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.000462728184553661,
      "loss": 3.0867,
      "theoretical_loss": 4.18645104207036,
      "tokens_seen": 276561920
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046271815446339014,
      "loss": 3.319,
      "theoretical_loss": 4.186333117973534,
      "tokens_seen": 276627456
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004627081243731194,
      "loss": 3.4047,
      "theoretical_loss": 4.1862152296312605,
      "tokens_seen": 276692992
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046269809428284856,
      "loss": 3.3944,
      "theoretical_loss": 4.1860973770242325,
      "tokens_seen": 276758528
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046268806419257774,
      "loss": 3.3308,
      "theoretical_loss": 4.185979560133161,
      "tokens_seen": 276824064
    },
    {
      "epoch": 0.08,
      "objective/train/advantage_avg": 0.49245086312294006,
      "objective/train/docs_used": 168545,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.9929311275482178,
      "objective/train/original_loss": 2.992931365966797,
      "objective/train/theoretical_loss": 4.185861778938769,
      "objective/train/tokens_used": 297349600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24263009428977966,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050478458404541,
      "objective/train/weighted_lm_loss": 3.144073009490967,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0338950157165527,
      "theoretical_loss": 4.185861778938769,
      "tokens_seen": 276889600
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004626780341023069,
      "loss": 3.4118,
      "theoretical_loss": 4.185861778938769,
      "tokens_seen": 276889600
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004626680040120361,
      "loss": 3.167,
      "theoretical_loss": 4.185744033421795,
      "tokens_seen": 276955136
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004626579739217653,
      "loss": 3.4273,
      "theoretical_loss": 4.185626323562996,
      "tokens_seen": 277020672
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004626479438314945,
      "loss": 3.336,
      "theoretical_loss": 4.18550864934314,
      "tokens_seen": 277086208
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046263791374122364,
      "loss": 3.3665,
      "theoretical_loss": 4.18539101074301,
      "tokens_seen": 277151744
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004626278836509529,
      "loss": 3.2692,
      "theoretical_loss": 4.185273407743407,
      "tokens_seen": 277217280
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.000462617853560682,
      "loss": 3.3031,
      "theoretical_loss": 4.185155840325143,
      "tokens_seen": 277282816
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046260782347041124,
      "loss": 3.2386,
      "theoretical_loss": 4.185038308469049,
      "tokens_seen": 277348352
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004625977933801404,
      "loss": 3.4065,
      "theoretical_loss": 4.184920812155967,
      "tokens_seen": 277413888
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004625877632898696,
      "loss": 3.3844,
      "theoretical_loss": 4.184803351366757,
      "tokens_seen": 277479424
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004625777331995988,
      "loss": 3.4669,
      "theoretical_loss": 4.184685926082291,
      "tokens_seen": 277544960
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.000462567703109328,
      "loss": 3.4112,
      "theoretical_loss": 4.184568536283457,
      "tokens_seen": 277610496
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046255767301905715,
      "loss": 3.0162,
      "theoretical_loss": 4.184451181951159,
      "tokens_seen": 277676032
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004625476429287864,
      "loss": 3.2547,
      "theoretical_loss": 4.184333863066316,
      "tokens_seen": 277741568
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004625376128385155,
      "loss": 2.9586,
      "theoretical_loss": 4.184216579609857,
      "tokens_seen": 277807104
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046252758274824475,
      "loss": 3.0603,
      "theoretical_loss": 4.184099331562732,
      "tokens_seen": 277872640
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004625175526579739,
      "loss": 3.1834,
      "theoretical_loss": 4.183982118905902,
      "tokens_seen": 277938176
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004625075225677031,
      "loss": 3.2652,
      "theoretical_loss": 4.183864941620344,
      "tokens_seen": 278003712
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004624974924774323,
      "loss": 3.0352,
      "theoretical_loss": 4.183747799687049,
      "tokens_seen": 278069248
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046248746238716147,
      "loss": 3.0865,
      "theoretical_loss": 4.183630693087024,
      "tokens_seen": 278134784
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046247743229689065,
      "loss": 3.5208,
      "theoretical_loss": 4.183513621801289,
      "tokens_seen": 278200320
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004624674022066199,
      "loss": 3.4281,
      "theoretical_loss": 4.183396585810881,
      "tokens_seen": 278265856
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046245737211634907,
      "loss": 3.2548,
      "theoretical_loss": 4.183279585096847,
      "tokens_seen": 278331392
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046244734202607825,
      "loss": 3.0439,
      "theoretical_loss": 4.183162619640255,
      "tokens_seen": 278396928
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004624373119358075,
      "loss": 3.1924,
      "theoretical_loss": 4.1830456894221815,
      "tokens_seen": 278462464
    },
    {
      "epoch": 0.08,
      "objective/train/advantage_avg": 0.49369385838508606,
      "objective/train/docs_used": 169743,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.9155867099761963,
      "objective/train/original_loss": 2.9155869483947754,
      "objective/train/theoretical_loss": 4.182928794423724,
      "objective/train/tokens_used": 298988000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24379518628120422,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.05060875415802,
      "objective/train/weighted_lm_loss": 3.0630016326904297,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0425469875335693,
      "theoretical_loss": 4.182928794423724,
      "tokens_seen": 278528000
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004624272818455366,
      "loss": 3.1675,
      "theoretical_loss": 4.182928794423724,
      "tokens_seen": 278528000
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046241725175526585,
      "loss": 3.1461,
      "theoretical_loss": 4.182811934625987,
      "tokens_seen": 278593536
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.000462407221664995,
      "loss": 3.2707,
      "theoretical_loss": 4.182695110010097,
      "tokens_seen": 278659072
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004623971915747242,
      "loss": 3.1884,
      "theoretical_loss": 4.182578320557189,
      "tokens_seen": 278724608
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004623871614844534,
      "loss": 3.17,
      "theoretical_loss": 4.182461566248418,
      "tokens_seen": 278790144
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046237713139418257,
      "loss": 3.2483,
      "theoretical_loss": 4.1823448470649485,
      "tokens_seen": 278855680
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046236710130391175,
      "loss": 3.2039,
      "theoretical_loss": 4.182228162987963,
      "tokens_seen": 278921216
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046235707121364093,
      "loss": 3.0097,
      "theoretical_loss": 4.182111513998657,
      "tokens_seen": 278986752
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004623470411233701,
      "loss": 3.1308,
      "theoretical_loss": 4.181994900078241,
      "tokens_seen": 279052288
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046233701103309935,
      "loss": 3.0647,
      "theoretical_loss": 4.181878321207939,
      "tokens_seen": 279117824
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004623269809428285,
      "loss": 3.3369,
      "theoretical_loss": 4.18176177736899,
      "tokens_seen": 279183360
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004623169508525577,
      "loss": 3.1524,
      "theoretical_loss": 4.181645268542649,
      "tokens_seen": 279248896
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046230692076228684,
      "loss": 3.1137,
      "theoretical_loss": 4.181528794710183,
      "tokens_seen": 279314432
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004622968906720161,
      "loss": 3.3977,
      "theoretical_loss": 4.181412355852875,
      "tokens_seen": 279379968
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046228686058174525,
      "loss": 3.3142,
      "theoretical_loss": 4.181295951952022,
      "tokens_seen": 279445504
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046227683049147444,
      "loss": 3.3139,
      "theoretical_loss": 4.181179582988934,
      "tokens_seen": 279511040
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004622668004012036,
      "loss": 3.217,
      "theoretical_loss": 4.181063248944938,
      "tokens_seen": 279576576
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046225677031093285,
      "loss": 3.4387,
      "theoretical_loss": 4.180946949801374,
      "tokens_seen": 279642112
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.000462246740220662,
      "loss": 3.3528,
      "theoretical_loss": 4.180830685539596,
      "tokens_seen": 279707648
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004622367101303912,
      "loss": 3.3839,
      "theoretical_loss": 4.180714456140972,
      "tokens_seen": 279773184
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046222668004012034,
      "loss": 3.1407,
      "theoretical_loss": 4.180598261586886,
      "tokens_seen": 279838720
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004622166499498496,
      "loss": 3.1683,
      "theoretical_loss": 4.180482101858734,
      "tokens_seen": 279904256
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046220661985957876,
      "loss": 3.2959,
      "theoretical_loss": 4.18036597693793,
      "tokens_seen": 279969792
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046219658976930794,
      "loss": 3.0762,
      "theoretical_loss": 4.180249886805897,
      "tokens_seen": 280035328
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004621865596790371,
      "loss": 3.372,
      "theoretical_loss": 4.180133831444077,
      "tokens_seen": 280100864
    },
    {
      "epoch": 0.08,
      "objective/train/advantage_avg": 0.4938663840293884,
      "objective/train/docs_used": 170343,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.6836788654327393,
      "objective/train/original_loss": 2.6836791038513184,
      "objective/train/theoretical_loss": 4.180017810833923,
      "objective/train/tokens_used": 300626400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24397186934947968,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506267547607422,
      "objective/train/weighted_lm_loss": 2.819705009460449,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 4.180017810833923,
      "tokens_seen": 280166400
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004621765295887663,
      "loss": 3.2465,
      "theoretical_loss": 4.180017810833923,
      "tokens_seen": 280166400
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004621664994984955,
      "loss": 3.1848,
      "theoretical_loss": 4.179901824956904,
      "tokens_seen": 280231936
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004621564694082247,
      "loss": 3.4641,
      "theoretical_loss": 4.179785873794503,
      "tokens_seen": 280297472
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.00046214643931795384,
      "loss": 3.2272,
      "theoretical_loss": 4.179669957328218,
      "tokens_seen": 280363008
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004621364092276831,
      "loss": 3.4378,
      "theoretical_loss": 4.179554075539558,
      "tokens_seen": 280428544
    },
    {
      "epoch": 0.08,
      "learning_rate": 0.0004621263791374122,
      "loss": 3.3459,
      "theoretical_loss": 4.17943822841005,
      "tokens_seen": 280494080
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00046211634904714144,
      "loss": 3.1166,
      "theoretical_loss": 4.1793224159212325,
      "tokens_seen": 280559616
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004621063189568706,
      "loss": 3.0973,
      "theoretical_loss": 4.1792066380546595,
      "tokens_seen": 280625152
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004620962888665998,
      "loss": 3.1715,
      "theoretical_loss": 4.179090894791899,
      "tokens_seen": 280690688
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.000462086258776329,
      "loss": 3.1837,
      "theoretical_loss": 4.178975186114532,
      "tokens_seen": 280756224
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004620762286860582,
      "loss": 3.3385,
      "theoretical_loss": 4.178859512004156,
      "tokens_seen": 280821760
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00046206619859578735,
      "loss": 2.9198,
      "theoretical_loss": 4.178743872442379,
      "tokens_seen": 280887296
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004620561685055166,
      "loss": 2.9687,
      "theoretical_loss": 4.178628267410826,
      "tokens_seen": 280952832
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004620461384152457,
      "loss": 3.5715,
      "theoretical_loss": 4.178512696891136,
      "tokens_seen": 281018368
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00046203610832497495,
      "loss": 3.4402,
      "theoretical_loss": 4.178397160864961,
      "tokens_seen": 281083904
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004620260782347041,
      "loss": 3.4549,
      "theoretical_loss": 4.178281659313965,
      "tokens_seen": 281149440
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004620160481444333,
      "loss": 3.2466,
      "theoretical_loss": 4.178166192219832,
      "tokens_seen": 281214976
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004620060180541625,
      "loss": 3.0088,
      "theoretical_loss": 4.178050759564254,
      "tokens_seen": 281280512
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00046199598796389167,
      "loss": 3.3511,
      "theoretical_loss": 4.177935361328939,
      "tokens_seen": 281346048
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00046198595787362085,
      "loss": 3.2493,
      "theoretical_loss": 4.17781999749561,
      "tokens_seen": 281411584
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004619759277833501,
      "loss": 3.4212,
      "theoretical_loss": 4.177704668046003,
      "tokens_seen": 281477120
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004619658976930792,
      "loss": 3.2769,
      "theoretical_loss": 4.177589372961869,
      "tokens_seen": 281542656
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00046195586760280845,
      "loss": 3.0635,
      "theoretical_loss": 4.177474112224969,
      "tokens_seen": 281608192
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00046194583751253763,
      "loss": 3.3448,
      "theoretical_loss": 4.177358885817085,
      "tokens_seen": 281673728
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004619358074222668,
      "loss": 3.1526,
      "theoretical_loss": 4.177243693720007,
      "tokens_seen": 281739264
    },
    {
      "epoch": 0.09,
      "objective/train/advantage_avg": 0.49510812759399414,
      "objective/train/docs_used": 171008,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.952528238296509,
      "objective/train/original_loss": 2.952528476715088,
      "objective/train/theoretical_loss": 4.177128535915539,
      "objective/train/tokens_used": 302264800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24523445963859558,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507574081420898,
      "objective/train/weighted_lm_loss": 3.102332830429077,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0338950157165527,
      "theoretical_loss": 4.177128535915539,
      "tokens_seen": 281804800
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.000461925777331996,
      "loss": 3.1273,
      "theoretical_loss": 4.177128535915539,
      "tokens_seen": 281804800
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004619157472417252,
      "loss": 3.1255,
      "theoretical_loss": 4.177013412385504,
      "tokens_seen": 281870336
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00046190571715145435,
      "loss": 3.2387,
      "theoretical_loss": 4.176898323111734,
      "tokens_seen": 281935872
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004618956870611836,
      "loss": 3.2214,
      "theoretical_loss": 4.1767832680760755,
      "tokens_seen": 282001408
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004618856569709127,
      "loss": 3.371,
      "theoretical_loss": 4.176668247260391,
      "tokens_seen": 282066944
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00046187562688064195,
      "loss": 3.2691,
      "theoretical_loss": 4.176553260646555,
      "tokens_seen": 282132480
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004618655967903711,
      "loss": 3.3124,
      "theoretical_loss": 4.176438308216456,
      "tokens_seen": 282198016
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004618555667001003,
      "loss": 3.1317,
      "theoretical_loss": 4.176323389951998,
      "tokens_seen": 282263552
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004618455366098295,
      "loss": 3.2619,
      "theoretical_loss": 4.1762085058350955,
      "tokens_seen": 282329088
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004618355065195587,
      "loss": 3.2733,
      "theoretical_loss": 4.176093655847679,
      "tokens_seen": 282394624
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00046182547642928786,
      "loss": 3.4296,
      "theoretical_loss": 4.175978839971694,
      "tokens_seen": 282460160
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00046181544633901704,
      "loss": 3.0844,
      "theoretical_loss": 4.175864058189096,
      "tokens_seen": 282525696
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004618054162487462,
      "loss": 3.488,
      "theoretical_loss": 4.175749310481859,
      "tokens_seen": 282591232
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00046179538615847545,
      "loss": 3.2821,
      "theoretical_loss": 4.175634596831965,
      "tokens_seen": 282656768
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004617853560682046,
      "loss": 3.1651,
      "theoretical_loss": 4.175519917221415,
      "tokens_seen": 282722304
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004617753259779338,
      "loss": 3.3875,
      "theoretical_loss": 4.17540527163222,
      "tokens_seen": 282787840
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.000461765295887663,
      "loss": 2.9688,
      "theoretical_loss": 4.175290660046407,
      "tokens_seen": 282853376
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004617552657973922,
      "loss": 3.3278,
      "theoretical_loss": 4.175176082446017,
      "tokens_seen": 282918912
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00046174523570712136,
      "loss": 3.3985,
      "theoretical_loss": 4.175061538813101,
      "tokens_seen": 282984448
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00046173520561685054,
      "loss": 3.3514,
      "theoretical_loss": 4.174947029129728,
      "tokens_seen": 283049984
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004617251755265797,
      "loss": 3.1419,
      "theoretical_loss": 4.174832553377978,
      "tokens_seen": 283115520
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00046171514543630896,
      "loss": 3.452,
      "theoretical_loss": 4.174718111539945,
      "tokens_seen": 283181056
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00046170511534603814,
      "loss": 3.086,
      "theoretical_loss": 4.174603703597737,
      "tokens_seen": 283246592
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004616950852557673,
      "loss": 3.3773,
      "theoretical_loss": 4.1744893295334755,
      "tokens_seen": 283312128
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004616850551654965,
      "loss": 3.0911,
      "theoretical_loss": 4.174374989329296,
      "tokens_seen": 283377664
    },
    {
      "epoch": 0.09,
      "objective/train/advantage_avg": 0.4932396411895752,
      "objective/train/docs_used": 172282,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.1940572261810303,
      "objective/train/original_loss": 3.1940574645996094,
      "objective/train/theoretical_loss": 4.174260682967347,
      "objective/train/tokens_used": 303903200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24337509274482727,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505610704421997,
      "objective/train/weighted_lm_loss": 3.355576515197754,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0363596677780151,
      "theoretical_loss": 4.174260682967347,
      "tokens_seen": 283443200
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004616750250752257,
      "loss": 3.1633,
      "theoretical_loss": 4.174260682967347,
      "tokens_seen": 283443200
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004616649949849549,
      "loss": 3.0392,
      "theoretical_loss": 4.17414641042979,
      "tokens_seen": 283508736
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00046165496489468404,
      "loss": 3.3046,
      "theoretical_loss": 4.1740321716988005,
      "tokens_seen": 283574272
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004616449348044133,
      "loss": 3.3233,
      "theoretical_loss": 4.1739179667565685,
      "tokens_seen": 283639808
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004616349047141424,
      "loss": 3.1206,
      "theoretical_loss": 4.173803795585296,
      "tokens_seen": 283705344
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00046162487462387164,
      "loss": 2.9978,
      "theoretical_loss": 4.173689658167199,
      "tokens_seen": 283770880
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004616148445336008,
      "loss": 3.383,
      "theoretical_loss": 4.173575554484509,
      "tokens_seen": 283836416
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00046160481444333,
      "loss": 2.9492,
      "theoretical_loss": 4.173461484519466,
      "tokens_seen": 283901952
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004615947843530592,
      "loss": 3.2222,
      "theoretical_loss": 4.1733474482543285,
      "tokens_seen": 283967488
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004615847542627884,
      "loss": 3.3725,
      "theoretical_loss": 4.173233445671365,
      "tokens_seen": 284033024
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00046157472417251755,
      "loss": 3.2613,
      "theoretical_loss": 4.173119476752861,
      "tokens_seen": 284098560
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004615646940822468,
      "loss": 3.1465,
      "theoretical_loss": 4.173005541481111,
      "tokens_seen": 284164096
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004615546639919759,
      "loss": 3.123,
      "theoretical_loss": 4.1728916398384275,
      "tokens_seen": 284229632
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00046154463390170515,
      "loss": 3.3936,
      "theoretical_loss": 4.172777771807132,
      "tokens_seen": 284295168
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004615346038114343,
      "loss": 3.3148,
      "theoretical_loss": 4.1726639373695615,
      "tokens_seen": 284360704
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004615245737211635,
      "loss": 2.948,
      "theoretical_loss": 4.172550136508066,
      "tokens_seen": 284426240
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004615145436308927,
      "loss": 3.4253,
      "theoretical_loss": 4.172436369205011,
      "tokens_seen": 284491776
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00046150451354062187,
      "loss": 3.3413,
      "theoretical_loss": 4.172322635442772,
      "tokens_seen": 284557312
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00046149448345035105,
      "loss": 3.5713,
      "theoretical_loss": 4.172208935203739,
      "tokens_seen": 284622848
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004614844533600803,
      "loss": 3.1813,
      "theoretical_loss": 4.172095268470316,
      "tokens_seen": 284688384
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004614744232698094,
      "loss": 3.0591,
      "theoretical_loss": 4.1719816352249195,
      "tokens_seen": 284753920
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00046146439317953865,
      "loss": 2.921,
      "theoretical_loss": 4.17186803544998,
      "tokens_seen": 284819456
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00046145436308926783,
      "loss": 3.1219,
      "theoretical_loss": 4.17175446912794,
      "tokens_seen": 284884992
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.000461444332998997,
      "loss": 3.2106,
      "theoretical_loss": 4.171640936241256,
      "tokens_seen": 284950528
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004614343029087262,
      "loss": 3.0987,
      "theoretical_loss": 4.171527436772399,
      "tokens_seen": 285016064
    },
    {
      "epoch": 0.09,
      "objective/train/advantage_avg": 0.4917828440666199,
      "objective/train/docs_used": 172985,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.203778028488159,
      "objective/train/original_loss": 3.20377779006958,
      "objective/train/theoretical_loss": 4.171413970703851,
      "objective/train/tokens_used": 305541600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24194203317165375,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050408124923706,
      "objective/train/weighted_lm_loss": 3.3651092052459717,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0417574644088745,
      "theoretical_loss": 4.171413970703851,
      "tokens_seen": 285081600
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004614242728184554,
      "loss": 3.207,
      "theoretical_loss": 4.171413970703851,
      "tokens_seen": 285081600
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00046141424272818455,
      "loss": 2.969,
      "theoretical_loss": 4.171300538018109,
      "tokens_seen": 285147136
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004614042126379138,
      "loss": 3.4524,
      "theoretical_loss": 4.1711871386976815,
      "tokens_seen": 285212672
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004613941825476429,
      "loss": 3.1723,
      "theoretical_loss": 4.171073772725091,
      "tokens_seen": 285278208
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00046138415245737215,
      "loss": 3.5167,
      "theoretical_loss": 4.170960440082874,
      "tokens_seen": 285343744
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004613741223671013,
      "loss": 3.3531,
      "theoretical_loss": 4.17084714075358,
      "tokens_seen": 285409280
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004613640922768305,
      "loss": 3.1496,
      "theoretical_loss": 4.170733874719769,
      "tokens_seen": 285474816
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004613540621865597,
      "loss": 3.2564,
      "theoretical_loss": 4.1706206419640175,
      "tokens_seen": 285540352
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004613440320962889,
      "loss": 3.4347,
      "theoretical_loss": 4.170507442468914,
      "tokens_seen": 285605888
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00046133400200601806,
      "loss": 3.2164,
      "theoretical_loss": 4.170394276217059,
      "tokens_seen": 285671424
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00046132397191574724,
      "loss": 3.3484,
      "theoretical_loss": 4.170281143191067,
      "tokens_seen": 285736960
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004613139418254764,
      "loss": 2.9886,
      "theoretical_loss": 4.170168043373567,
      "tokens_seen": 285802496
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00046130391173520566,
      "loss": 3.3677,
      "theoretical_loss": 4.170054976747199,
      "tokens_seen": 285868032
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004612938816449348,
      "loss": 3.402,
      "theoretical_loss": 4.169941943294615,
      "tokens_seen": 285933568
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.000461283851554664,
      "loss": 3.1542,
      "theoretical_loss": 4.1698289429984845,
      "tokens_seen": 285999104
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004612738214643932,
      "loss": 3.0731,
      "theoretical_loss": 4.169715975841485,
      "tokens_seen": 286064640
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004612637913741224,
      "loss": 3.3602,
      "theoretical_loss": 4.16960304180631,
      "tokens_seen": 286130176
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00046125376128385156,
      "loss": 3.3458,
      "theoretical_loss": 4.169490140875665,
      "tokens_seen": 286195712
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00046124373119358074,
      "loss": 3.1519,
      "theoretical_loss": 4.16937727303227,
      "tokens_seen": 286261248
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004612337011033099,
      "loss": 3.0964,
      "theoretical_loss": 4.169264438258854,
      "tokens_seen": 286326784
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00046122367101303916,
      "loss": 3.1504,
      "theoretical_loss": 4.169151636538165,
      "tokens_seen": 286392320
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004612136409227683,
      "loss": 3.4792,
      "theoretical_loss": 4.169038867852958,
      "tokens_seen": 286457856
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004612036108324975,
      "loss": 3.0519,
      "theoretical_loss": 4.168926132186005,
      "tokens_seen": 286523392
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00046119358074222665,
      "loss": 3.2737,
      "theoretical_loss": 4.16881342952009,
      "tokens_seen": 286588928
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004611835506519559,
      "loss": 3.2422,
      "theoretical_loss": 4.168700759838007,
      "tokens_seen": 286654464
    },
    {
      "epoch": 0.09,
      "objective/train/advantage_avg": 0.4953973591327667,
      "objective/train/docs_used": 174327,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.775333881378174,
      "objective/train/original_loss": 2.7753334045410156,
      "objective/train/theoretical_loss": 4.168588123122568,
      "objective/train/tokens_used": 307180000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24546678364276886,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507874488830566,
      "objective/train/weighted_lm_loss": 2.916236639022827,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0417574644088745,
      "theoretical_loss": 4.168588123122568,
      "tokens_seen": 286720000
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00046117352056168506,
      "loss": 3.3224,
      "theoretical_loss": 4.168588123122568,
      "tokens_seen": 286720000
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00046116349047141425,
      "loss": 3.1143,
      "theoretical_loss": 4.168475519356592,
      "tokens_seen": 286785536
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004611534603811434,
      "loss": 3.363,
      "theoretical_loss": 4.168362948522918,
      "tokens_seen": 286851072
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004611434302908726,
      "loss": 3.3857,
      "theoretical_loss": 4.168250410604391,
      "tokens_seen": 286916608
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004611334002006018,
      "loss": 3.1504,
      "theoretical_loss": 4.1681379055838725,
      "tokens_seen": 286982144
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.000461123370110331,
      "loss": 3.3721,
      "theoretical_loss": 4.168025433444237,
      "tokens_seen": 287047680
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00046111334002006015,
      "loss": 3.3487,
      "theoretical_loss": 4.16791299416837,
      "tokens_seen": 287113216
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004611033099297894,
      "loss": 3.1417,
      "theoretical_loss": 4.167800587739171,
      "tokens_seen": 287178752
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00046109327983951857,
      "loss": 3.1342,
      "theoretical_loss": 4.167688214139552,
      "tokens_seen": 287244288
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00046108324974924775,
      "loss": 3.4337,
      "theoretical_loss": 4.167575873352437,
      "tokens_seen": 287309824
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00046107321965897693,
      "loss": 3.1221,
      "theoretical_loss": 4.167463565360766,
      "tokens_seen": 287375360
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004610631895687061,
      "loss": 3.0469,
      "theoretical_loss": 4.1673512901474865,
      "tokens_seen": 287440896
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004610531594784353,
      "loss": 3.097,
      "theoretical_loss": 4.1672390476955625,
      "tokens_seen": 287506432
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004610431293881645,
      "loss": 3.3626,
      "theoretical_loss": 4.167126837987971,
      "tokens_seen": 287571968
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00046103309929789365,
      "loss": 3.4152,
      "theoretical_loss": 4.167014661007699,
      "tokens_seen": 287637504
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004610230692076229,
      "loss": 3.2178,
      "theoretical_loss": 4.16690251673775,
      "tokens_seen": 287703040
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.000461013039117352,
      "loss": 3.1857,
      "theoretical_loss": 4.1667904051611355,
      "tokens_seen": 287768576
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00046100300902708125,
      "loss": 3.0459,
      "theoretical_loss": 4.166678326260882,
      "tokens_seen": 287834112
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00046099297893681043,
      "loss": 3.3215,
      "theoretical_loss": 4.166566280020032,
      "tokens_seen": 287899648
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004609829488465396,
      "loss": 3.5272,
      "theoretical_loss": 4.166454266421635,
      "tokens_seen": 287965184
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004609729187562688,
      "loss": 3.1807,
      "theoretical_loss": 4.1663422854487555,
      "tokens_seen": 288030720
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00046096288866599803,
      "loss": 2.9599,
      "theoretical_loss": 4.166230337084471,
      "tokens_seen": 288096256
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004609528585757272,
      "loss": 3.2918,
      "theoretical_loss": 4.166118421311873,
      "tokens_seen": 288161792
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004609428284854564,
      "loss": 3.5227,
      "theoretical_loss": 4.166006538114062,
      "tokens_seen": 288227328
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004609327983951856,
      "loss": 3.0671,
      "theoretical_loss": 4.165894687474155,
      "tokens_seen": 288292864
    },
    {
      "epoch": 0.09,
      "objective/train/advantage_avg": 0.49518096446990967,
      "objective/train/docs_used": 174865,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.859485149383545,
      "objective/train/original_loss": 2.859485149383545,
      "objective/train/theoretical_loss": 4.165782869375278,
      "objective/train/tokens_used": 308818400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24526914954185486,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507649183273315,
      "objective/train/weighted_lm_loss": 3.004514694213867,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0442860126495361,
      "theoretical_loss": 4.165782869375278,
      "tokens_seen": 288358400
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00046092276830491475,
      "loss": 3.1043,
      "theoretical_loss": 4.165782869375278,
      "tokens_seen": 288358400
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.000460912738214644,
      "loss": 3.0814,
      "theoretical_loss": 4.1656710838005715,
      "tokens_seen": 288423936
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004609027081243731,
      "loss": 3.4661,
      "theoretical_loss": 4.165559330733188,
      "tokens_seen": 288489472
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00046089267803410235,
      "loss": 3.0244,
      "theoretical_loss": 4.165447610156296,
      "tokens_seen": 288555008
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004608826479438315,
      "loss": 3.2406,
      "theoretical_loss": 4.165335922053069,
      "tokens_seen": 288620544
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004608726178535607,
      "loss": 3.2875,
      "theoretical_loss": 4.165224266406701,
      "tokens_seen": 288686080
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004608625877632899,
      "loss": 3.1823,
      "theoretical_loss": 4.165112643200393,
      "tokens_seen": 288751616
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004608525576730191,
      "loss": 3.3071,
      "theoretical_loss": 4.165001052417361,
      "tokens_seen": 288817152
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00046084252758274826,
      "loss": 3.2087,
      "theoretical_loss": 4.164889494040834,
      "tokens_seen": 288882688
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00046083249749247744,
      "loss": 3.0823,
      "theoretical_loss": 4.16477796805405,
      "tokens_seen": 288948224
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004608224674022066,
      "loss": 3.1614,
      "theoretical_loss": 4.1646664744402635,
      "tokens_seen": 289013760
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00046081243731193586,
      "loss": 2.9463,
      "theoretical_loss": 4.16455501318274,
      "tokens_seen": 289079296
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.000460802407221665,
      "loss": 3.2557,
      "theoretical_loss": 4.164443584264758,
      "tokens_seen": 289144832
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004607923771313942,
      "loss": 3.0191,
      "theoretical_loss": 4.164332187669605,
      "tokens_seen": 289210368
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004607823470411234,
      "loss": 3.2773,
      "theoretical_loss": 4.164220823380587,
      "tokens_seen": 289275904
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004607723169508526,
      "loss": 3.129,
      "theoretical_loss": 4.164109491381017,
      "tokens_seen": 289341440
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00046076228686058176,
      "loss": 3.0788,
      "theoretical_loss": 4.163998191654223,
      "tokens_seen": 289406976
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00046075225677031094,
      "loss": 3.256,
      "theoretical_loss": 4.163886924183544,
      "tokens_seen": 289472512
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004607422266800401,
      "loss": 3.3299,
      "theoretical_loss": 4.163775688952334,
      "tokens_seen": 289538048
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00046073219658976936,
      "loss": 2.9007,
      "theoretical_loss": 4.163664485943956,
      "tokens_seen": 289603584
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004607221664994985,
      "loss": 3.0751,
      "theoretical_loss": 4.163553315141788,
      "tokens_seen": 289669120
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004607121364092277,
      "loss": 3.3586,
      "theoretical_loss": 4.163442176529218,
      "tokens_seen": 289734656
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00046070210631895685,
      "loss": 3.1344,
      "theoretical_loss": 4.163331070089649,
      "tokens_seen": 289800192
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004606920762286861,
      "loss": 3.0003,
      "theoretical_loss": 4.163219995806494,
      "tokens_seen": 289865728
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00046068204613841526,
      "loss": 3.3748,
      "theoretical_loss": 4.163108953663179,
      "tokens_seen": 289931264
    },
    {
      "epoch": 0.09,
      "objective/train/advantage_avg": 0.49260759353637695,
      "objective/train/docs_used": 176137,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.9476935863494873,
      "objective/train/original_loss": 2.947693109512329,
      "objective/train/theoretical_loss": 4.162997943643143,
      "objective/train/tokens_used": 310456800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24272336065769196,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0504944324493408,
      "objective/train/weighted_lm_loss": 3.0964951515197754,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0437887907028198,
      "theoretical_loss": 4.162997943643143,
      "tokens_seen": 289996800
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00046067201604814445,
      "loss": 3.2034,
      "theoretical_loss": 4.162997943643143,
      "tokens_seen": 289996800
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004606619859578736,
      "loss": 3.3927,
      "theoretical_loss": 4.162886965729837,
      "tokens_seen": 290062336
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004606519558676028,
      "loss": 3.2285,
      "theoretical_loss": 4.162776019906723,
      "tokens_seen": 290127872
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.000460641925777332,
      "loss": 3.4212,
      "theoretical_loss": 4.162665106157277,
      "tokens_seen": 290193408
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004606318956870612,
      "loss": 3.3201,
      "theoretical_loss": 4.162554224464987,
      "tokens_seen": 290258944
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00046062186559679035,
      "loss": 3.3027,
      "theoretical_loss": 4.162443374813352,
      "tokens_seen": 290324480
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004606118355065196,
      "loss": 3.1967,
      "theoretical_loss": 4.1623325571858825,
      "tokens_seen": 290390016
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00046060180541624877,
      "loss": 2.9457,
      "theoretical_loss": 4.162221771566105,
      "tokens_seen": 290455552
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00046059177532597795,
      "loss": 3.2158,
      "theoretical_loss": 4.162111017937555,
      "tokens_seen": 290521088
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00046058174523570713,
      "loss": 3.1727,
      "theoretical_loss": 4.1620002962837805,
      "tokens_seen": 290586624
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004605717151454363,
      "loss": 3.1289,
      "theoretical_loss": 4.161889606588343,
      "tokens_seen": 290652160
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004605616850551655,
      "loss": 3.113,
      "theoretical_loss": 4.161778948834815,
      "tokens_seen": 290717696
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004605516549648947,
      "loss": 3.2518,
      "theoretical_loss": 4.161668323006781,
      "tokens_seen": 290783232
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00046054162487462385,
      "loss": 3.075,
      "theoretical_loss": 4.161557729087839,
      "tokens_seen": 290848768
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004605315947843531,
      "loss": 3.1984,
      "theoretical_loss": 4.161447167061597,
      "tokens_seen": 290914304
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004605215646940822,
      "loss": 2.9163,
      "theoretical_loss": 4.1613366369116775,
      "tokens_seen": 290979840
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00046051153460381145,
      "loss": 3.0189,
      "theoretical_loss": 4.161226138621714,
      "tokens_seen": 291045376
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00046050150451354063,
      "loss": 3.0326,
      "theoretical_loss": 4.1611156721753515,
      "tokens_seen": 291110912
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004604914744232698,
      "loss": 3.1866,
      "theoretical_loss": 4.161005237556247,
      "tokens_seen": 291176448
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.000460481444332999,
      "loss": 3.1874,
      "theoretical_loss": 4.160894834748071,
      "tokens_seen": 291241984
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00046047141424272823,
      "loss": 3.1885,
      "theoretical_loss": 4.1607844637345055,
      "tokens_seen": 291307520
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00046046138415245736,
      "loss": 3.3941,
      "theoretical_loss": 4.160674124499243,
      "tokens_seen": 291373056
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004604513540621866,
      "loss": 3.3172,
      "theoretical_loss": 4.160563817025991,
      "tokens_seen": 291438592
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004604413239719157,
      "loss": 3.3411,
      "theoretical_loss": 4.160453541298465,
      "tokens_seen": 291504128
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00046043129388164495,
      "loss": 3.4265,
      "theoretical_loss": 4.160343297300397,
      "tokens_seen": 291569664
    },
    {
      "epoch": 0.09,
      "objective/train/advantage_avg": 0.49207547307014465,
      "objective/train/docs_used": 176478,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.7176215648651123,
      "objective/train/original_loss": 2.717621326446533,
      "objective/train/theoretical_loss": 4.160233085015529,
      "objective/train/tokens_used": 312095200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24225589632987976,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0504388809204102,
      "objective/train/weighted_lm_loss": 2.854548692703247,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0402631759643555,
      "theoretical_loss": 4.160233085015529,
      "tokens_seen": 291635200
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00046042126379137414,
      "loss": 3.143,
      "theoretical_loss": 4.160233085015529,
      "tokens_seen": 291635200
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004604112337011033,
      "loss": 3.3143,
      "theoretical_loss": 4.160122904427611,
      "tokens_seen": 291700736
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004604012036108325,
      "loss": 3.4049,
      "theoretical_loss": 4.160012755520413,
      "tokens_seen": 291766272
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004603911735205617,
      "loss": 3.181,
      "theoretical_loss": 4.159902638277712,
      "tokens_seen": 291831808
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00046038114343029086,
      "loss": 3.3577,
      "theoretical_loss": 4.159792552683296,
      "tokens_seen": 291897344
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004603711133400201,
      "loss": 3.3411,
      "theoretical_loss": 4.159682498720967,
      "tokens_seen": 291962880
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004603610832497492,
      "loss": 3.2096,
      "theoretical_loss": 4.15957247637454,
      "tokens_seen": 292028416
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00046035105315947846,
      "loss": 3.227,
      "theoretical_loss": 4.159462485627838,
      "tokens_seen": 292093952
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004603410230692076,
      "loss": 3.1602,
      "theoretical_loss": 4.1593525264647,
      "tokens_seen": 292159488
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004603309929789368,
      "loss": 3.3422,
      "theoretical_loss": 4.159242598868976,
      "tokens_seen": 292225024
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.000460320962888666,
      "loss": 3.5436,
      "theoretical_loss": 4.159132702824524,
      "tokens_seen": 292290560
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004603109327983952,
      "loss": 3.4198,
      "theoretical_loss": 4.15902283831522,
      "tokens_seen": 292356096
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00046030090270812436,
      "loss": 3.1836,
      "theoretical_loss": 4.158913005324949,
      "tokens_seen": 292421632
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004602908726178536,
      "loss": 3.2477,
      "theoretical_loss": 4.158803203837605,
      "tokens_seen": 292487168
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004602808425275827,
      "loss": 3.2248,
      "theoretical_loss": 4.158693433837098,
      "tokens_seen": 292552704
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00046027081243731196,
      "loss": 3.3148,
      "theoretical_loss": 4.158583695307349,
      "tokens_seen": 292618240
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004602607823470411,
      "loss": 3.5238,
      "theoretical_loss": 4.158473988232288,
      "tokens_seen": 292683776
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004602507522567703,
      "loss": 3.1445,
      "theoretical_loss": 4.158364312595862,
      "tokens_seen": 292749312
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004602407221664995,
      "loss": 3.1778,
      "theoretical_loss": 4.158254668382024,
      "tokens_seen": 292814848
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004602306920762287,
      "loss": 3.5461,
      "theoretical_loss": 4.158145055574742,
      "tokens_seen": 292880384
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00046022066198595787,
      "loss": 2.9906,
      "theoretical_loss": 4.158035474157996,
      "tokens_seen": 292945920
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00046021063189568705,
      "loss": 3.3653,
      "theoretical_loss": 4.157925924115778,
      "tokens_seen": 293011456
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004602006018054163,
      "loss": 3.2421,
      "theoretical_loss": 4.157816405432088,
      "tokens_seen": 293076992
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00046019057171514546,
      "loss": 3.1906,
      "theoretical_loss": 4.157706918090943,
      "tokens_seen": 293142528
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00046018054162487465,
      "loss": 3.2258,
      "theoretical_loss": 4.157597462076368,
      "tokens_seen": 293208064
    },
    {
      "epoch": 0.09,
      "objective/train/advantage_avg": 0.4939468801021576,
      "objective/train/docs_used": 177825,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.9082913398742676,
      "objective/train/original_loss": 2.9082915782928467,
      "objective/train/theoretical_loss": 4.157488037372401,
      "objective/train/tokens_used": 313733600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2441297173500061,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506356954574585,
      "objective/train/weighted_lm_loss": 3.055884838104248,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0440459251403809,
      "theoretical_loss": 4.157488037372401,
      "tokens_seen": 293273600
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004601705115346038,
      "loss": 3.0193,
      "theoretical_loss": 4.157488037372401,
      "tokens_seen": 293273600
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.000460160481444333,
      "loss": 3.1495,
      "theoretical_loss": 4.157378643963092,
      "tokens_seen": 293339136
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004601504513540622,
      "loss": 3.3651,
      "theoretical_loss": 4.157269281832502,
      "tokens_seen": 293404672
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004601404212637914,
      "loss": 3.3585,
      "theoretical_loss": 4.157159950964704,
      "tokens_seen": 293470208
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00046013039117352055,
      "loss": 3.1344,
      "theoretical_loss": 4.157050651343782,
      "tokens_seen": 293535744
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004601203610832498,
      "loss": 3.3926,
      "theoretical_loss": 4.156941382953835,
      "tokens_seen": 293601280
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00046011033099297897,
      "loss": 3.409,
      "theoretical_loss": 4.156832145778967,
      "tokens_seen": 293666816
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00046010030090270815,
      "loss": 2.9701,
      "theoretical_loss": 4.1567229398033,
      "tokens_seen": 293732352
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00046009027081243733,
      "loss": 3.3238,
      "theoretical_loss": 4.156613765010965,
      "tokens_seen": 293797888
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004600802407221665,
      "loss": 3.0614,
      "theoretical_loss": 4.156504621386103,
      "tokens_seen": 293863424
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004600702106318957,
      "loss": 3.2344,
      "theoretical_loss": 4.156395508912872,
      "tokens_seen": 293928960
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00046006018054162493,
      "loss": 2.9499,
      "theoretical_loss": 4.156286427575434,
      "tokens_seen": 293994496
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00046005015045135405,
      "loss": 2.8901,
      "theoretical_loss": 4.156177377357969,
      "tokens_seen": 294060032
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004600401203610833,
      "loss": 3.4233,
      "theoretical_loss": 4.156068358244665,
      "tokens_seen": 294125568
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004600300902708124,
      "loss": 3.3287,
      "theoretical_loss": 4.155959370219724,
      "tokens_seen": 294191104
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00046002006018054165,
      "loss": 3.4622,
      "theoretical_loss": 4.155850413267357,
      "tokens_seen": 294256640
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00046001003009027083,
      "loss": 3.2243,
      "theoretical_loss": 4.155741487371789,
      "tokens_seen": 294322176
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00046,
      "loss": 3.0126,
      "theoretical_loss": 4.155632592517254,
      "tokens_seen": 294387712
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004599899699097292,
      "loss": 3.3283,
      "theoretical_loss": 4.1555237286879985,
      "tokens_seen": 294453248
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045997993981945843,
      "loss": 3.4396,
      "theoretical_loss": 4.155414895868282,
      "tokens_seen": 294518784
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045996990972918756,
      "loss": 3.3629,
      "theoretical_loss": 4.155306094042373,
      "tokens_seen": 294584320
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004599598796389168,
      "loss": 3.3268,
      "theoretical_loss": 4.155197323194555,
      "tokens_seen": 294649856
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004599498495486459,
      "loss": 3.0429,
      "theoretical_loss": 4.155088583309118,
      "tokens_seen": 294715392
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045993981945837515,
      "loss": 3.0929,
      "theoretical_loss": 4.154979874370367,
      "tokens_seen": 294780928
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045992978936810434,
      "loss": 3.3871,
      "theoretical_loss": 4.154871196362619,
      "tokens_seen": 294846464
    },
    {
      "epoch": 0.09,
      "objective/train/advantage_avg": 0.4941478371620178,
      "objective/train/docs_used": 179027,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.354177713394165,
      "objective/train/original_loss": 3.354177474975586,
      "objective/train/theoretical_loss": 4.154762549270199,
      "objective/train/tokens_used": 315372000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24423769116401672,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506564378738403,
      "objective/train/weighted_lm_loss": 3.523916244506836,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0417574644088745,
      "theoretical_loss": 4.154762549270199,
      "tokens_seen": 294912000
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004599197592778335,
      "loss": 3.2319,
      "theoretical_loss": 4.154762549270199,
      "tokens_seen": 294912000
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004599097291875627,
      "loss": 3.1799,
      "theoretical_loss": 4.154653933077447,
      "tokens_seen": 294977536
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004598996990972919,
      "loss": 3.4499,
      "theoretical_loss": 4.154545347768712,
      "tokens_seen": 295043072
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045988966900702106,
      "loss": 3.3284,
      "theoretical_loss": 4.1544367933283555,
      "tokens_seen": 295108608
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004598796389167503,
      "loss": 3.0164,
      "theoretical_loss": 4.154328269740751,
      "tokens_seen": 295174144
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004598696088264794,
      "loss": 3.2188,
      "theoretical_loss": 4.154219776990281,
      "tokens_seen": 295239680
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045985957873620866,
      "loss": 3.2124,
      "theoretical_loss": 4.154111315061342,
      "tokens_seen": 295305216
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004598495486459378,
      "loss": 3.2435,
      "theoretical_loss": 4.1540028839383405,
      "tokens_seen": 295370752
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.000459839518555667,
      "loss": 3.3532,
      "theoretical_loss": 4.153894483605694,
      "tokens_seen": 295436288
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004598294884653962,
      "loss": 3.4898,
      "theoretical_loss": 4.153786114047834,
      "tokens_seen": 295501824
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004598194583751254,
      "loss": 3.1416,
      "theoretical_loss": 4.153677775249198,
      "tokens_seen": 295567360
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045980942828485456,
      "loss": 3.4476,
      "theoretical_loss": 4.153569467194241,
      "tokens_seen": 295632896
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004597993981945838,
      "loss": 3.3665,
      "theoretical_loss": 4.153461189867425,
      "tokens_seen": 295698432
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004597893681043129,
      "loss": 3.0998,
      "theoretical_loss": 4.153352943253225,
      "tokens_seen": 295763968
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045977933801404216,
      "loss": 3.2748,
      "theoretical_loss": 4.153244727336126,
      "tokens_seen": 295829504
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004597693079237713,
      "loss": 3.3594,
      "theoretical_loss": 4.153136542100627,
      "tokens_seen": 295895040
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004597592778335005,
      "loss": 2.9525,
      "theoretical_loss": 4.153028387531236,
      "tokens_seen": 295960576
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004597492477432297,
      "loss": 3.3239,
      "theoretical_loss": 4.152920263612472,
      "tokens_seen": 296026112
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004597392176529589,
      "loss": 3.1925,
      "theoretical_loss": 4.1528121703288665,
      "tokens_seen": 296091648
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045972918756268807,
      "loss": 3.1402,
      "theoretical_loss": 4.152704107664961,
      "tokens_seen": 296157184
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045971915747241725,
      "loss": 3.2106,
      "theoretical_loss": 4.15259607560531,
      "tokens_seen": 296222720
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045970912738214643,
      "loss": 3.0982,
      "theoretical_loss": 4.152488074134478,
      "tokens_seen": 296288256
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045969909729187566,
      "loss": 3.2595,
      "theoretical_loss": 4.15238010323704,
      "tokens_seen": 296353792
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004596890672016048,
      "loss": 3.2504,
      "theoretical_loss": 4.152272162897584,
      "tokens_seen": 296419328
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.000459679037111334,
      "loss": 3.224,
      "theoretical_loss": 4.1521642531007075,
      "tokens_seen": 296484864
    },
    {
      "epoch": 0.09,
      "objective/train/advantage_avg": 0.49464574456214905,
      "objective/train/docs_used": 179648,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.4784085750579834,
      "objective/train/original_loss": 3.4784085750579834,
      "objective/train/theoretical_loss": 4.15205637383102,
      "objective/train/tokens_used": 317010400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.244748055934906,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507086515426636,
      "objective/train/weighted_lm_loss": 3.6547677516937256,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0338950157165527,
      "theoretical_loss": 4.15205637383102,
      "tokens_seen": 296550400
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045966900702106315,
      "loss": 3.2406,
      "theoretical_loss": 4.15205637383102,
      "tokens_seen": 296550400
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004596589769307924,
      "loss": 3.3079,
      "theoretical_loss": 4.151948525073142,
      "tokens_seen": 296615936
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045964894684052157,
      "loss": 2.9897,
      "theoretical_loss": 4.151840706811706,
      "tokens_seen": 296681472
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045963891675025075,
      "loss": 3.1013,
      "theoretical_loss": 4.151732919031354,
      "tokens_seen": 296747008
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045962888665997993,
      "loss": 3.2655,
      "theoretical_loss": 4.15162516171674,
      "tokens_seen": 296812544
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045961885656970917,
      "loss": 3.2057,
      "theoretical_loss": 4.151517434852528,
      "tokens_seen": 296878080
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004596088264794383,
      "loss": 3.2951,
      "theoretical_loss": 4.151409738423396,
      "tokens_seen": 296943616
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045959879638916753,
      "loss": 3.2759,
      "theoretical_loss": 4.151302072414031,
      "tokens_seen": 297009152
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045958876629889666,
      "loss": 3.2748,
      "theoretical_loss": 4.15119443680913,
      "tokens_seen": 297074688
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004595787362086259,
      "loss": 3.4099,
      "theoretical_loss": 4.151086831593402,
      "tokens_seen": 297140224
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004595687061183551,
      "loss": 3.1563,
      "theoretical_loss": 4.15097925675157,
      "tokens_seen": 297205760
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045955867602808425,
      "loss": 3.3395,
      "theoretical_loss": 4.150871712268364,
      "tokens_seen": 297271296
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045954864593781344,
      "loss": 3.1918,
      "theoretical_loss": 4.150764198128526,
      "tokens_seen": 297336832
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004595386158475426,
      "loss": 3.444,
      "theoretical_loss": 4.15065671431681,
      "tokens_seen": 297402368
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004595285857572718,
      "loss": 3.3078,
      "theoretical_loss": 4.15054926081798,
      "tokens_seen": 297467904
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045951855566700103,
      "loss": 3.125,
      "theoretical_loss": 4.150441837616814,
      "tokens_seen": 297533440
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045950852557673016,
      "loss": 2.9565,
      "theoretical_loss": 4.150334444698096,
      "tokens_seen": 297598976
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004594984954864594,
      "loss": 3.1185,
      "theoretical_loss": 4.150227082046624,
      "tokens_seen": 297664512
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004594884653961885,
      "loss": 2.9613,
      "theoretical_loss": 4.150119749647207,
      "tokens_seen": 297730048
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045947843530591776,
      "loss": 3.1304,
      "theoretical_loss": 4.150012447484665,
      "tokens_seen": 297795584
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045946840521564694,
      "loss": 3.3143,
      "theoretical_loss": 4.149905175543827,
      "tokens_seen": 297861120
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004594583751253761,
      "loss": 3.1726,
      "theoretical_loss": 4.149797933809536,
      "tokens_seen": 297926656
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045944834503510536,
      "loss": 3.2535,
      "theoretical_loss": 4.149690722266644,
      "tokens_seen": 297992192
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045943831494483454,
      "loss": 3.2073,
      "theoretical_loss": 4.1495835409000135,
      "tokens_seen": 298057728
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004594282848545637,
      "loss": 3.1918,
      "theoretical_loss": 4.149476389694518,
      "tokens_seen": 298123264
    },
    {
      "epoch": 0.09,
      "objective/train/advantage_avg": 0.4930114150047302,
      "objective/train/docs_used": 180307,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.3320181369781494,
      "objective/train/original_loss": 3.3320186138153076,
      "objective/train/theoretical_loss": 4.149369268635046,
      "objective/train/tokens_used": 318648800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24316097795963287,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505372285842896,
      "objective/train/weighted_lm_loss": 3.5004477500915527,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0408108234405518,
      "theoretical_loss": 4.149369268635046,
      "tokens_seen": 298188800
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004594182547642929,
      "loss": 3.1023,
      "theoretical_loss": 4.149369268635046,
      "tokens_seen": 298188800
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004594082246740221,
      "loss": 3.1592,
      "theoretical_loss": 4.14926217770649,
      "tokens_seen": 298254336
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045939819458375126,
      "loss": 3.3868,
      "theoretical_loss": 4.149155116893757,
      "tokens_seen": 298319872
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004593881644934805,
      "loss": 3.0153,
      "theoretical_loss": 4.149048086181768,
      "tokens_seen": 298385408
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004593781344032096,
      "loss": 2.8557,
      "theoretical_loss": 4.148941085555448,
      "tokens_seen": 298450944
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045936810431293886,
      "loss": 3.0557,
      "theoretical_loss": 4.1488341149997385,
      "tokens_seen": 298516480
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.000459358074222668,
      "loss": 3.3412,
      "theoretical_loss": 4.148727174499589,
      "tokens_seen": 298582016
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004593480441323972,
      "loss": 3.3068,
      "theoretical_loss": 4.148620264039961,
      "tokens_seen": 298647552
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004593380140421264,
      "loss": 3.2231,
      "theoretical_loss": 4.148513383605826,
      "tokens_seen": 298713088
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004593279839518556,
      "loss": 3.1369,
      "theoretical_loss": 4.1484065331821665,
      "tokens_seen": 298778624
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045931795386158476,
      "loss": 3.1351,
      "theoretical_loss": 4.148299712753977,
      "tokens_seen": 298844160
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.000459307923771314,
      "loss": 3.3736,
      "theoretical_loss": 4.1481929223062615,
      "tokens_seen": 298909696
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004592978936810431,
      "loss": 3.1799,
      "theoretical_loss": 4.148086161824035,
      "tokens_seen": 298975232
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045928786359077236,
      "loss": 3.238,
      "theoretical_loss": 4.147979431292324,
      "tokens_seen": 299040768
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004592778335005015,
      "loss": 3.0854,
      "theoretical_loss": 4.147872730696165,
      "tokens_seen": 299106304
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004592678034102307,
      "loss": 3.1301,
      "theoretical_loss": 4.147766060020605,
      "tokens_seen": 299171840
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004592577733199599,
      "loss": 3.1008,
      "theoretical_loss": 4.147659419250703,
      "tokens_seen": 299237376
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004592477432296891,
      "loss": 3.3829,
      "theoretical_loss": 4.147552808371526,
      "tokens_seen": 299302912
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045923771313941827,
      "loss": 3.1832,
      "theoretical_loss": 4.147446227368155,
      "tokens_seen": 299368448
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045922768304914745,
      "loss": 3.238,
      "theoretical_loss": 4.147339676225681,
      "tokens_seen": 299433984
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045921765295887663,
      "loss": 3.4703,
      "theoretical_loss": 4.1472331549292045,
      "tokens_seen": 299499520
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045920762286860586,
      "loss": 3.2812,
      "theoretical_loss": 4.147126663463837,
      "tokens_seen": 299565056
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.000459197592778335,
      "loss": 3.1373,
      "theoretical_loss": 4.1470202018147,
      "tokens_seen": 299630592
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004591875626880642,
      "loss": 3.2132,
      "theoretical_loss": 4.146913769966929,
      "tokens_seen": 299696128
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045917753259779335,
      "loss": 3.3209,
      "theoretical_loss": 4.146807367905666,
      "tokens_seen": 299761664
    },
    {
      "epoch": 0.09,
      "objective/train/advantage_avg": 0.4944598078727722,
      "objective/train/docs_used": 181680,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.807593822479248,
      "objective/train/original_loss": 2.807594060897827,
      "objective/train/theoretical_loss": 4.146700995616065,
      "objective/train/tokens_used": 320287200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2445458024740219,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506889820098877,
      "objective/train/weighted_lm_loss": 2.950011968612671,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0396549701690674,
      "theoretical_loss": 4.146700995616065,
      "tokens_seen": 299827200
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004591675025075226,
      "loss": 3.1853,
      "theoretical_loss": 4.146700995616065,
      "tokens_seen": 299827200
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045915747241725177,
      "loss": 3.0443,
      "theoretical_loss": 4.146594653083293,
      "tokens_seen": 299892736
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045914744232698095,
      "loss": 2.896,
      "theoretical_loss": 4.146488340292523,
      "tokens_seen": 299958272
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045913741223671013,
      "loss": 3.0536,
      "theoretical_loss": 4.146382057228943,
      "tokens_seen": 300023808
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045912738214643937,
      "loss": 3.2963,
      "theoretical_loss": 4.146275803877751,
      "tokens_seen": 300089344
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004591173520561685,
      "loss": 3.2103,
      "theoretical_loss": 4.146169580224152,
      "tokens_seen": 300154880
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045910732196589773,
      "loss": 3.3239,
      "theoretical_loss": 4.146063386253366,
      "tokens_seen": 300220416
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045909729187562686,
      "loss": 2.9614,
      "theoretical_loss": 4.145957221950621,
      "tokens_seen": 300285952
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004590872617853561,
      "loss": 3.1674,
      "theoretical_loss": 4.145851087301156,
      "tokens_seen": 300351488
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004590772316950853,
      "loss": 3.081,
      "theoretical_loss": 4.145744982290221,
      "tokens_seen": 300417024
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045906720160481445,
      "loss": 3.1302,
      "theoretical_loss": 4.145638906903077,
      "tokens_seen": 300482560
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045905717151454364,
      "loss": 3.2926,
      "theoretical_loss": 4.145532861124994,
      "tokens_seen": 300548096
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004590471414242728,
      "loss": 3.0411,
      "theoretical_loss": 4.1454268449412535,
      "tokens_seen": 300613632
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.000459037111334002,
      "loss": 3.1717,
      "theoretical_loss": 4.145320858337149,
      "tokens_seen": 300679168
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045902708124373123,
      "loss": 2.9692,
      "theoretical_loss": 4.145214901297981,
      "tokens_seen": 300744704
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045901705115346036,
      "loss": 3.0895,
      "theoretical_loss": 4.145108973809064,
      "tokens_seen": 300810240
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004590070210631896,
      "loss": 3.0902,
      "theoretical_loss": 4.145003075855721,
      "tokens_seen": 300875776
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004589969909729187,
      "loss": 2.7719,
      "theoretical_loss": 4.144897207423284,
      "tokens_seen": 300941312
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045898696088264796,
      "loss": 3.1439,
      "theoretical_loss": 4.144791368497101,
      "tokens_seen": 301006848
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045897693079237714,
      "loss": 3.3897,
      "theoretical_loss": 4.144685559062525,
      "tokens_seen": 301072384
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004589669007021063,
      "loss": 2.9246,
      "theoretical_loss": 4.144579779104921,
      "tokens_seen": 301137920
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004589568706118355,
      "loss": 3.2362,
      "theoretical_loss": 4.1444740286096655,
      "tokens_seen": 301203456
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045894684052156474,
      "loss": 2.9183,
      "theoretical_loss": 4.144368307562145,
      "tokens_seen": 301268992
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045893681043129386,
      "loss": 3.3752,
      "theoretical_loss": 4.144262615947755,
      "tokens_seen": 301334528
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004589267803410231,
      "loss": 3.343,
      "theoretical_loss": 4.144156953751904,
      "tokens_seen": 301400064
    },
    {
      "epoch": 0.09,
      "objective/train/advantage_avg": 0.4936721622943878,
      "objective/train/docs_used": 182385,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.5783615112304688,
      "objective/train/original_loss": 3.578361749649048,
      "objective/train/theoretical_loss": 4.144051320960009,
      "objective/train/tokens_used": 321925600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24378986656665802,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506064891815186,
      "objective/train/weighted_lm_loss": 3.759338855743408,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0363596677780151,
      "theoretical_loss": 4.144051320960009,
      "tokens_seen": 301465600
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004589167502507522,
      "loss": 3.3679,
      "theoretical_loss": 4.144051320960009,
      "tokens_seen": 301465600
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045890672016048146,
      "loss": 3.0401,
      "theoretical_loss": 4.143945717557497,
      "tokens_seen": 301531136
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045889669007021064,
      "loss": 3.0808,
      "theoretical_loss": 4.143840143529808,
      "tokens_seen": 301596672
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004588866599799398,
      "loss": 3.2758,
      "theoretical_loss": 4.143734598862389,
      "tokens_seen": 301662208
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.000458876629889669,
      "loss": 3.2737,
      "theoretical_loss": 4.1436290835407,
      "tokens_seen": 301727744
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004588665997993982,
      "loss": 3.0187,
      "theoretical_loss": 4.14352359755021,
      "tokens_seen": 301793280
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045885656970912737,
      "loss": 3.1584,
      "theoretical_loss": 4.143418140876399,
      "tokens_seen": 301858816
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004588465396188566,
      "loss": 3.2063,
      "theoretical_loss": 4.1433127135047565,
      "tokens_seen": 301924352
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045883650952858573,
      "loss": 3.1823,
      "theoretical_loss": 4.143207315420783,
      "tokens_seen": 301989888
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045882647943831496,
      "loss": 3.0626,
      "theoretical_loss": 4.143101946609991,
      "tokens_seen": 302055424
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045881644934804415,
      "loss": 3.2819,
      "theoretical_loss": 4.142996607057899,
      "tokens_seen": 302120960
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004588064192577733,
      "loss": 3.1907,
      "theoretical_loss": 4.142891296750039,
      "tokens_seen": 302186496
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004587963891675025,
      "loss": 3.5116,
      "theoretical_loss": 4.142786015671954,
      "tokens_seen": 302252032
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004587863590772317,
      "loss": 3.2795,
      "theoretical_loss": 4.142680763809194,
      "tokens_seen": 302317568
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045877632898696087,
      "loss": 3.3761,
      "theoretical_loss": 4.142575541147323,
      "tokens_seen": 302383104
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004587662988966901,
      "loss": 3.2568,
      "theoretical_loss": 4.142470347671912,
      "tokens_seen": 302448640
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045875626880641923,
      "loss": 3.3883,
      "theoretical_loss": 4.142365183368544,
      "tokens_seen": 302514176
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045874623871614847,
      "loss": 3.2544,
      "theoretical_loss": 4.142260048222812,
      "tokens_seen": 302579712
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004587362086258776,
      "loss": 3.1924,
      "theoretical_loss": 4.142154942220319,
      "tokens_seen": 302645248
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045872617853560683,
      "loss": 3.228,
      "theoretical_loss": 4.14204986534668,
      "tokens_seen": 302710784
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.000458716148445336,
      "loss": 3.1529,
      "theoretical_loss": 4.141944817587516,
      "tokens_seen": 302776320
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004587061183550652,
      "loss": 3.1921,
      "theoretical_loss": 4.141839798928462,
      "tokens_seen": 302841856
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004586960882647944,
      "loss": 3.2566,
      "theoretical_loss": 4.141734809355163,
      "tokens_seen": 302907392
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045868605817452355,
      "loss": 3.3708,
      "theoretical_loss": 4.141629848853272,
      "tokens_seen": 302972928
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004586760280842528,
      "loss": 3.1547,
      "theoretical_loss": 4.141524917408454,
      "tokens_seen": 303038464
    },
    {
      "epoch": 0.09,
      "objective/train/advantage_avg": 0.4953043460845947,
      "objective/train/docs_used": 183646,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.079941511154175,
      "objective/train/original_loss": 3.079941749572754,
      "objective/train/theoretical_loss": 4.141420015006382,
      "objective/train/tokens_used": 323564000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24538274109363556,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507779121398926,
      "objective/train/weighted_lm_loss": 3.236419677734375,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0253151655197144,
      "theoretical_loss": 4.141420015006382,
      "tokens_seen": 303104000
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045866599799398197,
      "loss": 3.2873,
      "theoretical_loss": 4.141420015006382,
      "tokens_seen": 303104000
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045865596790371115,
      "loss": 3.2929,
      "theoretical_loss": 4.1413151416327425,
      "tokens_seen": 303169536
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045864593781344033,
      "loss": 3.1995,
      "theoretical_loss": 4.14121029727323,
      "tokens_seen": 303235072
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045863590772316957,
      "loss": 3.2236,
      "theoretical_loss": 4.141105481913549,
      "tokens_seen": 303300608
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004586258776328987,
      "loss": 3.3889,
      "theoretical_loss": 4.1410006955394145,
      "tokens_seen": 303366144
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045861584754262793,
      "loss": 3.0037,
      "theoretical_loss": 4.140895938136552,
      "tokens_seen": 303431680
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045860581745235706,
      "loss": 3.4215,
      "theoretical_loss": 4.140791209690697,
      "tokens_seen": 303497216
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004585957873620863,
      "loss": 3.1879,
      "theoretical_loss": 4.140686510187595,
      "tokens_seen": 303562752
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004585857572718155,
      "loss": 3.4289,
      "theoretical_loss": 4.1405818396130005,
      "tokens_seen": 303628288
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045857572718154465,
      "loss": 3.188,
      "theoretical_loss": 4.140477197952681,
      "tokens_seen": 303693824
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045856569709127384,
      "loss": 3.2247,
      "theoretical_loss": 4.14037258519241,
      "tokens_seen": 303759360
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.000458555667001003,
      "loss": 3.2047,
      "theoretical_loss": 4.140268001317976,
      "tokens_seen": 303824896
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004585456369107322,
      "loss": 3.3746,
      "theoretical_loss": 4.140163446315172,
      "tokens_seen": 303890432
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045853560682046143,
      "loss": 3.327,
      "theoretical_loss": 4.140058920169805,
      "tokens_seen": 303955968
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045852557673019056,
      "loss": 3.1427,
      "theoretical_loss": 4.139954422867691,
      "tokens_seen": 304021504
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004585155466399198,
      "loss": 3.3642,
      "theoretical_loss": 4.1398499543946565,
      "tokens_seen": 304087040
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004585055165496489,
      "loss": 3.3256,
      "theoretical_loss": 4.139745514736537,
      "tokens_seen": 304152576
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045849548645937816,
      "loss": 3.0138,
      "theoretical_loss": 4.139641103879177,
      "tokens_seen": 304218112
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045848545636910734,
      "loss": 3.3602,
      "theoretical_loss": 4.139536721808435,
      "tokens_seen": 304283648
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004584754262788365,
      "loss": 3.2038,
      "theoretical_loss": 4.1394323685101755,
      "tokens_seen": 304349184
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004584653961885657,
      "loss": 3.1007,
      "theoretical_loss": 4.139328043970274,
      "tokens_seen": 304414720
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045845536609829494,
      "loss": 3.3215,
      "theoretical_loss": 4.139223748174618,
      "tokens_seen": 304480256
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045844533600802406,
      "loss": 3.2047,
      "theoretical_loss": 4.139119481109101,
      "tokens_seen": 304545792
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004584353059177533,
      "loss": 3.0249,
      "theoretical_loss": 4.139015242759631,
      "tokens_seen": 304611328
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004584252758274824,
      "loss": 3.2101,
      "theoretical_loss": 4.138911033112123,
      "tokens_seen": 304676864
    },
    {
      "epoch": 0.09,
      "objective/train/advantage_avg": 0.4959307014942169,
      "objective/train/docs_used": 184075,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.6364943981170654,
      "objective/train/original_loss": 2.6364941596984863,
      "objective/train/theoretical_loss": 4.138806852152502,
      "objective/train/tokens_used": 325202400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24600711464881897,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0508437156677246,
      "objective/train/weighted_lm_loss": 2.7703769207000732,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 4.138806852152502,
      "tokens_seen": 304742400
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045841524573721166,
      "loss": 3.046,
      "theoretical_loss": 4.138806852152502,
      "tokens_seen": 304742400
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045840521564694084,
      "loss": 3.2201,
      "theoretical_loss": 4.1387026998667045,
      "tokens_seen": 304807936
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045839518555667,
      "loss": 3.236,
      "theoretical_loss": 4.138598576240675,
      "tokens_seen": 304873472
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004583851554663992,
      "loss": 3.0105,
      "theoretical_loss": 4.138494481260371,
      "tokens_seen": 304939008
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004583751253761284,
      "loss": 3.1695,
      "theoretical_loss": 4.138390414911756,
      "tokens_seen": 305004544
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045836509528585757,
      "loss": 3.3289,
      "theoretical_loss": 4.138286377180805,
      "tokens_seen": 305070080
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004583550651955868,
      "loss": 3.2879,
      "theoretical_loss": 4.138182368053505,
      "tokens_seen": 305135616
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045834503510531593,
      "loss": 2.9967,
      "theoretical_loss": 4.138078387515849,
      "tokens_seen": 305201152
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045833500501504516,
      "loss": 3.3028,
      "theoretical_loss": 4.1379744355538435,
      "tokens_seen": 305266688
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045832497492477435,
      "loss": 2.996,
      "theoretical_loss": 4.137870512153501,
      "tokens_seen": 305332224
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004583149448345035,
      "loss": 3.4446,
      "theoretical_loss": 4.137766617300849,
      "tokens_seen": 305397760
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004583049147442327,
      "loss": 3.1792,
      "theoretical_loss": 4.137662750981919,
      "tokens_seen": 305463296
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004582948846539619,
      "loss": 3.2759,
      "theoretical_loss": 4.137558913182758,
      "tokens_seen": 305528832
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045828485456369107,
      "loss": 3.2115,
      "theoretical_loss": 4.137455103889417,
      "tokens_seen": 305594368
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004582748244734203,
      "loss": 3.1487,
      "theoretical_loss": 4.137351323087962,
      "tokens_seen": 305659904
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045826479438314943,
      "loss": 3.192,
      "theoretical_loss": 4.137247570764467,
      "tokens_seen": 305725440
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045825476429287867,
      "loss": 3.2435,
      "theoretical_loss": 4.137143846905014,
      "tokens_seen": 305790976
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004582447342026078,
      "loss": 3.1739,
      "theoretical_loss": 4.137040151495696,
      "tokens_seen": 305856512
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045823470411233703,
      "loss": 3.1901,
      "theoretical_loss": 4.136936484522616,
      "tokens_seen": 305922048
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004582246740220662,
      "loss": 3.0141,
      "theoretical_loss": 4.136832845971888,
      "tokens_seen": 305987584
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004582146439317954,
      "loss": 3.1706,
      "theoretical_loss": 4.136729235829634,
      "tokens_seen": 306053120
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045820461384152457,
      "loss": 3.1651,
      "theoretical_loss": 4.136625654081986,
      "tokens_seen": 306118656
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045819458375125375,
      "loss": 3.4558,
      "theoretical_loss": 4.136522100715087,
      "tokens_seen": 306184192
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045818455366098294,
      "loss": 3.0496,
      "theoretical_loss": 4.136418575715086,
      "tokens_seen": 306249728
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045817452357071217,
      "loss": 3.3183,
      "theoretical_loss": 4.136315079068147,
      "tokens_seen": 306315264
    },
    {
      "epoch": 0.09,
      "objective/train/advantage_avg": 0.4936089813709259,
      "objective/train/docs_used": 185377,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.4112157821655273,
      "objective/train/original_loss": 3.4112162590026855,
      "objective/train/theoretical_loss": 4.136211610760441,
      "objective/train/tokens_used": 326840800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24371224641799927,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505996942520142,
      "objective/train/weighted_lm_loss": 3.583704710006714,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.044721245765686,
      "theoretical_loss": 4.136211610760441,
      "tokens_seen": 306380800
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004581644934804413,
      "loss": 3.4296,
      "theoretical_loss": 4.136211610760441,
      "tokens_seen": 306380800
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045815446339017053,
      "loss": 2.9308,
      "theoretical_loss": 4.136108170778147,
      "tokens_seen": 306446336
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004581444332998997,
      "loss": 3.1479,
      "theoretical_loss": 4.136004759107458,
      "tokens_seen": 306511872
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004581344032096289,
      "loss": 3.2352,
      "theoretical_loss": 4.135901375734571,
      "tokens_seen": 306577408
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004581243731193581,
      "loss": 3.1353,
      "theoretical_loss": 4.1357980206457,
      "tokens_seen": 306642944
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045811434302908726,
      "loss": 3.0906,
      "theoretical_loss": 4.135694693827061,
      "tokens_seen": 306708480
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045810431293881644,
      "loss": 3.3038,
      "theoretical_loss": 4.135591395264884,
      "tokens_seen": 306774016
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004580942828485457,
      "loss": 3.2125,
      "theoretical_loss": 4.135488124945409,
      "tokens_seen": 306839552
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004580842527582748,
      "loss": 3.2093,
      "theoretical_loss": 4.135384882854884,
      "tokens_seen": 306905088
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045807422266800404,
      "loss": 3.0561,
      "theoretical_loss": 4.135281668979566,
      "tokens_seen": 306970624
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045806419257773316,
      "loss": 3.1817,
      "theoretical_loss": 4.135178483305725,
      "tokens_seen": 307036160
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004580541624874624,
      "loss": 3.1498,
      "theoretical_loss": 4.135075325819637,
      "tokens_seen": 307101696
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004580441323971916,
      "loss": 3.3522,
      "theoretical_loss": 4.134972196507588,
      "tokens_seen": 307167232
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045803410230692076,
      "loss": 2.9599,
      "theoretical_loss": 4.134869095355876,
      "tokens_seen": 307232768
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045802407221664994,
      "loss": 3.1732,
      "theoretical_loss": 4.134766022350807,
      "tokens_seen": 307298304
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004580140421263791,
      "loss": 3.14,
      "theoretical_loss": 4.134662977478697,
      "tokens_seen": 307363840
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004580040120361083,
      "loss": 3.3053,
      "theoretical_loss": 4.13455996072587,
      "tokens_seen": 307429376
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045799398194583754,
      "loss": 3.002,
      "theoretical_loss": 4.134456972078661,
      "tokens_seen": 307494912
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045798395185556667,
      "loss": 3.2952,
      "theoretical_loss": 4.134354011523417,
      "tokens_seen": 307560448
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004579739217652959,
      "loss": 3.2191,
      "theoretical_loss": 4.134251079046488,
      "tokens_seen": 307625984
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004579638916750251,
      "loss": 3.3311,
      "theoretical_loss": 4.13414817463424,
      "tokens_seen": 307691520
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045795386158475426,
      "loss": 3.1899,
      "theoretical_loss": 4.134045298273046,
      "tokens_seen": 307757056
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004579438314944835,
      "loss": 3.3186,
      "theoretical_loss": 4.133942449949287,
      "tokens_seen": 307822592
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004579338014042126,
      "loss": 3.0354,
      "theoretical_loss": 4.133839629649357,
      "tokens_seen": 307888128
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045792377131394186,
      "loss": 3.1377,
      "theoretical_loss": 4.133736837359656,
      "tokens_seen": 307953664
    },
    {
      "epoch": 0.09,
      "objective/train/advantage_avg": 0.4946558177471161,
      "objective/train/docs_used": 186163,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.0586376190185547,
      "objective/train/original_loss": 3.0586376190185547,
      "objective/train/theoretical_loss": 4.133634073066595,
      "objective/train/tokens_used": 328479200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24475303292274475,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.05070960521698,
      "objective/train/weighted_lm_loss": 3.2135303020477295,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0454314947128296,
      "theoretical_loss": 4.133634073066595,
      "tokens_seen": 308019200
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045791374122367104,
      "loss": 3.0786,
      "theoretical_loss": 4.133634073066595,
      "tokens_seen": 308019200
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004579037111334002,
      "loss": 3.4014,
      "theoretical_loss": 4.133531336756596,
      "tokens_seen": 308084736
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004578936810431294,
      "loss": 3.4464,
      "theoretical_loss": 4.1334286284160875,
      "tokens_seen": 308150272
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004578836509528586,
      "loss": 3.1843,
      "theoretical_loss": 4.1333259480315085,
      "tokens_seen": 308215808
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045787362086258777,
      "loss": 3.0826,
      "theoretical_loss": 4.1332232955893105,
      "tokens_seen": 308281344
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.000457863590772317,
      "loss": 3.2394,
      "theoretical_loss": 4.133120671075949,
      "tokens_seen": 308346880
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045785356068204613,
      "loss": 3.5627,
      "theoretical_loss": 4.133018074477894,
      "tokens_seen": 308412416
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045784353059177536,
      "loss": 3.2675,
      "theoretical_loss": 4.132915505781621,
      "tokens_seen": 308477952
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045783350050150455,
      "loss": 3.0116,
      "theoretical_loss": 4.132812964973617,
      "tokens_seen": 308543488
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004578234704112337,
      "loss": 3.2707,
      "theoretical_loss": 4.132710452040378,
      "tokens_seen": 308609024
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004578134403209629,
      "loss": 3.0037,
      "theoretical_loss": 4.132607966968411,
      "tokens_seen": 308674560
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004578034102306921,
      "loss": 3.3508,
      "theoretical_loss": 4.132505509744229,
      "tokens_seen": 308740096
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045779338014042127,
      "loss": 3.0627,
      "theoretical_loss": 4.132403080354357,
      "tokens_seen": 308805632
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004577833500501505,
      "loss": 3.0307,
      "theoretical_loss": 4.132300678785328,
      "tokens_seen": 308871168
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045777331995987963,
      "loss": 3.2376,
      "theoretical_loss": 4.132198305023686,
      "tokens_seen": 308936704
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045776328986960887,
      "loss": 2.9473,
      "theoretical_loss": 4.132095959055983,
      "tokens_seen": 309002240
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.000457753259779338,
      "loss": 3.1971,
      "theoretical_loss": 4.13199364086878,
      "tokens_seen": 309067776
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045774322968906723,
      "loss": 3.5178,
      "theoretical_loss": 4.131891350448649,
      "tokens_seen": 309133312
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004577331995987964,
      "loss": 3.2639,
      "theoretical_loss": 4.1317890877821695,
      "tokens_seen": 309198848
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004577231695085256,
      "loss": 3.0813,
      "theoretical_loss": 4.131686852855932,
      "tokens_seen": 309264384
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004577131394182548,
      "loss": 3.113,
      "theoretical_loss": 4.131584645656535,
      "tokens_seen": 309329920
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045770310932798395,
      "loss": 3.4369,
      "theoretical_loss": 4.131482466170587,
      "tokens_seen": 309395456
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045769307923771314,
      "loss": 3.3522,
      "theoretical_loss": 4.131380314384706,
      "tokens_seen": 309460992
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045768304914744237,
      "loss": 3.2489,
      "theoretical_loss": 4.131278190285518,
      "tokens_seen": 309526528
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004576730190571715,
      "loss": 3.1892,
      "theoretical_loss": 4.1311760938596604,
      "tokens_seen": 309592064
    },
    {
      "epoch": 0.09,
      "objective/train/advantage_avg": 0.4930472671985626,
      "objective/train/docs_used": 187126,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.9618749618530273,
      "objective/train/original_loss": 2.9618752002716064,
      "objective/train/theoretical_loss": 4.131074025093778,
      "objective/train/tokens_used": 330117600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24322275817394257,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505410432815552,
      "objective/train/weighted_lm_loss": 3.1117260456085205,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 4.131074025093778,
      "tokens_seen": 309657600
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045766298896690073,
      "loss": 3.2609,
      "theoretical_loss": 4.131074025093778,
      "tokens_seen": 309657600
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004576529588766299,
      "loss": 3.265,
      "theoretical_loss": 4.130971983974526,
      "tokens_seen": 309723136
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004576429287863591,
      "loss": 3.4806,
      "theoretical_loss": 4.130869970488567,
      "tokens_seen": 309788672
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004576328986960883,
      "loss": 3.1186,
      "theoretical_loss": 4.130767984622577,
      "tokens_seen": 309854208
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045762286860581746,
      "loss": 3.2661,
      "theoretical_loss": 4.130666026363236,
      "tokens_seen": 309919744
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045761283851554664,
      "loss": 2.8587,
      "theoretical_loss": 4.1305640956972365,
      "tokens_seen": 309985280
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004576028084252759,
      "loss": 3.0386,
      "theoretical_loss": 4.13046219261128,
      "tokens_seen": 310050816
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.000457592778335005,
      "loss": 3.3416,
      "theoretical_loss": 4.130360317092077,
      "tokens_seen": 310116352
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045758274824473424,
      "loss": 3.2494,
      "theoretical_loss": 4.130258469126346,
      "tokens_seen": 310181888
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045757271815446336,
      "loss": 3.2814,
      "theoretical_loss": 4.130156648700816,
      "tokens_seen": 310247424
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004575626880641926,
      "loss": 3.3083,
      "theoretical_loss": 4.130054855802225,
      "tokens_seen": 310312960
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004575526579739218,
      "loss": 3.266,
      "theoretical_loss": 4.129953090417319,
      "tokens_seen": 310378496
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045754262788365096,
      "loss": 3.2362,
      "theoretical_loss": 4.129851352532856,
      "tokens_seen": 310444032
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045753259779338014,
      "loss": 3.1762,
      "theoretical_loss": 4.129749642135599,
      "tokens_seen": 310509568
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004575225677031093,
      "loss": 3.1072,
      "theoretical_loss": 4.129647959212325,
      "tokens_seen": 310575104
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004575125376128385,
      "loss": 3.2866,
      "theoretical_loss": 4.129546303749816,
      "tokens_seen": 310640640
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045750250752256774,
      "loss": 3.043,
      "theoretical_loss": 4.129444675734866,
      "tokens_seen": 310706176
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045749247743229687,
      "loss": 3.1933,
      "theoretical_loss": 4.129343075154275,
      "tokens_seen": 310771712
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004574824473420261,
      "loss": 3.0318,
      "theoretical_loss": 4.129241501994857,
      "tokens_seen": 310837248
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004574724172517553,
      "loss": 3.3049,
      "theoretical_loss": 4.129139956243429,
      "tokens_seen": 310902784
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045746238716148446,
      "loss": 3.3115,
      "theoretical_loss": 4.129038437886822,
      "tokens_seen": 310968320
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045745235707121364,
      "loss": 3.2526,
      "theoretical_loss": 4.128936946911875,
      "tokens_seen": 311033856
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004574423269809428,
      "loss": 3.5231,
      "theoretical_loss": 4.1288354833054335,
      "tokens_seen": 311099392
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.000457432296890672,
      "loss": 3.3652,
      "theoretical_loss": 4.1287340470543565,
      "tokens_seen": 311164928
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045742226680040124,
      "loss": 3.3688,
      "theoretical_loss": 4.128632638145508,
      "tokens_seen": 311230464
    },
    {
      "epoch": 0.09,
      "objective/train/advantage_avg": 0.49221256375312805,
      "objective/train/docs_used": 187875,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.9318759441375732,
      "objective/train/original_loss": 2.9318759441375732,
      "objective/train/theoretical_loss": 4.128531256565763,
      "objective/train/tokens_used": 331756000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24235175549983978,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0504530668258667,
      "objective/train/weighted_lm_loss": 3.0795204639434814,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0457526445388794,
      "theoretical_loss": 4.128531256565763,
      "tokens_seen": 311296000
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045741223671013037,
      "loss": 3.1001,
      "theoretical_loss": 4.128531256565763,
      "tokens_seen": 311296000
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004574022066198596,
      "loss": 3.0941,
      "theoretical_loss": 4.128429902302006,
      "tokens_seen": 311361536
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045739217652958873,
      "loss": 3.4387,
      "theoretical_loss": 4.128328575341129,
      "tokens_seen": 311427072
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045738214643931797,
      "loss": 3.3511,
      "theoretical_loss": 4.128227275670034,
      "tokens_seen": 311492608
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045737211634904715,
      "loss": 3.2345,
      "theoretical_loss": 4.128126003275633,
      "tokens_seen": 311558144
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045736208625877633,
      "loss": 3.3359,
      "theoretical_loss": 4.128024758144845,
      "tokens_seen": 311623680
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004573520561685055,
      "loss": 3.3666,
      "theoretical_loss": 4.127923540264598,
      "tokens_seen": 311689216
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045734202607823475,
      "loss": 3.3691,
      "theoretical_loss": 4.127822349621832,
      "tokens_seen": 311754752
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045733199598796387,
      "loss": 3.3418,
      "theoretical_loss": 4.127721186203494,
      "tokens_seen": 311820288
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004573219658976931,
      "loss": 3.2518,
      "theoretical_loss": 4.127620049996538,
      "tokens_seen": 311885824
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045731193580742223,
      "loss": 3.3968,
      "theoretical_loss": 4.12751894098793,
      "tokens_seen": 311951360
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045730190571715147,
      "loss": 3.5266,
      "theoretical_loss": 4.127417859164645,
      "tokens_seen": 312016896
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045729187562688065,
      "loss": 3.1812,
      "theoretical_loss": 4.127316804513664,
      "tokens_seen": 312082432
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045728184553660983,
      "loss": 3.2292,
      "theoretical_loss": 4.127215777021981,
      "tokens_seen": 312147968
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.000457271815446339,
      "loss": 2.9507,
      "theoretical_loss": 4.127114776676595,
      "tokens_seen": 312213504
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004572617853560682,
      "loss": 3.4361,
      "theoretical_loss": 4.127013803464516,
      "tokens_seen": 312279040
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004572517552657974,
      "loss": 2.9842,
      "theoretical_loss": 4.126912857372765,
      "tokens_seen": 312344576
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004572417251755266,
      "loss": 3.3224,
      "theoretical_loss": 4.126811938388367,
      "tokens_seen": 312410112
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045723169508525574,
      "loss": 3.2996,
      "theoretical_loss": 4.12671104649836,
      "tokens_seen": 312475648
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.000457221664994985,
      "loss": 3.3792,
      "theoretical_loss": 4.12661018168979,
      "tokens_seen": 312541184
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004572116349047141,
      "loss": 3.0353,
      "theoretical_loss": 4.126509343949709,
      "tokens_seen": 312606720
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045720160481444334,
      "loss": 3.4325,
      "theoretical_loss": 4.126408533265183,
      "tokens_seen": 312672256
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045719157472417257,
      "loss": 3.0987,
      "theoretical_loss": 4.126307749623283,
      "tokens_seen": 312737792
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004571815446339017,
      "loss": 3.1949,
      "theoretical_loss": 4.12620699301109,
      "tokens_seen": 312803328
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045717151454363093,
      "loss": 3.4552,
      "theoretical_loss": 4.1261062634156955,
      "tokens_seen": 312868864
    },
    {
      "epoch": 0.09,
      "objective/train/advantage_avg": 0.49421873688697815,
      "objective/train/docs_used": 188927,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.302163600921631,
      "objective/train/original_loss": 3.302163600921631,
      "objective/train/theoretical_loss": 4.126005560824196,
      "objective/train/tokens_used": 333394400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24432355165481567,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506638288497925,
      "objective/train/weighted_lm_loss": 3.4693210124969482,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0396549701690674,
      "theoretical_loss": 4.126005560824196,
      "tokens_seen": 312934400
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004571614844533601,
      "loss": 3.5266,
      "theoretical_loss": 4.126005560824196,
      "tokens_seen": 312934400
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004571514543630893,
      "loss": 3.3728,
      "theoretical_loss": 4.125904885223701,
      "tokens_seen": 312999936
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004571414242728185,
      "loss": 3.5407,
      "theoretical_loss": 4.125804236601327,
      "tokens_seen": 313065472
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045713139418254766,
      "loss": 3.3108,
      "theoretical_loss": 4.1257036149441975,
      "tokens_seen": 313131008
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045712136409227684,
      "loss": 3.085,
      "theoretical_loss": 4.125603020239449,
      "tokens_seen": 313196544
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004571113340020061,
      "loss": 3.334,
      "theoretical_loss": 4.125502452474223,
      "tokens_seen": 313262080
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.0004571013039117352,
      "loss": 3.2478,
      "theoretical_loss": 4.125401911635674,
      "tokens_seen": 313327616
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045709127382146444,
      "loss": 3.4575,
      "theoretical_loss": 4.125301397710959,
      "tokens_seen": 313393152
    },
    {
      "epoch": 0.09,
      "learning_rate": 0.00045708124373119356,
      "loss": 3.4235,
      "theoretical_loss": 4.12520091068725,
      "tokens_seen": 313458688
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004570712136409228,
      "loss": 3.2982,
      "theoretical_loss": 4.125100450551725,
      "tokens_seen": 313524224
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.000457061183550652,
      "loss": 3.0426,
      "theoretical_loss": 4.12500001729157,
      "tokens_seen": 313589760
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045705115346038116,
      "loss": 3.3525,
      "theoretical_loss": 4.124899610893983,
      "tokens_seen": 313655296
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045704112337011034,
      "loss": 3.2531,
      "theoretical_loss": 4.1247992313461666,
      "tokens_seen": 313720832
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004570310932798395,
      "loss": 3.4441,
      "theoretical_loss": 4.124698878635335,
      "tokens_seen": 313786368
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004570210631895687,
      "loss": 3.355,
      "theoretical_loss": 4.12459855274871,
      "tokens_seen": 313851904
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045701103309929794,
      "loss": 3.1559,
      "theoretical_loss": 4.124498253673523,
      "tokens_seen": 313917440
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045700100300902707,
      "loss": 3.5971,
      "theoretical_loss": 4.124397981397014,
      "tokens_seen": 313982976
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004569909729187563,
      "loss": 3.3509,
      "theoretical_loss": 4.124297735906431,
      "tokens_seen": 314048512
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004569809428284855,
      "loss": 3.2984,
      "theoretical_loss": 4.124197517189032,
      "tokens_seen": 314114048
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045697091273821466,
      "loss": 3.4643,
      "theoretical_loss": 4.124097325232081,
      "tokens_seen": 314179584
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045696088264794384,
      "loss": 3.4059,
      "theoretical_loss": 4.123997160022854,
      "tokens_seen": 314245120
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.000456950852557673,
      "loss": 3.4262,
      "theoretical_loss": 4.1238970215486335,
      "tokens_seen": 314310656
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004569408224674022,
      "loss": 3.2704,
      "theoretical_loss": 4.123796909796713,
      "tokens_seen": 314376192
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045693079237713144,
      "loss": 3.1966,
      "theoretical_loss": 4.123696824754391,
      "tokens_seen": 314441728
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045692076228686057,
      "loss": 3.2509,
      "theoretical_loss": 4.123596766408978,
      "tokens_seen": 314507264
    },
    {
      "epoch": 0.1,
      "objective/train/advantage_avg": 0.4950125515460968,
      "objective/train/docs_used": 189547,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.0951547622680664,
      "objective/train/original_loss": 3.0951550006866455,
      "objective/train/theoretical_loss": 4.123496734747793,
      "objective/train/tokens_used": 335032800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24510212242603302,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507473945617676,
      "objective/train/weighted_lm_loss": 3.2522311210632324,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0408108234405518,
      "theoretical_loss": 4.123496734747793,
      "tokens_seen": 314572800
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004569107321965898,
      "loss": 3.3962,
      "theoretical_loss": 4.123496734747793,
      "tokens_seen": 314572800
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045690070210631893,
      "loss": 3.3836,
      "theoretical_loss": 4.12339672975816,
      "tokens_seen": 314638336
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045689067201604817,
      "loss": 3.1269,
      "theoretical_loss": 4.123296751427417,
      "tokens_seen": 314703872
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045688064192577735,
      "loss": 3.303,
      "theoretical_loss": 4.123196799742907,
      "tokens_seen": 314769408
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045687061183550653,
      "loss": 3.4024,
      "theoretical_loss": 4.123096874691981,
      "tokens_seen": 314834944
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004568605817452357,
      "loss": 3.07,
      "theoretical_loss": 4.122996976262003,
      "tokens_seen": 314900480
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045685055165496495,
      "loss": 3.3981,
      "theoretical_loss": 4.122897104440343,
      "tokens_seen": 314966016
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045684052156469407,
      "loss": 3.1816,
      "theoretical_loss": 4.1227972592143765,
      "tokens_seen": 315031552
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004568304914744233,
      "loss": 3.3893,
      "theoretical_loss": 4.122697440571493,
      "tokens_seen": 315097088
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045682046138415243,
      "loss": 3.3504,
      "theoretical_loss": 4.122597648499087,
      "tokens_seen": 315162624
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045681043129388167,
      "loss": 3.4764,
      "theoretical_loss": 4.122497882984564,
      "tokens_seen": 315228160
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045680040120361085,
      "loss": 3.1921,
      "theoretical_loss": 4.122398144015335,
      "tokens_seen": 315293696
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045679037111334003,
      "loss": 3.0886,
      "theoretical_loss": 4.122298431578824,
      "tokens_seen": 315359232
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004567803410230692,
      "loss": 3.266,
      "theoretical_loss": 4.12219874566246,
      "tokens_seen": 315424768
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004567703109327984,
      "loss": 3.1911,
      "theoretical_loss": 4.122099086253681,
      "tokens_seen": 315490304
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004567602808425276,
      "loss": 3.3766,
      "theoretical_loss": 4.121999453339935,
      "tokens_seen": 315555840
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004567502507522568,
      "loss": 3.3135,
      "theoretical_loss": 4.121899846908677,
      "tokens_seen": 315621376
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045674022066198594,
      "loss": 3.497,
      "theoretical_loss": 4.121800266947372,
      "tokens_seen": 315686912
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004567301905717152,
      "loss": 3.3702,
      "theoretical_loss": 4.121700713443492,
      "tokens_seen": 315752448
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004567201604814443,
      "loss": 3.3264,
      "theoretical_loss": 4.121601186384519,
      "tokens_seen": 315817984
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045671013039117354,
      "loss": 3.6497,
      "theoretical_loss": 4.121501685757943,
      "tokens_seen": 315883520
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004567001003009027,
      "loss": 3.0948,
      "theoretical_loss": 4.121402211551262,
      "tokens_seen": 315949056
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004566900702106319,
      "loss": 3.2845,
      "theoretical_loss": 4.121302763751983,
      "tokens_seen": 316014592
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004566800401203611,
      "loss": 3.354,
      "theoretical_loss": 4.12120334234762,
      "tokens_seen": 316080128
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004566700100300903,
      "loss": 3.1835,
      "theoretical_loss": 4.121103947325699,
      "tokens_seen": 316145664
    },
    {
      "epoch": 0.1,
      "objective/train/advantage_avg": 0.4945513904094696,
      "objective/train/docs_used": 190733,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.089888095855713,
      "objective/train/original_loss": 3.089887857437134,
      "objective/train/theoretical_loss": 4.121004578673752,
      "objective/train/tokens_used": 336671200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24464742839336395,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506986379623413,
      "objective/train/weighted_lm_loss": 3.2467257976531982,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0338950157165527,
      "theoretical_loss": 4.121004578673752,
      "tokens_seen": 316211200
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045665997993981944,
      "loss": 3.2415,
      "theoretical_loss": 4.121004578673752,
      "tokens_seen": 316211200
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004566499498495487,
      "loss": 3.1542,
      "theoretical_loss": 4.120905236379318,
      "tokens_seen": 316276736
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004566399197592778,
      "loss": 3.3418,
      "theoretical_loss": 4.120805920429949,
      "tokens_seen": 316342272
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045662988966900704,
      "loss": 3.3717,
      "theoretical_loss": 4.120706630813201,
      "tokens_seen": 316407808
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004566198595787362,
      "loss": 3.3556,
      "theoretical_loss": 4.12060736751664,
      "tokens_seen": 316473344
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004566098294884654,
      "loss": 3.2698,
      "theoretical_loss": 4.120508130527841,
      "tokens_seen": 316538880
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004565997993981946,
      "loss": 3.3375,
      "theoretical_loss": 4.120408919834388,
      "tokens_seen": 316604416
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045658976930792376,
      "loss": 3.4123,
      "theoretical_loss": 4.120309735423871,
      "tokens_seen": 316669952
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045657973921765294,
      "loss": 3.3231,
      "theoretical_loss": 4.120210577283891,
      "tokens_seen": 316735488
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004565697091273822,
      "loss": 3.1469,
      "theoretical_loss": 4.120111445402056,
      "tokens_seen": 316801024
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004565596790371113,
      "loss": 3.1852,
      "theoretical_loss": 4.1200123397659825,
      "tokens_seen": 316866560
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045654964894684054,
      "loss": 3.3178,
      "theoretical_loss": 4.119913260363296,
      "tokens_seen": 316932096
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045653961885656967,
      "loss": 3.3539,
      "theoretical_loss": 4.11981420718163,
      "tokens_seen": 316997632
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004565295887662989,
      "loss": 3.2191,
      "theoretical_loss": 4.119715180208626,
      "tokens_seen": 317063168
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004565195586760281,
      "loss": 3.1961,
      "theoretical_loss": 4.119616179431935,
      "tokens_seen": 317128704
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045650952858575727,
      "loss": 3.3728,
      "theoretical_loss": 4.119517204839215,
      "tokens_seen": 317194240
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045649949849548645,
      "loss": 3.027,
      "theoretical_loss": 4.119418256418133,
      "tokens_seen": 317259776
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004564894684052157,
      "loss": 3.4308,
      "theoretical_loss": 4.119319334156366,
      "tokens_seen": 317325312
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004564794383149448,
      "loss": 3.159,
      "theoretical_loss": 4.119220438041595,
      "tokens_seen": 317390848
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045646940822467405,
      "loss": 3.2895,
      "theoretical_loss": 4.119121568061515,
      "tokens_seen": 317456384
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045645937813440317,
      "loss": 3.0865,
      "theoretical_loss": 4.119022724203824,
      "tokens_seen": 317521920
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004564493480441324,
      "loss": 3.3676,
      "theoretical_loss": 4.1189239064562315,
      "tokens_seen": 317587456
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045643931795386164,
      "loss": 3.5299,
      "theoretical_loss": 4.118825114806455,
      "tokens_seen": 317652992
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045642928786359077,
      "loss": 3.314,
      "theoretical_loss": 4.118726349242221,
      "tokens_seen": 317718528
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045641925777332,
      "loss": 3.1838,
      "theoretical_loss": 4.118627609751259,
      "tokens_seen": 317784064
    },
    {
      "epoch": 0.1,
      "objective/train/advantage_avg": 0.49288105964660645,
      "objective/train/docs_used": 191330,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.2269513607025146,
      "objective/train/original_loss": 3.2269515991210938,
      "objective/train/theoretical_loss": 4.118528896321316,
      "objective/train/tokens_used": 338309600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2430429458618164,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505235195159912,
      "objective/train/weighted_lm_loss": 3.39017391204834,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0408108234405518,
      "theoretical_loss": 4.118528896321316,
      "tokens_seen": 317849600
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045640922768304913,
      "loss": 3.2356,
      "theoretical_loss": 4.118528896321316,
      "tokens_seen": 317849600
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045639919759277837,
      "loss": 3.0653,
      "theoretical_loss": 4.118430208940139,
      "tokens_seen": 317915136
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045638916750250755,
      "loss": 3.3407,
      "theoretical_loss": 4.1183315475954885,
      "tokens_seen": 317980672
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045637913741223673,
      "loss": 3.1603,
      "theoretical_loss": 4.118232912275129,
      "tokens_seen": 318046208
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004563691073219659,
      "loss": 3.3116,
      "theoretical_loss": 4.118134302966838,
      "tokens_seen": 318111744
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045635907723169515,
      "loss": 3.2453,
      "theoretical_loss": 4.118035719658397,
      "tokens_seen": 318177280
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045634904714142427,
      "loss": 3.2941,
      "theoretical_loss": 4.117937162337598,
      "tokens_seen": 318242816
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004563390170511535,
      "loss": 3.4641,
      "theoretical_loss": 4.117838630992242,
      "tokens_seen": 318308352
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045632898696088263,
      "loss": 3.355,
      "theoretical_loss": 4.117740125610136,
      "tokens_seen": 318373888
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045631895687061187,
      "loss": 3.3953,
      "theoretical_loss": 4.117641646179096,
      "tokens_seen": 318439424
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045630892678034105,
      "loss": 3.1287,
      "theoretical_loss": 4.117543192686947,
      "tokens_seen": 318504960
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045629889669007023,
      "loss": 3.3661,
      "theoretical_loss": 4.1174447651215225,
      "tokens_seen": 318570496
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004562888665997994,
      "loss": 3.3495,
      "theoretical_loss": 4.117346363470663,
      "tokens_seen": 318636032
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004562788365095286,
      "loss": 3.1497,
      "theoretical_loss": 4.117247987722216,
      "tokens_seen": 318701568
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004562688064192578,
      "loss": 3.201,
      "theoretical_loss": 4.117149637864041,
      "tokens_seen": 318767104
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.000456258776328987,
      "loss": 3.3687,
      "theoretical_loss": 4.117051313884003,
      "tokens_seen": 318832640
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045624874623871614,
      "loss": 3.6204,
      "theoretical_loss": 4.116953015769975,
      "tokens_seen": 318898176
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004562387161484454,
      "loss": 3.0276,
      "theoretical_loss": 4.11685474350984,
      "tokens_seen": 318963712
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004562286860581745,
      "loss": 3.3238,
      "theoretical_loss": 4.116756497091487,
      "tokens_seen": 319029248
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045621865596790374,
      "loss": 3.2917,
      "theoretical_loss": 4.116658276502815,
      "tokens_seen": 319094784
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004562086258776329,
      "loss": 3.4155,
      "theoretical_loss": 4.116560081731729,
      "tokens_seen": 319160320
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004561985957873621,
      "loss": 3.3653,
      "theoretical_loss": 4.1164619127661455,
      "tokens_seen": 319225856
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004561885656970913,
      "loss": 3.2161,
      "theoretical_loss": 4.116363769593986,
      "tokens_seen": 319291392
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004561785356068205,
      "loss": 3.4103,
      "theoretical_loss": 4.116265652203181,
      "tokens_seen": 319356928
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045616850551654964,
      "loss": 3.2816,
      "theoretical_loss": 4.116167560581669,
      "tokens_seen": 319422464
    },
    {
      "epoch": 0.1,
      "objective/train/advantage_avg": 0.4940725564956665,
      "objective/train/docs_used": 192764,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.9938433170318604,
      "objective/train/original_loss": 2.9938435554504395,
      "objective/train/theoretical_loss": 4.116069494717399,
      "objective/train/tokens_used": 339948000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24415349960327148,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506483316421509,
      "objective/train/weighted_lm_loss": 3.1455249786376953,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0396549701690674,
      "theoretical_loss": 4.116069494717399,
      "tokens_seen": 319488000
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004561584754262789,
      "loss": 3.1974,
      "theoretical_loss": 4.116069494717399,
      "tokens_seen": 319488000
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.000456148445336008,
      "loss": 3.4421,
      "theoretical_loss": 4.115971454598324,
      "tokens_seen": 319553536
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045613841524573724,
      "loss": 3.5459,
      "theoretical_loss": 4.115873440212409,
      "tokens_seen": 319619072
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004561283851554664,
      "loss": 3.2374,
      "theoretical_loss": 4.115775451547624,
      "tokens_seen": 319684608
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004561183550651956,
      "loss": 3.3727,
      "theoretical_loss": 4.11567748859195,
      "tokens_seen": 319750144
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004561083249749248,
      "loss": 3.093,
      "theoretical_loss": 4.115579551333372,
      "tokens_seen": 319815680
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045609829488465396,
      "loss": 3.3255,
      "theoretical_loss": 4.115481639759887,
      "tokens_seen": 319881216
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045608826479438314,
      "loss": 3.4634,
      "theoretical_loss": 4.115383753859499,
      "tokens_seen": 319946752
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004560782347041124,
      "loss": 3.06,
      "theoretical_loss": 4.115285893620219,
      "tokens_seen": 320012288
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004560682046138415,
      "loss": 3.3641,
      "theoretical_loss": 4.115188059030066,
      "tokens_seen": 320077824
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045605817452357074,
      "loss": 3.4872,
      "theoretical_loss": 4.11509025007707,
      "tokens_seen": 320143360
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045604814443329987,
      "loss": 3.1255,
      "theoretical_loss": 4.114992466749264,
      "tokens_seen": 320208896
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004560381143430291,
      "loss": 3.1656,
      "theoretical_loss": 4.114894709034695,
      "tokens_seen": 320274432
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004560280842527583,
      "loss": 3.4827,
      "theoretical_loss": 4.114796976921412,
      "tokens_seen": 320339968
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045601805416248747,
      "loss": 3.2219,
      "theoretical_loss": 4.114699270397477,
      "tokens_seen": 320405504
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045600802407221665,
      "loss": 3.4352,
      "theoretical_loss": 4.114601589450956,
      "tokens_seen": 320471040
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004559979939819459,
      "loss": 3.3502,
      "theoretical_loss": 4.114503934069926,
      "tokens_seen": 320536576
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.000455987963891675,
      "loss": 3.1037,
      "theoretical_loss": 4.114406304242472,
      "tokens_seen": 320602112
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045597793380140425,
      "loss": 3.3243,
      "theoretical_loss": 4.114308699956684,
      "tokens_seen": 320667648
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045596790371113337,
      "loss": 3.2887,
      "theoretical_loss": 4.114211121200662,
      "tokens_seen": 320733184
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004559578736208626,
      "loss": 3.4391,
      "theoretical_loss": 4.114113567962514,
      "tokens_seen": 320798720
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004559478435305918,
      "loss": 2.9804,
      "theoretical_loss": 4.114016040230357,
      "tokens_seen": 320864256
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045593781344032097,
      "loss": 3.0817,
      "theoretical_loss": 4.113918537992314,
      "tokens_seen": 320929792
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045592778335005015,
      "loss": 3.4465,
      "theoretical_loss": 4.113821061236516,
      "tokens_seen": 320995328
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045591775325977933,
      "loss": 3.2143,
      "theoretical_loss": 4.113723609951103,
      "tokens_seen": 321060864
    },
    {
      "epoch": 0.1,
      "objective/train/advantage_avg": 0.493630975484848,
      "objective/train/docs_used": 193449,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.531113386154175,
      "objective/train/original_loss": 2.531113862991333,
      "objective/train/theoretical_loss": 4.113626184124224,
      "objective/train/tokens_used": 341586400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2437637895345688,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506023168563843,
      "objective/train/weighted_lm_loss": 2.65856671333313,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0428944826126099,
      "theoretical_loss": 4.113626184124224,
      "tokens_seen": 321126400
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004559077231695085,
      "loss": 3.1719,
      "theoretical_loss": 4.113626184124224,
      "tokens_seen": 321126400
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045589769307923775,
      "loss": 3.1002,
      "theoretical_loss": 4.113528783744033,
      "tokens_seen": 321191936
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004558876629889669,
      "loss": 3.2276,
      "theoretical_loss": 4.113431408798694,
      "tokens_seen": 321257472
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004558776328986961,
      "loss": 3.2762,
      "theoretical_loss": 4.113334059276378,
      "tokens_seen": 321323008
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045586760280842524,
      "loss": 3.2593,
      "theoretical_loss": 4.113236735165264,
      "tokens_seen": 321388544
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004558575727181545,
      "loss": 3.2012,
      "theoretical_loss": 4.11313943645354,
      "tokens_seen": 321454080
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045584754262788365,
      "loss": 3.3874,
      "theoretical_loss": 4.113042163129402,
      "tokens_seen": 321519616
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045583751253761284,
      "loss": 3.0882,
      "theoretical_loss": 4.112944915181051,
      "tokens_seen": 321585152
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.000455827482447342,
      "loss": 3.0639,
      "theoretical_loss": 4.112847692596698,
      "tokens_seen": 321650688
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045581745235707125,
      "loss": 3.4719,
      "theoretical_loss": 4.112750495364563,
      "tokens_seen": 321716224
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004558074222668004,
      "loss": 3.204,
      "theoretical_loss": 4.112653323472872,
      "tokens_seen": 321781760
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004557973921765296,
      "loss": 3.2263,
      "theoretical_loss": 4.11255617690986,
      "tokens_seen": 321847296
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045578736208625874,
      "loss": 3.1233,
      "theoretical_loss": 4.112459055663768,
      "tokens_seen": 321912832
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.000455777331995988,
      "loss": 3.2675,
      "theoretical_loss": 4.112361959722849,
      "tokens_seen": 321978368
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045576730190571716,
      "loss": 3.1484,
      "theoretical_loss": 4.112264889075357,
      "tokens_seen": 322043904
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045575727181544634,
      "loss": 3.2129,
      "theoretical_loss": 4.11216784370956,
      "tokens_seen": 322109440
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004557472417251755,
      "loss": 3.2221,
      "theoretical_loss": 4.112070823613734,
      "tokens_seen": 322174976
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004557372116349047,
      "loss": 3.0387,
      "theoretical_loss": 4.111973828776156,
      "tokens_seen": 322240512
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004557271815446339,
      "loss": 3.3356,
      "theoretical_loss": 4.111876859185118,
      "tokens_seen": 322306048
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004557171514543631,
      "loss": 3.3476,
      "theoretical_loss": 4.111779914828917,
      "tokens_seen": 322371584
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045570712136409224,
      "loss": 3.5165,
      "theoretical_loss": 4.111682995695858,
      "tokens_seen": 322437120
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004556970912738215,
      "loss": 3.1581,
      "theoretical_loss": 4.111586101774253,
      "tokens_seen": 322502656
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004556870611835507,
      "loss": 3.2191,
      "theoretical_loss": 4.111489233052423,
      "tokens_seen": 322568192
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045567703109327984,
      "loss": 3.3375,
      "theoretical_loss": 4.111392389518697,
      "tokens_seen": 322633728
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004556670010030091,
      "loss": 3.3192,
      "theoretical_loss": 4.111295571161409,
      "tokens_seen": 322699264
    },
    {
      "epoch": 0.1,
      "objective/train/advantage_avg": 0.4942582845687866,
      "objective/train/docs_used": 194727,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.3616836071014404,
      "objective/train/original_loss": 3.3616836071014404,
      "objective/train/theoretical_loss": 4.111198777968905,
      "objective/train/tokens_used": 343224800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24434004724025726,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506678819656372,
      "objective/train/weighted_lm_loss": 3.531949996948242,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0417574644088745,
      "theoretical_loss": 4.111198777968905,
      "tokens_seen": 322764800
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004556569709127382,
      "loss": 3.0729,
      "theoretical_loss": 4.111198777968905,
      "tokens_seen": 322764800
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045564694082246744,
      "loss": 3.2476,
      "theoretical_loss": 4.111102009929535,
      "tokens_seen": 322830336
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004556369107321966,
      "loss": 3.0567,
      "theoretical_loss": 4.11100526703166,
      "tokens_seen": 322895872
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004556268806419258,
      "loss": 3.1686,
      "theoretical_loss": 4.110908549263647,
      "tokens_seen": 322961408
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.000455616850551655,
      "loss": 3.0047,
      "theoretical_loss": 4.11081185661387,
      "tokens_seen": 323026944
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045560682046138416,
      "loss": 3.1798,
      "theoretical_loss": 4.110715189070711,
      "tokens_seen": 323092480
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045559679037111334,
      "loss": 3.3536,
      "theoretical_loss": 4.110618546622561,
      "tokens_seen": 323158016
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004555867602808426,
      "loss": 3.0663,
      "theoretical_loss": 4.110521929257819,
      "tokens_seen": 323223552
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004555767301905717,
      "loss": 3.1462,
      "theoretical_loss": 4.110425336964889,
      "tokens_seen": 323289088
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045556670010030094,
      "loss": 3.2886,
      "theoretical_loss": 4.110328769732186,
      "tokens_seen": 323354624
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045555667001003007,
      "loss": 3.0746,
      "theoretical_loss": 4.110232227548131,
      "tokens_seen": 323420160
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004555466399197593,
      "loss": 3.2308,
      "theoretical_loss": 4.110135710401153,
      "tokens_seen": 323485696
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004555366098294885,
      "loss": 3.3975,
      "theoretical_loss": 4.110039218279687,
      "tokens_seen": 323551232
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045552657973921767,
      "loss": 3.436,
      "theoretical_loss": 4.109942751172179,
      "tokens_seen": 323616768
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045551654964894685,
      "loss": 3.3749,
      "theoretical_loss": 4.10984630906708,
      "tokens_seen": 323682304
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004555065195586761,
      "loss": 3.2608,
      "theoretical_loss": 4.109749891952849,
      "tokens_seen": 323747840
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004554964894684052,
      "loss": 3.3375,
      "theoretical_loss": 4.109653499817955,
      "tokens_seen": 323813376
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045548645937813445,
      "loss": 3.4934,
      "theoretical_loss": 4.109557132650871,
      "tokens_seen": 323878912
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045547642928786357,
      "loss": 2.8896,
      "theoretical_loss": 4.109460790440082,
      "tokens_seen": 323944448
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004554663991975928,
      "loss": 3.3047,
      "theoretical_loss": 4.109364473174075,
      "tokens_seen": 324009984
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.000455456369107322,
      "loss": 3.2197,
      "theoretical_loss": 4.109268180841351,
      "tokens_seen": 324075520
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045544633901705117,
      "loss": 2.7977,
      "theoretical_loss": 4.109171913430414,
      "tokens_seen": 324141056
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045543630892678035,
      "loss": 3.1398,
      "theoretical_loss": 4.109075670929776,
      "tokens_seen": 324206592
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045542627883650953,
      "loss": 3.2519,
      "theoretical_loss": 4.108979453327959,
      "tokens_seen": 324272128
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004554162487462387,
      "loss": 3.2578,
      "theoretical_loss": 4.108883260613491,
      "tokens_seen": 324337664
    },
    {
      "epoch": 0.1,
      "objective/train/advantage_avg": 0.4942469298839569,
      "objective/train/docs_used": 195463,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.305856943130493,
      "objective/train/original_loss": 3.3058559894561768,
      "objective/train/theoretical_loss": 4.108787092774909,
      "objective/train/tokens_used": 344863200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24433614313602448,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506666898727417,
      "objective/train/weighted_lm_loss": 3.473337173461914,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0462769269943237,
      "theoretical_loss": 4.108787092774909,
      "tokens_seen": 324403200
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045540621865596795,
      "loss": 3.2131,
      "theoretical_loss": 4.108787092774909,
      "tokens_seen": 324403200
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004553961885656971,
      "loss": 3.2292,
      "theoretical_loss": 4.108690949800755,
      "tokens_seen": 324468736
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004553861584754263,
      "loss": 3.1056,
      "theoretical_loss": 4.1085948316795795,
      "tokens_seen": 324534272
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045537612838515544,
      "loss": 3.1599,
      "theoretical_loss": 4.108498738399943,
      "tokens_seen": 324599808
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004553660982948847,
      "loss": 3.1467,
      "theoretical_loss": 4.108402669950411,
      "tokens_seen": 324665344
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045535606820461385,
      "loss": 3.2135,
      "theoretical_loss": 4.108306626319557,
      "tokens_seen": 324730880
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045534603811434304,
      "loss": 2.9753,
      "theoretical_loss": 4.108210607495963,
      "tokens_seen": 324796416
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004553360080240722,
      "loss": 3.2048,
      "theoretical_loss": 4.108114613468217,
      "tokens_seen": 324861952
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045532597793380145,
      "loss": 3.2152,
      "theoretical_loss": 4.1080186442249165,
      "tokens_seen": 324927488
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004553159478435306,
      "loss": 3.2527,
      "theoretical_loss": 4.107922699754665,
      "tokens_seen": 324993024
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004553059177532598,
      "loss": 3.0946,
      "theoretical_loss": 4.107826780046074,
      "tokens_seen": 325058560
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045529588766298894,
      "loss": 3.2085,
      "theoretical_loss": 4.107730885087763,
      "tokens_seen": 325124096
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004552858575727182,
      "loss": 3.2909,
      "theoretical_loss": 4.107635014868358,
      "tokens_seen": 325189632
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045527582748244736,
      "loss": 2.9999,
      "theoretical_loss": 4.107539169376494,
      "tokens_seen": 325255168
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045526579739217654,
      "loss": 3.2132,
      "theoretical_loss": 4.107443348600812,
      "tokens_seen": 325320704
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004552557673019057,
      "loss": 3.0767,
      "theoretical_loss": 4.107347552529961,
      "tokens_seen": 325386240
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004552457372116349,
      "loss": 3.0053,
      "theoretical_loss": 4.107251781152598,
      "tokens_seen": 325451776
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004552357071213641,
      "loss": 3.2347,
      "theoretical_loss": 4.107156034457387,
      "tokens_seen": 325517312
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004552256770310933,
      "loss": 3.1149,
      "theoretical_loss": 4.107060312433,
      "tokens_seen": 325582848
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045521564694082244,
      "loss": 3.1885,
      "theoretical_loss": 4.106964615068115,
      "tokens_seen": 325648384
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004552056168505517,
      "loss": 3.0704,
      "theoretical_loss": 4.10686894235142,
      "tokens_seen": 325713920
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004551955867602808,
      "loss": 3.1798,
      "theoretical_loss": 4.106773294271608,
      "tokens_seen": 325779456
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045518555667001004,
      "loss": 2.8809,
      "theoretical_loss": 4.1066776708173816,
      "tokens_seen": 325844992
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004551755265797392,
      "loss": 3.0213,
      "theoretical_loss": 4.106582071977449,
      "tokens_seen": 325910528
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004551654964894684,
      "loss": 3.2495,
      "theoretical_loss": 4.106486497740526,
      "tokens_seen": 325976064
    },
    {
      "epoch": 0.1,
      "objective/train/advantage_avg": 0.4947391152381897,
      "objective/train/docs_used": 196731,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.307795524597168,
      "objective/train/original_loss": 3.3077950477600098,
      "objective/train/theoretical_loss": 4.106390948095338,
      "objective/train/tokens_used": 346501600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2448437660932541,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507185459136963,
      "objective/train/weighted_lm_loss": 3.4755401611328125,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0304546356201172,
      "theoretical_loss": 4.106390948095338,
      "tokens_seen": 326041600
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004551554663991976,
      "loss": 3.2447,
      "theoretical_loss": 4.106390948095338,
      "tokens_seen": 326041600
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004551454363089268,
      "loss": 3.3116,
      "theoretical_loss": 4.106295423030614,
      "tokens_seen": 326107136
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045513540621865595,
      "loss": 3.0965,
      "theoretical_loss": 4.106199922535095,
      "tokens_seen": 326172672
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004551253761283852,
      "loss": 3.1587,
      "theoretical_loss": 4.106104446597525,
      "tokens_seen": 326238208
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004551153460381143,
      "loss": 3.2697,
      "theoretical_loss": 4.10600899520666,
      "tokens_seen": 326303744
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045510531594784354,
      "loss": 3.3029,
      "theoretical_loss": 4.1059135683512595,
      "tokens_seen": 326369280
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004550952858575727,
      "loss": 3.287,
      "theoretical_loss": 4.105818166020091,
      "tokens_seen": 326434816
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004550852557673019,
      "loss": 3.3233,
      "theoretical_loss": 4.105722788201932,
      "tokens_seen": 326500352
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004550752256770311,
      "loss": 3.2712,
      "theoretical_loss": 4.105627434885564,
      "tokens_seen": 326565888
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045506519558676027,
      "loss": 3.1566,
      "theoretical_loss": 4.10553210605978,
      "tokens_seen": 326631424
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045505516549648945,
      "loss": 3.2229,
      "theoretical_loss": 4.105436801713375,
      "tokens_seen": 326696960
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004550451354062187,
      "loss": 3.1581,
      "theoretical_loss": 4.105341521835155,
      "tokens_seen": 326762496
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004550351053159478,
      "loss": 3.2165,
      "theoretical_loss": 4.105246266413934,
      "tokens_seen": 326828032
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045502507522567705,
      "loss": 3.0829,
      "theoretical_loss": 4.10515103543853,
      "tokens_seen": 326893568
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045501504513540623,
      "loss": 2.992,
      "theoretical_loss": 4.105055828897772,
      "tokens_seen": 326959104
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004550050150451354,
      "loss": 3.1753,
      "theoretical_loss": 4.104960646780494,
      "tokens_seen": 327024640
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004549949849548646,
      "loss": 3.0873,
      "theoretical_loss": 4.104865489075538,
      "tokens_seen": 327090176
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045498495486459377,
      "loss": 3.292,
      "theoretical_loss": 4.104770355771754,
      "tokens_seen": 327155712
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045497492477432295,
      "loss": 3.0732,
      "theoretical_loss": 4.104675246857997,
      "tokens_seen": 327221248
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004549648946840522,
      "loss": 3.1762,
      "theoretical_loss": 4.104580162323132,
      "tokens_seen": 327286784
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045495486459378137,
      "loss": 3.2361,
      "theoretical_loss": 4.104485102156031,
      "tokens_seen": 327352320
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045494483450351055,
      "loss": 3.3166,
      "theoretical_loss": 4.104390066345571,
      "tokens_seen": 327417856
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045493480441323973,
      "loss": 3.1559,
      "theoretical_loss": 4.104295054880639,
      "tokens_seen": 327483392
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004549247743229689,
      "loss": 2.9865,
      "theoretical_loss": 4.10420006775013,
      "tokens_seen": 327548928
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045491474423269815,
      "loss": 3.2961,
      "theoretical_loss": 4.10410510494294,
      "tokens_seen": 327614464
    },
    {
      "debugging/Self-BLEU-5": 0.48445945981218463,
      "debugging/distinct-1-grams": 0.7948268571619983,
      "debugging/distinct-2-grams": 0.9739697199580383,
      "debugging/entropy-1-grams": 6.01135557249139,
      "debugging/entropy-2-grams": 6.827639403579008,
      "debugging/length": 512.9230769230769,
      "debugging/num_segments": 13,
      "debugging/raw_token_scores_avg": 0.00634933914989233,
      "debugging/raw_token_scores_std": 0.008881147019565105,
      "debugging/score": 0.008299263930001547,
      "debugging/score_std": 0.00590965264802656,
      "epoch": 0.1,
      "objective/train/advantage_avg": 0.4936501979827881,
      "objective/train/docs_used": 197513,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.1297130584716797,
      "objective/train/original_loss": 3.1297128200531006,
      "objective/train/theoretical_loss": 4.10401016644798,
      "objective/train/tokens_used": 348140000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2437693178653717,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050604224205017,
      "objective/train/weighted_lm_loss": 3.2880423069000244,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 4.10401016644798,
      "tokens_seen": 327680000
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004549047141424273,
      "loss": 3.2995,
      "theoretical_loss": 4.10401016644798,
      "tokens_seen": 327680000
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004548946840521565,
      "loss": 3.0897,
      "theoretical_loss": 4.103915252254165,
      "tokens_seen": 327745536
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045488465396188564,
      "loss": 3.1175,
      "theoretical_loss": 4.103820362350415,
      "tokens_seen": 327811072
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004548746238716149,
      "loss": 3.1878,
      "theoretical_loss": 4.103725496725662,
      "tokens_seen": 327876608
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045486459378134405,
      "loss": 3.1716,
      "theoretical_loss": 4.103630655368843,
      "tokens_seen": 327942144
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045485456369107324,
      "loss": 3.1857,
      "theoretical_loss": 4.1035358382689004,
      "tokens_seen": 328007680
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004548445336008024,
      "loss": 3.1851,
      "theoretical_loss": 4.103441045414786,
      "tokens_seen": 328073216
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045483450351053165,
      "loss": 3.1422,
      "theoretical_loss": 4.103346276795459,
      "tokens_seen": 328138752
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004548244734202608,
      "loss": 2.7748,
      "theoretical_loss": 4.103251532399884,
      "tokens_seen": 328204288
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045481444332999,
      "loss": 2.948,
      "theoretical_loss": 4.103156812217035,
      "tokens_seen": 328269824
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045480441323971914,
      "loss": 3.4204,
      "theoretical_loss": 4.1030621162358925,
      "tokens_seen": 328335360
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004547943831494484,
      "loss": 3.1408,
      "theoretical_loss": 4.102967444445444,
      "tokens_seen": 328400896
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045478435305917756,
      "loss": 3.2126,
      "theoretical_loss": 4.102872796834683,
      "tokens_seen": 328466432
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045477432296890674,
      "loss": 3.4342,
      "theoretical_loss": 4.102778173392611,
      "tokens_seen": 328531968
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004547642928786359,
      "loss": 3.3203,
      "theoretical_loss": 4.10268357410824,
      "tokens_seen": 328597504
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004547542627883651,
      "loss": 3.144,
      "theoretical_loss": 4.1025889989705835,
      "tokens_seen": 328663040
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004547442326980943,
      "loss": 3.1109,
      "theoretical_loss": 4.102494447968666,
      "tokens_seen": 328728576
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004547342026078235,
      "loss": 3.3399,
      "theoretical_loss": 4.102399921091518,
      "tokens_seen": 328794112
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045472417251755264,
      "loss": 3.246,
      "theoretical_loss": 4.102305418328176,
      "tokens_seen": 328859648
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004547141424272819,
      "loss": 3.2695,
      "theoretical_loss": 4.102210939667687,
      "tokens_seen": 328925184
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.000454704112337011,
      "loss": 3.1946,
      "theoretical_loss": 4.102116485099103,
      "tokens_seen": 328990720
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045469408224674024,
      "loss": 3.2342,
      "theoretical_loss": 4.102022054611481,
      "tokens_seen": 329056256
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004546840521564694,
      "loss": 3.2628,
      "theoretical_loss": 4.1019276481938896,
      "tokens_seen": 329121792
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004546740220661986,
      "loss": 3.2101,
      "theoretical_loss": 4.101833265835402,
      "tokens_seen": 329187328
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004546639919759278,
      "loss": 3.173,
      "theoretical_loss": 4.101738907525098,
      "tokens_seen": 329252864
    },
    {
      "epoch": 0.1,
      "objective/train/advantage_avg": 0.49452680349349976,
      "objective/train/docs_used": 198138,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.4164631366729736,
      "objective/train/original_loss": 3.4164628982543945,
      "objective/train/theoretical_loss": 4.101644573252066,
      "objective/train/tokens_used": 349778400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2446473091840744,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506962537765503,
      "objective/train/weighted_lm_loss": 3.5895261764526367,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0253151655197144,
      "theoretical_loss": 4.101644573252066,
      "tokens_seen": 329318400
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.000454653961885657,
      "loss": 2.9853,
      "theoretical_loss": 4.101644573252066,
      "tokens_seen": 329318400
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045464393179538615,
      "loss": 3.1771,
      "theoretical_loss": 4.1015502630054,
      "tokens_seen": 329383936
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004546339017051154,
      "loss": 3.1553,
      "theoretical_loss": 4.101455976774204,
      "tokens_seen": 329449472
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004546238716148445,
      "loss": 3.1629,
      "theoretical_loss": 4.101361714547585,
      "tokens_seen": 329515008
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045461384152457374,
      "loss": 3.3843,
      "theoretical_loss": 4.1012674763146615,
      "tokens_seen": 329580544
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004546038114343029,
      "loss": 3.1498,
      "theoretical_loss": 4.101173262064555,
      "tokens_seen": 329646080
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004545937813440321,
      "loss": 3.4985,
      "theoretical_loss": 4.101079071786397,
      "tokens_seen": 329711616
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004545837512537613,
      "loss": 3.4214,
      "theoretical_loss": 4.100984905469324,
      "tokens_seen": 329777152
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045457372116349047,
      "loss": 3.0725,
      "theoretical_loss": 4.100890763102481,
      "tokens_seen": 329842688
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045456369107321965,
      "loss": 3.4852,
      "theoretical_loss": 4.10079664467502,
      "tokens_seen": 329908224
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004545536609829489,
      "loss": 2.9479,
      "theoretical_loss": 4.1007025501761,
      "tokens_seen": 329973760
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.000454543630892678,
      "loss": 3.3623,
      "theoretical_loss": 4.100608479594887,
      "tokens_seen": 330039296
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045453360080240725,
      "loss": 3.3719,
      "theoretical_loss": 4.100514432920552,
      "tokens_seen": 330104832
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045452357071213643,
      "loss": 3.2256,
      "theoretical_loss": 4.100420410142277,
      "tokens_seen": 330170368
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004545135406218656,
      "loss": 3.1896,
      "theoretical_loss": 4.100326411249248,
      "tokens_seen": 330235904
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004545035105315948,
      "loss": 3.1364,
      "theoretical_loss": 4.100232436230659,
      "tokens_seen": 330301440
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045449348044132397,
      "loss": 3.1254,
      "theoretical_loss": 4.100138485075711,
      "tokens_seen": 330366976
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045448345035105315,
      "loss": 2.8876,
      "theoretical_loss": 4.1000445577736135,
      "tokens_seen": 330432512
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004544734202607824,
      "loss": 2.8058,
      "theoretical_loss": 4.099950654313579,
      "tokens_seen": 330498048
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004544633901705115,
      "loss": 3.1678,
      "theoretical_loss": 4.099856774684832,
      "tokens_seen": 330563584
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045445336008024075,
      "loss": 2.9185,
      "theoretical_loss": 4.0997629188766,
      "tokens_seen": 330629120
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004544433299899699,
      "loss": 3.1936,
      "theoretical_loss": 4.099669086878119,
      "tokens_seen": 330694656
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004544332998996991,
      "loss": 3.0409,
      "theoretical_loss": 4.099575278678634,
      "tokens_seen": 330760192
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004544232698094283,
      "loss": 3.4225,
      "theoretical_loss": 4.099481494267393,
      "tokens_seen": 330825728
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004544132397191575,
      "loss": 3.0742,
      "theoretical_loss": 4.0993877336336535,
      "tokens_seen": 330891264
    },
    {
      "epoch": 0.1,
      "objective/train/advantage_avg": 0.4929335415363312,
      "objective/train/docs_used": 199416,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.983245611190796,
      "objective/train/original_loss": 2.983245611190796,
      "objective/train/theoretical_loss": 4.099293996766681,
      "objective/train/tokens_used": 351416800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24307529628276825,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505290031433105,
      "objective/train/weighted_lm_loss": 3.134032726287842,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 4.099293996766681,
      "tokens_seen": 330956800
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045440320962888666,
      "loss": 2.9901,
      "theoretical_loss": 4.099293996766681,
      "tokens_seen": 330956800
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045439317953861584,
      "loss": 2.9767,
      "theoretical_loss": 4.099200283655745,
      "tokens_seen": 331022336
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.000454383149448345,
      "loss": 3.0532,
      "theoretical_loss": 4.099106594290124,
      "tokens_seen": 331087872
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045437311935807425,
      "loss": 3.5069,
      "theoretical_loss": 4.099012928659102,
      "tokens_seen": 331153408
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004543630892678034,
      "loss": 3.3961,
      "theoretical_loss": 4.098919286751972,
      "tokens_seen": 331218944
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004543530591775326,
      "loss": 2.9655,
      "theoretical_loss": 4.0988256685580335,
      "tokens_seen": 331284480
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004543430290872618,
      "loss": 3.3582,
      "theoretical_loss": 4.098732074066591,
      "tokens_seen": 331350016
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.000454332998996991,
      "loss": 3.1087,
      "theoretical_loss": 4.098638503266957,
      "tokens_seen": 331415552
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045432296890672016,
      "loss": 3.017,
      "theoretical_loss": 4.098544956148452,
      "tokens_seen": 331481088
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045431293881644934,
      "loss": 3.3377,
      "theoretical_loss": 4.098451432700403,
      "tokens_seen": 331546624
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004543029087261785,
      "loss": 3.5497,
      "theoretical_loss": 4.098357932912142,
      "tokens_seen": 331612160
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045429287863590776,
      "loss": 3.344,
      "theoretical_loss": 4.098264456773012,
      "tokens_seen": 331677696
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004542828485456369,
      "loss": 3.1696,
      "theoretical_loss": 4.098171004272357,
      "tokens_seen": 331743232
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004542728184553661,
      "loss": 3.3481,
      "theoretical_loss": 4.098077575399533,
      "tokens_seen": 331808768
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045426278836509525,
      "loss": 3.2664,
      "theoretical_loss": 4.097984170143901,
      "tokens_seen": 331874304
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004542527582748245,
      "loss": 3.3043,
      "theoretical_loss": 4.09789078849483,
      "tokens_seen": 331939840
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045424272818455366,
      "loss": 3.0657,
      "theoretical_loss": 4.097797430441693,
      "tokens_seen": 332005376
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045423269809428284,
      "loss": 3.1071,
      "theoretical_loss": 4.097704095973873,
      "tokens_seen": 332070912
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.000454222668004012,
      "loss": 2.9425,
      "theoretical_loss": 4.097610785080757,
      "tokens_seen": 332136448
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004542126379137412,
      "loss": 3.055,
      "theoretical_loss": 4.097517497751743,
      "tokens_seen": 332201984
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045420260782347044,
      "loss": 3.0797,
      "theoretical_loss": 4.0974242339762315,
      "tokens_seen": 332267520
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004541925777331996,
      "loss": 3.3614,
      "theoretical_loss": 4.097330993743633,
      "tokens_seen": 332333056
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004541825476429288,
      "loss": 3.0125,
      "theoretical_loss": 4.097237777043363,
      "tokens_seen": 332398592
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.000454172517552658,
      "loss": 3.0913,
      "theoretical_loss": 4.097144583864844,
      "tokens_seen": 332464128
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004541624874623872,
      "loss": 3.176,
      "theoretical_loss": 4.097051414197506,
      "tokens_seen": 332529664
    },
    {
      "epoch": 0.1,
      "objective/train/advantage_avg": 0.4942306578159332,
      "objective/train/docs_used": 200524,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.114231586456299,
      "objective/train/original_loss": 3.114231824874878,
      "objective/train/theoretical_loss": 4.096958268030786,
      "objective/train/tokens_used": 353055200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2443409264087677,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506651401519775,
      "objective/train/weighted_lm_loss": 3.271825075149536,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0417574644088745,
      "theoretical_loss": 4.096958268030786,
      "tokens_seen": 332595200
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045415245737211635,
      "loss": 3.3424,
      "theoretical_loss": 4.096958268030786,
      "tokens_seen": 332595200
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004541424272818456,
      "loss": 2.9604,
      "theoretical_loss": 4.096865145354127,
      "tokens_seen": 332660736
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004541323971915747,
      "loss": 3.2767,
      "theoretical_loss": 4.09677204615698,
      "tokens_seen": 332726272
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045412236710130395,
      "loss": 3.0937,
      "theoretical_loss": 4.096678970428801,
      "tokens_seen": 332791808
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004541123370110331,
      "loss": 2.9344,
      "theoretical_loss": 4.096585918159054,
      "tokens_seen": 332857344
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004541023069207623,
      "loss": 3.308,
      "theoretical_loss": 4.096492889337211,
      "tokens_seen": 332922880
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004540922768304915,
      "loss": 3.4674,
      "theoretical_loss": 4.096399883952747,
      "tokens_seen": 332988416
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045408224674022067,
      "loss": 3.0363,
      "theoretical_loss": 4.096306901995149,
      "tokens_seen": 333053952
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045407221664994985,
      "loss": 3.3096,
      "theoretical_loss": 4.096213943453906,
      "tokens_seen": 333119488
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004540621865596791,
      "loss": 3.0111,
      "theoretical_loss": 4.096121008318517,
      "tokens_seen": 333185024
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004540521564694082,
      "loss": 3.2563,
      "theoretical_loss": 4.096028096578486,
      "tokens_seen": 333250560
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045404212637913745,
      "loss": 2.9846,
      "theoretical_loss": 4.095935208223324,
      "tokens_seen": 333316096
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045403209628886663,
      "loss": 3.0259,
      "theoretical_loss": 4.09584234324255,
      "tokens_seen": 333381632
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004540220661985958,
      "loss": 3.047,
      "theoretical_loss": 4.095749501625689,
      "tokens_seen": 333447168
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.000454012036108325,
      "loss": 3.3406,
      "theoretical_loss": 4.095656683362272,
      "tokens_seen": 333512704
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045400200601805417,
      "loss": 3.0607,
      "theoretical_loss": 4.095563888441838,
      "tokens_seen": 333578240
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045399197592778335,
      "loss": 3.2887,
      "theoretical_loss": 4.095471116853931,
      "tokens_seen": 333643776
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004539819458375126,
      "loss": 3.0465,
      "theoretical_loss": 4.095378368588103,
      "tokens_seen": 333709312
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004539719157472417,
      "loss": 3.2766,
      "theoretical_loss": 4.095285643633915,
      "tokens_seen": 333774848
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045396188565697095,
      "loss": 3.0804,
      "theoretical_loss": 4.095192941980929,
      "tokens_seen": 333840384
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004539518555667001,
      "loss": 3.2992,
      "theoretical_loss": 4.095100263618718,
      "tokens_seen": 333905920
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004539418254764293,
      "loss": 3.3428,
      "theoretical_loss": 4.0950076085368625,
      "tokens_seen": 333971456
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004539317953861585,
      "loss": 3.0463,
      "theoretical_loss": 4.0949149767249455,
      "tokens_seen": 334036992
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004539217652958877,
      "loss": 3.2038,
      "theoretical_loss": 4.0948223681725615,
      "tokens_seen": 334102528
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045391173520561686,
      "loss": 3.453,
      "theoretical_loss": 4.094729782869308,
      "tokens_seen": 334168064
    },
    {
      "epoch": 0.1,
      "objective/train/advantage_avg": 0.4939344823360443,
      "objective/train/docs_used": 201196,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.278226852416992,
      "objective/train/original_loss": 3.2782273292541504,
      "objective/train/theoretical_loss": 4.09463722080479,
      "objective/train/tokens_used": 354693600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24409575760364532,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506342649459839,
      "objective/train/weighted_lm_loss": 3.4443368911743164,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0304546356201172,
      "theoretical_loss": 4.09463722080479,
      "tokens_seen": 334233600
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045390170511534604,
      "loss": 3.4534,
      "theoretical_loss": 4.09463722080479,
      "tokens_seen": 334233600
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004538916750250752,
      "loss": 3.255,
      "theoretical_loss": 4.094544681968621,
      "tokens_seen": 334299136
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045388164493480445,
      "loss": 3.1367,
      "theoretical_loss": 4.094452166350419,
      "tokens_seen": 334364672
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004538716148445336,
      "loss": 3.1219,
      "theoretical_loss": 4.0943596739398105,
      "tokens_seen": 334430208
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004538615847542628,
      "loss": 3.0472,
      "theoretical_loss": 4.094267204726426,
      "tokens_seen": 334495744
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.000453851554663992,
      "loss": 3.2621,
      "theoretical_loss": 4.094174758699907,
      "tokens_seen": 334561280
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004538415245737212,
      "loss": 3.1038,
      "theoretical_loss": 4.0940823358498974,
      "tokens_seen": 334626816
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045383149448345036,
      "loss": 3.5836,
      "theoretical_loss": 4.093989936166049,
      "tokens_seen": 334692352
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045382146439317954,
      "loss": 3.2494,
      "theoretical_loss": 4.093897559638021,
      "tokens_seen": 334757888
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004538114343029087,
      "loss": 3.1535,
      "theoretical_loss": 4.09380520625548,
      "tokens_seen": 334823424
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045380140421263796,
      "loss": 3.1256,
      "theoretical_loss": 4.093712876008097,
      "tokens_seen": 334888960
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004537913741223671,
      "loss": 3.1111,
      "theoretical_loss": 4.093620568885552,
      "tokens_seen": 334954496
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004537813440320963,
      "loss": 3.3528,
      "theoretical_loss": 4.093528284877528,
      "tokens_seen": 335020032
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045377131394182545,
      "loss": 3.0471,
      "theoretical_loss": 4.093436023973719,
      "tokens_seen": 335085568
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004537612838515547,
      "loss": 3.4041,
      "theoretical_loss": 4.093343786163823,
      "tokens_seen": 335151104
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045375125376128386,
      "loss": 3.079,
      "theoretical_loss": 4.093251571437545,
      "tokens_seen": 335216640
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045374122367101304,
      "loss": 3.1973,
      "theoretical_loss": 4.093159379784597,
      "tokens_seen": 335282176
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004537311935807422,
      "loss": 3.3399,
      "theoretical_loss": 4.093067211194698,
      "tokens_seen": 335347712
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004537211634904714,
      "loss": 3.2896,
      "theoretical_loss": 4.092975065657572,
      "tokens_seen": 335413248
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004537111334002006,
      "loss": 3.3429,
      "theoretical_loss": 4.092882943162952,
      "tokens_seen": 335478784
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004537011033099298,
      "loss": 3.2923,
      "theoretical_loss": 4.092790843700574,
      "tokens_seen": 335544320
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045369107321965895,
      "loss": 3.376,
      "theoretical_loss": 4.092698767260185,
      "tokens_seen": 335609856
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004536810431293882,
      "loss": 3.3776,
      "theoretical_loss": 4.092606713831534,
      "tokens_seen": 335675392
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045367101303911737,
      "loss": 3.222,
      "theoretical_loss": 4.0925146834043815,
      "tokens_seen": 335740928
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045366098294884655,
      "loss": 3.3332,
      "theoretical_loss": 4.092422675968489,
      "tokens_seen": 335806464
    },
    {
      "epoch": 0.1,
      "objective/train/advantage_avg": 0.49477964639663696,
      "objective/train/docs_used": 202460,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.3561666011810303,
      "objective/train/original_loss": 3.356166362762451,
      "objective/train/theoretical_loss": 4.092330691513631,
      "objective/train/tokens_used": 356332000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24487102031707764,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507228374481201,
      "objective/train/weighted_lm_loss": 3.5263192653656006,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0408108234405518,
      "theoretical_loss": 4.092330691513631,
      "tokens_seen": 335872000
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045365095285857573,
      "loss": 3.1706,
      "theoretical_loss": 4.092330691513631,
      "tokens_seen": 335872000
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004536409227683049,
      "loss": 3.2802,
      "theoretical_loss": 4.092238730029582,
      "tokens_seen": 335937536
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004536308926780341,
      "loss": 3.1988,
      "theoretical_loss": 4.092146791506127,
      "tokens_seen": 336003072
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004536208625877633,
      "loss": 3.2923,
      "theoretical_loss": 4.092054875933057,
      "tokens_seen": 336068608
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045361083249749245,
      "loss": 3.3353,
      "theoretical_loss": 4.091962983300168,
      "tokens_seen": 336134144
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004536008024072217,
      "loss": 3.0964,
      "theoretical_loss": 4.091871113597266,
      "tokens_seen": 336199680
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004535907723169508,
      "loss": 3.1608,
      "theoretical_loss": 4.091779266814158,
      "tokens_seen": 336265216
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045358074222668005,
      "loss": 3.3777,
      "theoretical_loss": 4.091687442940664,
      "tokens_seen": 336330752
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045357071213640923,
      "loss": 3.3847,
      "theoretical_loss": 4.091595641966604,
      "tokens_seen": 336396288
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004535606820461384,
      "loss": 2.9843,
      "theoretical_loss": 4.09150386388181,
      "tokens_seen": 336461824
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004535506519558676,
      "loss": 3.0636,
      "theoretical_loss": 4.091412108676117,
      "tokens_seen": 336527360
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045354062186559683,
      "loss": 3.3159,
      "theoretical_loss": 4.091320376339368,
      "tokens_seen": 336592896
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045353059177532596,
      "loss": 3.3452,
      "theoretical_loss": 4.091228666861413,
      "tokens_seen": 336658432
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004535205616850552,
      "loss": 3.154,
      "theoretical_loss": 4.091136980232107,
      "tokens_seen": 336723968
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004535105315947843,
      "loss": 3.4314,
      "theoretical_loss": 4.091045316441311,
      "tokens_seen": 336789504
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045350050150451355,
      "loss": 3.1618,
      "theoretical_loss": 4.090953675478895,
      "tokens_seen": 336855040
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045349047141424274,
      "loss": 3.109,
      "theoretical_loss": 4.090862057334734,
      "tokens_seen": 336920576
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004534804413239719,
      "loss": 3.441,
      "theoretical_loss": 4.090770461998709,
      "tokens_seen": 336986112
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004534704112337011,
      "loss": 3.3112,
      "theoretical_loss": 4.090678889460707,
      "tokens_seen": 337051648
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004534603811434303,
      "loss": 3.071,
      "theoretical_loss": 4.090587339710625,
      "tokens_seen": 337117184
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004534503510531595,
      "loss": 3.2707,
      "theoretical_loss": 4.090495812738362,
      "tokens_seen": 337182720
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004534403209628887,
      "loss": 3.4787,
      "theoretical_loss": 4.090404308533825,
      "tokens_seen": 337248256
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004534302908726179,
      "loss": 2.8634,
      "theoretical_loss": 4.090312827086929,
      "tokens_seen": 337313792
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045342026078234706,
      "loss": 2.9138,
      "theoretical_loss": 4.090221368387593,
      "tokens_seen": 337379328
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045341023069207624,
      "loss": 3.2211,
      "theoretical_loss": 4.090129932425745,
      "tokens_seen": 337444864
    },
    {
      "epoch": 0.1,
      "objective/train/advantage_avg": 0.49518266320228577,
      "objective/train/docs_used": 203033,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.4789485931396484,
      "objective/train/original_loss": 3.4789483547210693,
      "objective/train/theoretical_loss": 4.0900385191913164,
      "objective/train/tokens_used": 357970400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24525649845600128,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050765037536621,
      "objective/train/weighted_lm_loss": 3.6554462909698486,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0304546356201172,
      "theoretical_loss": 4.0900385191913164,
      "tokens_seen": 337510400
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004534002006018054,
      "loss": 3.5438,
      "theoretical_loss": 4.0900385191913164,
      "tokens_seen": 337510400
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045339017051153465,
      "loss": 3.1413,
      "theoretical_loss": 4.089947128674248,
      "tokens_seen": 337575936
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004533801404212638,
      "loss": 3.2007,
      "theoretical_loss": 4.089855760864484,
      "tokens_seen": 337641472
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.000453370110330993,
      "loss": 3.0673,
      "theoretical_loss": 4.089764415751978,
      "tokens_seen": 337707008
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004533600802407222,
      "loss": 3.2954,
      "theoretical_loss": 4.089673093326688,
      "tokens_seen": 337772544
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004533500501504514,
      "loss": 3.5089,
      "theoretical_loss": 4.08958179357858,
      "tokens_seen": 337838080
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045334002006018056,
      "loss": 2.9505,
      "theoretical_loss": 4.089490516497624,
      "tokens_seen": 337903616
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045332998996990974,
      "loss": 2.9178,
      "theoretical_loss": 4.089399262073798,
      "tokens_seen": 337969152
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004533199598796389,
      "loss": 3.1305,
      "theoretical_loss": 4.089308030297087,
      "tokens_seen": 338034688
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045330992978936816,
      "loss": 3.0787,
      "theoretical_loss": 4.08921682115748,
      "tokens_seen": 338100224
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004532998996990973,
      "loss": 3.0737,
      "theoretical_loss": 4.089125634644975,
      "tokens_seen": 338165760
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004532898696088265,
      "loss": 3.1488,
      "theoretical_loss": 4.089034470749575,
      "tokens_seen": 338231296
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045327983951855565,
      "loss": 3.2827,
      "theoretical_loss": 4.08894332946129,
      "tokens_seen": 338296832
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004532698094282849,
      "loss": 3.4421,
      "theoretical_loss": 4.088852210770135,
      "tokens_seen": 338362368
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045325977933801406,
      "loss": 3.1565,
      "theoretical_loss": 4.088761114666133,
      "tokens_seen": 338427904
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045324974924774324,
      "loss": 3.0687,
      "theoretical_loss": 4.088670041139312,
      "tokens_seen": 338493440
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004532397191574724,
      "loss": 3.1835,
      "theoretical_loss": 4.088578990179708,
      "tokens_seen": 338558976
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004532296890672016,
      "loss": 2.7972,
      "theoretical_loss": 4.088487961777361,
      "tokens_seen": 338624512
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004532196589769308,
      "loss": 3.2381,
      "theoretical_loss": 4.0883969559223186,
      "tokens_seen": 338690048
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045320962888666,
      "loss": 3.108,
      "theoretical_loss": 4.088305972604636,
      "tokens_seen": 338755584
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045319959879638915,
      "loss": 3.1857,
      "theoretical_loss": 4.088215011814372,
      "tokens_seen": 338821120
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004531895687061184,
      "loss": 2.9521,
      "theoretical_loss": 4.088124073541594,
      "tokens_seen": 338886656
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045317953861584757,
      "loss": 3.3776,
      "theoretical_loss": 4.088033157776374,
      "tokens_seen": 338952192
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045316950852557675,
      "loss": 3.1436,
      "theoretical_loss": 4.087942264508792,
      "tokens_seen": 339017728
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045315947843530593,
      "loss": 3.424,
      "theoretical_loss": 4.087851393728933,
      "tokens_seen": 339083264
    },
    {
      "epoch": 0.1,
      "objective/train/advantage_avg": 0.4946117103099823,
      "objective/train/docs_used": 204232,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.6656346321105957,
      "objective/train/original_loss": 2.6656343936920166,
      "objective/train/theoretical_loss": 4.087760545426889,
      "objective/train/tokens_used": 359608800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24468956887722015,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050705075263977,
      "objective/train/weighted_lm_loss": 2.80072021484375,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0253151655197144,
      "theoretical_loss": 4.087760545426889,
      "tokens_seen": 339148800
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004531494483450351,
      "loss": 3.0198,
      "theoretical_loss": 4.087760545426889,
      "tokens_seen": 339148800
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004531394182547643,
      "loss": 3.1247,
      "theoretical_loss": 4.087669719592757,
      "tokens_seen": 339214336
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004531293881644935,
      "loss": 3.3697,
      "theoretical_loss": 4.087578916216642,
      "tokens_seen": 339279872
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045311935807422265,
      "loss": 3.4467,
      "theoretical_loss": 4.087488135288654,
      "tokens_seen": 339345408
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004531093279839519,
      "loss": 3.289,
      "theoretical_loss": 4.0873973767989105,
      "tokens_seen": 339410944
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.000453099297893681,
      "loss": 2.9835,
      "theoretical_loss": 4.087306640737534,
      "tokens_seen": 339476480
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045308926780341025,
      "loss": 3.2206,
      "theoretical_loss": 4.087215927094654,
      "tokens_seen": 339542016
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045307923771313943,
      "loss": 3.1815,
      "theoretical_loss": 4.087125235860405,
      "tokens_seen": 339607552
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004530692076228686,
      "loss": 3.2228,
      "theoretical_loss": 4.087034567024931,
      "tokens_seen": 339673088
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004530591775325978,
      "loss": 3.0433,
      "theoretical_loss": 4.086943920578378,
      "tokens_seen": 339738624
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045304914744232703,
      "loss": 3.1415,
      "theoretical_loss": 4.086853296510902,
      "tokens_seen": 339804160
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045303911735205616,
      "loss": 2.9373,
      "theoretical_loss": 4.086762694812661,
      "tokens_seen": 339869696
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004530290872617854,
      "loss": 3.1489,
      "theoretical_loss": 4.0866721154738235,
      "tokens_seen": 339935232
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004530190571715145,
      "loss": 3.2971,
      "theoretical_loss": 4.086581558484562,
      "tokens_seen": 340000768
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045300902708124375,
      "loss": 2.9705,
      "theoretical_loss": 4.086491023835055,
      "tokens_seen": 340066304
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045299899699097294,
      "loss": 3.2306,
      "theoretical_loss": 4.08640051151549,
      "tokens_seen": 340131840
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004529889669007021,
      "loss": 3.4619,
      "theoretical_loss": 4.086310021516056,
      "tokens_seen": 340197376
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004529789368104313,
      "loss": 3.131,
      "theoretical_loss": 4.086219553826952,
      "tokens_seen": 340262912
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004529689067201605,
      "loss": 3.2574,
      "theoretical_loss": 4.086129108438381,
      "tokens_seen": 340328448
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045295887662988966,
      "loss": 3.0174,
      "theoretical_loss": 4.0860386853405535,
      "tokens_seen": 340393984
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004529488465396189,
      "loss": 3.0264,
      "theoretical_loss": 4.0859482845236865,
      "tokens_seen": 340459520
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.000452938816449348,
      "loss": 3.0646,
      "theoretical_loss": 4.085857905978002,
      "tokens_seen": 340525056
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045292878635907726,
      "loss": 3.3768,
      "theoretical_loss": 4.085767549693728,
      "tokens_seen": 340590592
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004529187562688064,
      "loss": 3.0074,
      "theoretical_loss": 4.0856772156611,
      "tokens_seen": 340656128
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004529087261785356,
      "loss": 3.1243,
      "theoretical_loss": 4.085586903870359,
      "tokens_seen": 340721664
    },
    {
      "epoch": 0.1,
      "objective/train/advantage_avg": 0.4942024350166321,
      "objective/train/docs_used": 205032,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.3295276165008545,
      "objective/train/original_loss": 3.3295278549194336,
      "objective/train/theoretical_loss": 4.085496614311752,
      "objective/train/tokens_used": 361247200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24429498612880707,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506621599197388,
      "objective/train/weighted_lm_loss": 3.4982388019561768,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0396549701690674,
      "theoretical_loss": 4.085496614311752,
      "tokens_seen": 340787200
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004528986960882648,
      "loss": 3.3293,
      "theoretical_loss": 4.085496614311752,
      "tokens_seen": 340787200
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.000452888665997994,
      "loss": 3.424,
      "theoretical_loss": 4.085406346975532,
      "tokens_seen": 340852736
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045287863590772316,
      "loss": 3.5574,
      "theoretical_loss": 4.085316101851959,
      "tokens_seen": 340918272
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004528686058174524,
      "loss": 2.9932,
      "theoretical_loss": 4.085225878931297,
      "tokens_seen": 340983808
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004528585757271815,
      "loss": 3.2781,
      "theoretical_loss": 4.0851356782038195,
      "tokens_seen": 341049344
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045284854563691076,
      "loss": 3.3427,
      "theoretical_loss": 4.085045499659804,
      "tokens_seen": 341114880
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004528385155466399,
      "loss": 3.1079,
      "theoretical_loss": 4.084955343289534,
      "tokens_seen": 341180416
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004528284854563691,
      "loss": 3.1517,
      "theoretical_loss": 4.084865209083299,
      "tokens_seen": 341245952
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004528184553660983,
      "loss": 3.3246,
      "theoretical_loss": 4.084775097031397,
      "tokens_seen": 341311488
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004528084252758275,
      "loss": 2.8128,
      "theoretical_loss": 4.084685007124127,
      "tokens_seen": 341377024
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045279839518555667,
      "loss": 3.2542,
      "theoretical_loss": 4.084594939351801,
      "tokens_seen": 341442560
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045278836509528585,
      "loss": 3.1246,
      "theoretical_loss": 4.084504893704732,
      "tokens_seen": 341508096
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045277833500501503,
      "loss": 3.1539,
      "theoretical_loss": 4.084414870173241,
      "tokens_seen": 341573632
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045276830491474426,
      "loss": 3.2008,
      "theoretical_loss": 4.084324868747653,
      "tokens_seen": 341639168
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004527582748244734,
      "loss": 3.2703,
      "theoretical_loss": 4.084234889418303,
      "tokens_seen": 341704704
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004527482447342026,
      "loss": 3.1326,
      "theoretical_loss": 4.084144932175528,
      "tokens_seen": 341770240
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045273821464393175,
      "loss": 3.2727,
      "theoretical_loss": 4.084054997009675,
      "tokens_seen": 341835776
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.000452728184553661,
      "loss": 3.3821,
      "theoretical_loss": 4.083965083911093,
      "tokens_seen": 341901312
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045271815446339017,
      "loss": 3.4467,
      "theoretical_loss": 4.08387519287014,
      "tokens_seen": 341966848
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045270812437311935,
      "loss": 3.4216,
      "theoretical_loss": 4.08378532387718,
      "tokens_seen": 342032384
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004526980942828486,
      "loss": 3.1787,
      "theoretical_loss": 4.08369547692258,
      "tokens_seen": 342097920
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045268806419257777,
      "loss": 2.9833,
      "theoretical_loss": 4.0836056519967165,
      "tokens_seen": 342163456
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045267803410230695,
      "loss": 3.0739,
      "theoretical_loss": 4.083515849089971,
      "tokens_seen": 342228992
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045266800401203613,
      "loss": 3.3047,
      "theoretical_loss": 4.083426068192731,
      "tokens_seen": 342294528
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004526579739217653,
      "loss": 3.1136,
      "theoretical_loss": 4.0833363092953885,
      "tokens_seen": 342360064
    },
    {
      "epoch": 0.1,
      "objective/train/advantage_avg": 0.49365076422691345,
      "objective/train/docs_used": 206109,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.181786298751831,
      "objective/train/original_loss": 3.18178653717041,
      "objective/train/theoretical_loss": 4.083246572388344,
      "objective/train/tokens_used": 362885600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24375072121620178,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050604224205017,
      "objective/train/weighted_lm_loss": 3.342766284942627,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0437887907028198,
      "theoretical_loss": 4.083246572388344,
      "tokens_seen": 342425600
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004526479438314945,
      "loss": 3.3459,
      "theoretical_loss": 4.083246572388344,
      "tokens_seen": 342425600
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004526379137412237,
      "loss": 2.9513,
      "theoretical_loss": 4.0831568574620025,
      "tokens_seen": 342491136
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045262788365095285,
      "loss": 3.1345,
      "theoretical_loss": 4.083067164506776,
      "tokens_seen": 342556672
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004526178535606821,
      "loss": 3.3982,
      "theoretical_loss": 4.082977493513082,
      "tokens_seen": 342622208
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004526078234704112,
      "loss": 3.0157,
      "theoretical_loss": 4.082887844471342,
      "tokens_seen": 342687744
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045259779338014045,
      "loss": 3.2582,
      "theoretical_loss": 4.0827982173719874,
      "tokens_seen": 342753280
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045258776328986963,
      "loss": 3.1673,
      "theoretical_loss": 4.082708612205454,
      "tokens_seen": 342818816
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004525777331995988,
      "loss": 3.0302,
      "theoretical_loss": 4.082619028962182,
      "tokens_seen": 342884352
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.000452567703109328,
      "loss": 3.0578,
      "theoretical_loss": 4.08252946763262,
      "tokens_seen": 342949888
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045255767301905723,
      "loss": 3.0385,
      "theoretical_loss": 4.0824399282072195,
      "tokens_seen": 343015424
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045254764292878636,
      "loss": 3.3004,
      "theoretical_loss": 4.082350410676442,
      "tokens_seen": 343080960
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004525376128385156,
      "loss": 3.3134,
      "theoretical_loss": 4.082260915030751,
      "tokens_seen": 343146496
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004525275827482447,
      "loss": 2.9048,
      "theoretical_loss": 4.082171441260621,
      "tokens_seen": 343212032
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045251755265797395,
      "loss": 3.0472,
      "theoretical_loss": 4.082081989356525,
      "tokens_seen": 343277568
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045250752256770314,
      "loss": 3.1835,
      "theoretical_loss": 4.08199255930895,
      "tokens_seen": 343343104
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004524974924774323,
      "loss": 2.9354,
      "theoretical_loss": 4.081903151108384,
      "tokens_seen": 343408640
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004524874623871615,
      "loss": 3.0108,
      "theoretical_loss": 4.081813764745322,
      "tokens_seen": 343474176
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004524774322968907,
      "loss": 3.2906,
      "theoretical_loss": 4.081724400210264,
      "tokens_seen": 343539712
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045246740220661986,
      "loss": 3.226,
      "theoretical_loss": 4.0816350574937195,
      "tokens_seen": 343605248
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004524573721163491,
      "loss": 2.9688,
      "theoretical_loss": 4.0815457365862,
      "tokens_seen": 343670784
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004524473420260782,
      "loss": 3.2757,
      "theoretical_loss": 4.081456437478225,
      "tokens_seen": 343736320
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045243731193580746,
      "loss": 3.1258,
      "theoretical_loss": 4.081367160160319,
      "tokens_seen": 343801856
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004524272818455366,
      "loss": 3.0552,
      "theoretical_loss": 4.081277904623013,
      "tokens_seen": 343867392
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004524172517552658,
      "loss": 3.2081,
      "theoretical_loss": 4.081188670856844,
      "tokens_seen": 343932928
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.000452407221664995,
      "loss": 2.7924,
      "theoretical_loss": 4.081099458852354,
      "tokens_seen": 343998464
    },
    {
      "epoch": 0.1,
      "objective/train/advantage_avg": 0.4940275549888611,
      "objective/train/docs_used": 206760,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.370387315750122,
      "objective/train/original_loss": 3.3703877925872803,
      "objective/train/theoretical_loss": 4.081010268600093,
      "objective/train/tokens_used": 364524000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24414381384849548,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050643801689148,
      "objective/train/weighted_lm_loss": 3.541198492050171,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0253151655197144,
      "theoretical_loss": 4.081010268600093,
      "tokens_seen": 344064000
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004523971915747242,
      "loss": 3.1361,
      "theoretical_loss": 4.081010268600093,
      "tokens_seen": 344064000
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045238716148445336,
      "loss": 3.441,
      "theoretical_loss": 4.080921100090614,
      "tokens_seen": 344129536
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004523771313941826,
      "loss": 3.1194,
      "theoretical_loss": 4.080831953314478,
      "tokens_seen": 344195072
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004523671013039117,
      "loss": 3.126,
      "theoretical_loss": 4.080742828262251,
      "tokens_seen": 344260608
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045235707121364096,
      "loss": 3.2655,
      "theoretical_loss": 4.080653724924504,
      "tokens_seen": 344326144
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004523470411233701,
      "loss": 2.9922,
      "theoretical_loss": 4.080564643291818,
      "tokens_seen": 344391680
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004523370110330993,
      "loss": 3.0765,
      "theoretical_loss": 4.080475583354775,
      "tokens_seen": 344457216
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004523269809428285,
      "loss": 3.1867,
      "theoretical_loss": 4.0803865451039645,
      "tokens_seen": 344522752
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004523169508525577,
      "loss": 3.2326,
      "theoretical_loss": 4.080297528529983,
      "tokens_seen": 344588288
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045230692076228687,
      "loss": 3.0903,
      "theoretical_loss": 4.080208533623432,
      "tokens_seen": 344653824
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045229689067201605,
      "loss": 3.2604,
      "theoretical_loss": 4.080119560374918,
      "tokens_seen": 344719360
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045228686058174523,
      "loss": 3.2327,
      "theoretical_loss": 4.080030608775055,
      "tokens_seen": 344784896
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045227683049147446,
      "loss": 3.2709,
      "theoretical_loss": 4.079941678814462,
      "tokens_seen": 344850432
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004522668004012036,
      "loss": 2.9916,
      "theoretical_loss": 4.079852770483765,
      "tokens_seen": 344915968
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004522567703109328,
      "loss": 3.1725,
      "theoretical_loss": 4.079763883773593,
      "tokens_seen": 344981504
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045224674022066195,
      "loss": 3.2843,
      "theoretical_loss": 4.079675018674582,
      "tokens_seen": 345047040
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004522367101303912,
      "loss": 3.2698,
      "theoretical_loss": 4.079586175177377,
      "tokens_seen": 345112576
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045222668004012037,
      "loss": 2.7937,
      "theoretical_loss": 4.079497353272625,
      "tokens_seen": 345178112
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045221664994984955,
      "loss": 3.0016,
      "theoretical_loss": 4.07940855295098,
      "tokens_seen": 345243648
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045220661985957873,
      "loss": 2.9003,
      "theoretical_loss": 4.079319774203102,
      "tokens_seen": 345309184
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045219658976930797,
      "loss": 3.0928,
      "theoretical_loss": 4.079231017019657,
      "tokens_seen": 345374720
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004521865596790371,
      "loss": 3.3615,
      "theoretical_loss": 4.0791422813913165,
      "tokens_seen": 345440256
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045217652958876633,
      "loss": 3.0568,
      "theoretical_loss": 4.079053567308757,
      "tokens_seen": 345505792
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045216649949849546,
      "loss": 2.8698,
      "theoretical_loss": 4.078964874762663,
      "tokens_seen": 345571328
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004521564694082247,
      "loss": 3.3532,
      "theoretical_loss": 4.078876203743723,
      "tokens_seen": 345636864
    },
    {
      "epoch": 0.1,
      "objective/train/advantage_avg": 0.49371546506881714,
      "objective/train/docs_used": 207336,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.107851028442383,
      "objective/train/original_loss": 3.107851266860962,
      "objective/train/theoretical_loss": 4.0787875542426315,
      "objective/train/tokens_used": 366162400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24384531378746033,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050611138343811,
      "objective/train/weighted_lm_loss": 3.2650794982910156,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0338950157165527,
      "theoretical_loss": 4.0787875542426315,
      "tokens_seen": 345702400
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045214643931795387,
      "loss": 3.1527,
      "theoretical_loss": 4.0787875542426315,
      "tokens_seen": 345702400
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045213640922768305,
      "loss": 2.7998,
      "theoretical_loss": 4.078698926250089,
      "tokens_seen": 345767936
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045212637913741223,
      "loss": 3.1425,
      "theoretical_loss": 4.078610319756802,
      "tokens_seen": 345833472
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004521163490471414,
      "loss": 2.9752,
      "theoretical_loss": 4.078521734753482,
      "tokens_seen": 345899008
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004521063189568706,
      "loss": 3.1742,
      "theoretical_loss": 4.078433171230849,
      "tokens_seen": 345964544
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045209628886659983,
      "loss": 2.9096,
      "theoretical_loss": 4.078344629179623,
      "tokens_seen": 346030080
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045208625877632896,
      "loss": 3.3563,
      "theoretical_loss": 4.078256108590535,
      "tokens_seen": 346095616
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004520762286860582,
      "loss": 3.1417,
      "theoretical_loss": 4.078167609454322,
      "tokens_seen": 346161152
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004520661985957873,
      "loss": 3.1418,
      "theoretical_loss": 4.078079131761723,
      "tokens_seen": 346226688
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045205616850551656,
      "loss": 3.3255,
      "theoretical_loss": 4.077990675503484,
      "tokens_seen": 346292224
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.00045204613841524574,
      "loss": 3.1858,
      "theoretical_loss": 4.07790224067036,
      "tokens_seen": 346357760
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004520361083249749,
      "loss": 2.775,
      "theoretical_loss": 4.077813827253106,
      "tokens_seen": 346423296
    },
    {
      "epoch": 0.1,
      "learning_rate": 0.0004520260782347041,
      "loss": 3.0652,
      "theoretical_loss": 4.077725435242488,
      "tokens_seen": 346488832
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00045201604814443334,
      "loss": 2.8765,
      "theoretical_loss": 4.077637064629275,
      "tokens_seen": 346554368
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00045200601805416246,
      "loss": 2.9789,
      "theoretical_loss": 4.077548715404243,
      "tokens_seen": 346619904
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004519959879638917,
      "loss": 3.1644,
      "theoretical_loss": 4.077460387558171,
      "tokens_seen": 346685440
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004519859578736208,
      "loss": 3.3355,
      "theoretical_loss": 4.077372081081847,
      "tokens_seen": 346750976
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00045197592778335006,
      "loss": 3.3187,
      "theoretical_loss": 4.077283795966063,
      "tokens_seen": 346816512
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00045196589769307924,
      "loss": 3.1896,
      "theoretical_loss": 4.077195532201619,
      "tokens_seen": 346882048
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004519558676028084,
      "loss": 3.1733,
      "theoretical_loss": 4.077107289779317,
      "tokens_seen": 346947584
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00045194583751253766,
      "loss": 3.0587,
      "theoretical_loss": 4.0770190686899666,
      "tokens_seen": 347013120
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004519358074222668,
      "loss": 2.9534,
      "theoretical_loss": 4.076930868924384,
      "tokens_seen": 347078656
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.000451925777331996,
      "loss": 3.13,
      "theoretical_loss": 4.07684269047339,
      "tokens_seen": 347144192
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004519157472417252,
      "loss": 3.0977,
      "theoretical_loss": 4.07675453332781,
      "tokens_seen": 347209728
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004519057171514544,
      "loss": 3.0682,
      "theoretical_loss": 4.076666397478476,
      "tokens_seen": 347275264
    },
    {
      "epoch": 0.11,
      "objective/train/advantage_avg": 0.4924106299877167,
      "objective/train/docs_used": 208796,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.7873036861419678,
      "objective/train/original_loss": 2.787304162979126,
      "objective/train/theoretical_loss": 4.076578282916229,
      "objective/train/tokens_used": 367800800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2425767481327057,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0504740476608276,
      "objective/train/weighted_lm_loss": 2.9276061058044434,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0338950157165527,
      "theoretical_loss": 4.076578282916229,
      "tokens_seen": 347340800
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00045189568706118356,
      "loss": 3.0241,
      "theoretical_loss": 4.076578282916229,
      "tokens_seen": 347340800
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004518856569709128,
      "loss": 3.2381,
      "theoretical_loss": 4.07649018963191,
      "tokens_seen": 347406336
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004518756268806419,
      "loss": 3.2851,
      "theoretical_loss": 4.07640211761637,
      "tokens_seen": 347471872
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00045186559679037116,
      "loss": 3.2073,
      "theoretical_loss": 4.076314066860462,
      "tokens_seen": 347537408
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004518555667001003,
      "loss": 3.0534,
      "theoretical_loss": 4.076226037355047,
      "tokens_seen": 347602944
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004518455366098295,
      "loss": 2.964,
      "theoretical_loss": 4.076138029090993,
      "tokens_seen": 347668480
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004518355065195587,
      "loss": 3.1283,
      "theoretical_loss": 4.076050042059171,
      "tokens_seen": 347734016
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004518254764292879,
      "loss": 2.9722,
      "theoretical_loss": 4.075962076250458,
      "tokens_seen": 347799552
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00045181544633901707,
      "loss": 3.0587,
      "theoretical_loss": 4.075874131655738,
      "tokens_seen": 347865088
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00045180541624874625,
      "loss": 3.022,
      "theoretical_loss": 4.075786208265899,
      "tokens_seen": 347930624
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00045179538615847543,
      "loss": 3.1186,
      "theoretical_loss": 4.0756983060718355,
      "tokens_seen": 347996160
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00045178535606820466,
      "loss": 3.1827,
      "theoretical_loss": 4.075610425064449,
      "tokens_seen": 348061696
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004517753259779338,
      "loss": 3.3346,
      "theoretical_loss": 4.075522565234643,
      "tokens_seen": 348127232
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.000451765295887663,
      "loss": 3.0116,
      "theoretical_loss": 4.075434726573331,
      "tokens_seen": 348192768
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00045175526579739215,
      "loss": 3.0024,
      "theoretical_loss": 4.0753469090714285,
      "tokens_seen": 348258304
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004517452357071214,
      "loss": 2.935,
      "theoretical_loss": 4.075259112719859,
      "tokens_seen": 348323840
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00045173520561685057,
      "loss": 3.17,
      "theoretical_loss": 4.075171337509549,
      "tokens_seen": 348389376
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00045172517552657975,
      "loss": 2.9154,
      "theoretical_loss": 4.075083583431434,
      "tokens_seen": 348454912
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00045171514543630893,
      "loss": 2.9739,
      "theoretical_loss": 4.074995850476453,
      "tokens_seen": 348520448
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00045170511534603817,
      "loss": 3.131,
      "theoretical_loss": 4.074908138635551,
      "tokens_seen": 348585984
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004516950852557673,
      "loss": 3.0711,
      "theoretical_loss": 4.074820447899677,
      "tokens_seen": 348651520
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00045168505516549653,
      "loss": 2.9847,
      "theoretical_loss": 4.074732778259788,
      "tokens_seen": 348717056
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00045167502507522566,
      "loss": 3.2408,
      "theoretical_loss": 4.0746451297068464,
      "tokens_seen": 348782592
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004516649949849549,
      "loss": 3.2098,
      "theoretical_loss": 4.074557502231819,
      "tokens_seen": 348848128
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00045165496489468407,
      "loss": 3.1263,
      "theoretical_loss": 4.074469895825678,
      "tokens_seen": 348913664
    },
    {
      "epoch": 0.11,
      "objective/train/advantage_avg": 0.4918447732925415,
      "objective/train/docs_used": 209597,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.5819830894470215,
      "objective/train/original_loss": 2.5819830894470215,
      "objective/train/theoretical_loss": 4.074382310479402,
      "objective/train/tokens_used": 369439200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24205341935157776,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0504149198532104,
      "objective/train/weighted_lm_loss": 2.711848020553589,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0253151655197144,
      "theoretical_loss": 4.074382310479402,
      "tokens_seen": 348979200
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00045164493480441325,
      "loss": 2.7518,
      "theoretical_loss": 4.074382310479402,
      "tokens_seen": 348979200
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00045163490471414244,
      "loss": 3.051,
      "theoretical_loss": 4.074294746183976,
      "tokens_seen": 349044736
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004516248746238716,
      "loss": 3.1905,
      "theoretical_loss": 4.0742072029303875,
      "tokens_seen": 349110272
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004516148445336008,
      "loss": 3.1429,
      "theoretical_loss": 4.074119680709633,
      "tokens_seen": 349175808
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00045160481444333003,
      "loss": 3.2062,
      "theoretical_loss": 4.074032179512713,
      "tokens_seen": 349241344
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00045159478435305916,
      "loss": 3.0645,
      "theoretical_loss": 4.073944699330632,
      "tokens_seen": 349306880
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004515847542627884,
      "loss": 3.1375,
      "theoretical_loss": 4.073857240154403,
      "tokens_seen": 349372416
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004515747241725175,
      "loss": 3.0605,
      "theoretical_loss": 4.073769801975043,
      "tokens_seen": 349437952
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00045156469408224676,
      "loss": 3.1633,
      "theoretical_loss": 4.0736823847835755,
      "tokens_seen": 349503488
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00045155466399197594,
      "loss": 2.9865,
      "theoretical_loss": 4.073594988571027,
      "tokens_seen": 349569024
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004515446339017051,
      "loss": 3.0746,
      "theoretical_loss": 4.073507613328433,
      "tokens_seen": 349634560
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004515346038114343,
      "loss": 3.2476,
      "theoretical_loss": 4.07342025904683,
      "tokens_seen": 349700096
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00045152457372116354,
      "loss": 3.0982,
      "theoretical_loss": 4.073332925717265,
      "tokens_seen": 349765632
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00045151454363089266,
      "loss": 2.8938,
      "theoretical_loss": 4.073245613330788,
      "tokens_seen": 349831168
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004515045135406219,
      "loss": 2.9708,
      "theoretical_loss": 4.073158321878454,
      "tokens_seen": 349896704
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.000451494483450351,
      "loss": 3.0703,
      "theoretical_loss": 4.073071051351325,
      "tokens_seen": 349962240
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00045148445336008026,
      "loss": 3.0821,
      "theoretical_loss": 4.072983801740467,
      "tokens_seen": 350027776
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00045147442326980944,
      "loss": 3.2884,
      "theoretical_loss": 4.072896573036953,
      "tokens_seen": 350093312
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004514643931795386,
      "loss": 2.9824,
      "theoretical_loss": 4.072809365231859,
      "tokens_seen": 350158848
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004514543630892678,
      "loss": 3.1125,
      "theoretical_loss": 4.072722178316271,
      "tokens_seen": 350224384
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.000451444332998997,
      "loss": 2.9644,
      "theoretical_loss": 4.072635012281275,
      "tokens_seen": 350289920
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00045143430290872617,
      "loss": 2.9885,
      "theoretical_loss": 4.072547867117966,
      "tokens_seen": 350355456
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004514242728184554,
      "loss": 3.0825,
      "theoretical_loss": 4.072460742817444,
      "tokens_seen": 350420992
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00045141424272818453,
      "loss": 3.0779,
      "theoretical_loss": 4.072373639370814,
      "tokens_seen": 350486528
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00045140421263791376,
      "loss": 2.9119,
      "theoretical_loss": 4.0722865567691855,
      "tokens_seen": 350552064
    },
    {
      "epoch": 0.11,
      "objective/train/advantage_avg": 0.4937944710254669,
      "objective/train/docs_used": 210243,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.644829273223877,
      "objective/train/original_loss": 2.644829511642456,
      "objective/train/theoretical_loss": 4.072199495003675,
      "objective/train/tokens_used": 371077600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2439092993736267,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.05061936378479,
      "objective/train/weighted_lm_loss": 2.7789437770843506,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.047261118888855,
      "theoretical_loss": 4.072199495003675,
      "tokens_seen": 350617600
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00045139418254764294,
      "loss": 2.8746,
      "theoretical_loss": 4.072199495003675,
      "tokens_seen": 350617600
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004513841524573721,
      "loss": 3.0918,
      "theoretical_loss": 4.072112454065405,
      "tokens_seen": 350683136
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004513741223671013,
      "loss": 2.9621,
      "theoretical_loss": 4.0720254339455,
      "tokens_seen": 350748672
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004513640922768305,
      "loss": 2.8589,
      "theoretical_loss": 4.0719384346350935,
      "tokens_seen": 350814208
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00045135406218655967,
      "loss": 2.7925,
      "theoretical_loss": 4.071851456125324,
      "tokens_seen": 350879744
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004513440320962889,
      "loss": 3.2245,
      "theoretical_loss": 4.071764498407333,
      "tokens_seen": 350945280
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00045133400200601803,
      "loss": 3.1704,
      "theoretical_loss": 4.071677561472271,
      "tokens_seen": 351010816
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00045132397191574727,
      "loss": 3.1485,
      "theoretical_loss": 4.071590645311289,
      "tokens_seen": 351076352
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004513139418254764,
      "loss": 3.2833,
      "theoretical_loss": 4.07150374991555,
      "tokens_seen": 351141888
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00045130391173520563,
      "loss": 2.8283,
      "theoretical_loss": 4.071416875276216,
      "tokens_seen": 351207424
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004512938816449348,
      "loss": 2.9385,
      "theoretical_loss": 4.071330021384458,
      "tokens_seen": 351272960
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.000451283851554664,
      "loss": 3.1822,
      "theoretical_loss": 4.071243188231452,
      "tokens_seen": 351338496
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00045127382146439317,
      "loss": 3.075,
      "theoretical_loss": 4.071156375808378,
      "tokens_seen": 351404032
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00045126379137412235,
      "loss": 3.0462,
      "theoretical_loss": 4.071069584106424,
      "tokens_seen": 351469568
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00045125376128385153,
      "loss": 3.1695,
      "theoretical_loss": 4.070982813116781,
      "tokens_seen": 351535104
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00045124373119358077,
      "loss": 2.9864,
      "theoretical_loss": 4.070896062830645,
      "tokens_seen": 351600640
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004512337011033099,
      "loss": 2.9488,
      "theoretical_loss": 4.070809333239221,
      "tokens_seen": 351666176
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00045122367101303913,
      "loss": 3.0208,
      "theoretical_loss": 4.0707226243337145,
      "tokens_seen": 351731712
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004512136409227683,
      "loss": 3.1064,
      "theoretical_loss": 4.070635936105339,
      "tokens_seen": 351797248
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004512036108324975,
      "loss": 3.0029,
      "theoretical_loss": 4.0705492685453155,
      "tokens_seen": 351862784
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00045119358074222673,
      "loss": 3.2276,
      "theoretical_loss": 4.070462621644865,
      "tokens_seen": 351928320
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00045118355065195586,
      "loss": 2.9146,
      "theoretical_loss": 4.07037599539522,
      "tokens_seen": 351993856
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004511735205616851,
      "loss": 3.3412,
      "theoretical_loss": 4.070289389787613,
      "tokens_seen": 352059392
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004511634904714143,
      "loss": 3.3417,
      "theoretical_loss": 4.070202804813284,
      "tokens_seen": 352124928
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00045115346038114345,
      "loss": 3.091,
      "theoretical_loss": 4.070116240463479,
      "tokens_seen": 352190464
    },
    {
      "epoch": 0.11,
      "objective/train/advantage_avg": 0.49390873312950134,
      "objective/train/docs_used": 211560,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.1473708152770996,
      "objective/train/original_loss": 3.1473705768585205,
      "objective/train/theoretical_loss": 4.070029696729449,
      "objective/train/tokens_used": 372716000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.244071826338768,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506316423416138,
      "objective/train/weighted_lm_loss": 3.3067872524261475,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0408108234405518,
      "theoretical_loss": 4.070029696729449,
      "tokens_seen": 352256000
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00045114343029087264,
      "loss": 3.0553,
      "theoretical_loss": 4.070029696729449,
      "tokens_seen": 352256000
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004511334002006018,
      "loss": 3.0346,
      "theoretical_loss": 4.069943173602451,
      "tokens_seen": 352321536
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.000451123370110331,
      "loss": 3.2639,
      "theoretical_loss": 4.069856671073745,
      "tokens_seen": 352387072
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00045111334002006023,
      "loss": 2.8952,
      "theoretical_loss": 4.069770189134598,
      "tokens_seen": 352452608
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00045110330992978936,
      "loss": 3.1458,
      "theoretical_loss": 4.069683727776281,
      "tokens_seen": 352518144
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004510932798395186,
      "loss": 3.0688,
      "theoretical_loss": 4.069597286990075,
      "tokens_seen": 352583680
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004510832497492477,
      "loss": 2.8657,
      "theoretical_loss": 4.06951086676726,
      "tokens_seen": 352649216
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00045107321965897696,
      "loss": 2.9886,
      "theoretical_loss": 4.0694244670991235,
      "tokens_seen": 352714752
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00045106318956870614,
      "loss": 2.8924,
      "theoretical_loss": 4.069338087976961,
      "tokens_seen": 352780288
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004510531594784353,
      "loss": 3.1106,
      "theoretical_loss": 4.06925172939207,
      "tokens_seen": 352845824
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004510431293881645,
      "loss": 3.1364,
      "theoretical_loss": 4.069165391335753,
      "tokens_seen": 352911360
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00045103309929789374,
      "loss": 3.1691,
      "theoretical_loss": 4.0690790737993225,
      "tokens_seen": 352976896
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00045102306920762286,
      "loss": 3.0214,
      "theoretical_loss": 4.068992776774091,
      "tokens_seen": 353042432
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004510130391173521,
      "loss": 2.9989,
      "theoretical_loss": 4.068906500251378,
      "tokens_seen": 353107968
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004510030090270812,
      "loss": 3.2136,
      "theoretical_loss": 4.068820244222511,
      "tokens_seen": 353173504
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00045099297893681046,
      "loss": 3.0895,
      "theoretical_loss": 4.068734008678818,
      "tokens_seen": 353239040
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00045098294884653964,
      "loss": 3.1474,
      "theoretical_loss": 4.068647793611635,
      "tokens_seen": 353304576
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004509729187562688,
      "loss": 3.0729,
      "theoretical_loss": 4.068561599012304,
      "tokens_seen": 353370112
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.000450962888665998,
      "loss": 3.1904,
      "theoretical_loss": 4.068475424872171,
      "tokens_seen": 353435648
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004509528585757272,
      "loss": 2.8911,
      "theoretical_loss": 4.068389271182586,
      "tokens_seen": 353501184
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00045094282848545637,
      "loss": 3.1431,
      "theoretical_loss": 4.068303137934908,
      "tokens_seen": 353566720
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004509327983951856,
      "loss": 3.3511,
      "theoretical_loss": 4.0682170251204965,
      "tokens_seen": 353632256
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00045092276830491473,
      "loss": 3.0348,
      "theoretical_loss": 4.068130932730722,
      "tokens_seen": 353697792
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00045091273821464396,
      "loss": 2.9882,
      "theoretical_loss": 4.068044860756954,
      "tokens_seen": 353763328
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00045090270812437314,
      "loss": 3.1666,
      "theoretical_loss": 4.067958809190571,
      "tokens_seen": 353828864
    },
    {
      "epoch": 0.11,
      "objective/train/advantage_avg": 0.4944409728050232,
      "objective/train/docs_used": 212189,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.642113208770752,
      "objective/train/original_loss": 2.642112970352173,
      "objective/train/theoretical_loss": 4.0678727780229575,
      "objective/train/tokens_used": 374354400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24453206360340118,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506870746612549,
      "objective/train/weighted_lm_loss": 2.7759082317352295,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0413066148757935,
      "theoretical_loss": 4.0678727780229575,
      "tokens_seen": 353894400
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004508926780341023,
      "loss": 3.069,
      "theoretical_loss": 4.0678727780229575,
      "tokens_seen": 353894400
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004508826479438315,
      "loss": 3.3363,
      "theoretical_loss": 4.0677867672455,
      "tokens_seen": 353959936
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004508726178535607,
      "loss": 3.0128,
      "theoretical_loss": 4.067700776849592,
      "tokens_seen": 354025472
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00045086258776328987,
      "loss": 3.1482,
      "theoretical_loss": 4.067614806826633,
      "tokens_seen": 354091008
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004508525576730191,
      "loss": 2.9143,
      "theoretical_loss": 4.067528857168027,
      "tokens_seen": 354156544
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00045084252758274823,
      "loss": 3.0543,
      "theoretical_loss": 4.067442927865182,
      "tokens_seen": 354222080
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00045083249749247747,
      "loss": 2.8932,
      "theoretical_loss": 4.067357018909513,
      "tokens_seen": 354287616
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004508224674022066,
      "loss": 3.0629,
      "theoretical_loss": 4.06727113029244,
      "tokens_seen": 354353152
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00045081243731193583,
      "loss": 3.1787,
      "theoretical_loss": 4.0671852620053865,
      "tokens_seen": 354418688
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.000450802407221665,
      "loss": 2.7976,
      "theoretical_loss": 4.067099414039784,
      "tokens_seen": 354484224
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004507923771313942,
      "loss": 3.2697,
      "theoretical_loss": 4.067013586387065,
      "tokens_seen": 354549760
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00045078234704112337,
      "loss": 3.1622,
      "theoretical_loss": 4.066927779038673,
      "tokens_seen": 354615296
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00045077231695085255,
      "loss": 2.874,
      "theoretical_loss": 4.066841991986053,
      "tokens_seen": 354680832
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00045076228686058173,
      "loss": 3.0522,
      "theoretical_loss": 4.066756225220653,
      "tokens_seen": 354746368
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00045075225677031097,
      "loss": 3.0527,
      "theoretical_loss": 4.066670478733931,
      "tokens_seen": 354811904
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004507422266800401,
      "loss": 2.908,
      "theoretical_loss": 4.066584752517348,
      "tokens_seen": 354877440
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00045073219658976933,
      "loss": 3.0376,
      "theoretical_loss": 4.066499046562369,
      "tokens_seen": 354942976
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004507221664994985,
      "loss": 3.1262,
      "theoretical_loss": 4.066413360860468,
      "tokens_seen": 355008512
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004507121364092277,
      "loss": 3.1719,
      "theoretical_loss": 4.066327695403118,
      "tokens_seen": 355074048
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004507021063189569,
      "loss": 3.198,
      "theoretical_loss": 4.066242050181803,
      "tokens_seen": 355139584
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00045069207622868606,
      "loss": 3.2162,
      "theoretical_loss": 4.06615642518801,
      "tokens_seen": 355205120
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00045068204613841524,
      "loss": 3.0582,
      "theoretical_loss": 4.066070820413229,
      "tokens_seen": 355270656
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004506720160481445,
      "loss": 3.1384,
      "theoretical_loss": 4.065985235848959,
      "tokens_seen": 355336192
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004506619859578736,
      "loss": 3.0985,
      "theoretical_loss": 4.065899671486703,
      "tokens_seen": 355401728
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00045065195586760284,
      "loss": 3.037,
      "theoretical_loss": 4.0658141273179655,
      "tokens_seen": 355467264
    },
    {
      "epoch": 0.11,
      "objective/train/advantage_avg": 0.49388593435287476,
      "objective/train/docs_used": 213521,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.4977986812591553,
      "objective/train/original_loss": 2.497798204421997,
      "objective/train/theoretical_loss": 4.065728603334263,
      "objective/train/tokens_used": 375992800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24399062991142273,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050628900527954,
      "objective/train/weighted_lm_loss": 2.624354600906372,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0425469875335693,
      "theoretical_loss": 4.065728603334263,
      "tokens_seen": 355532800
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00045064192577733196,
      "loss": 2.9544,
      "theoretical_loss": 4.065728603334263,
      "tokens_seen": 355532800
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004506318956870612,
      "loss": 3.0897,
      "theoretical_loss": 4.065643099527109,
      "tokens_seen": 355598336
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004506218655967904,
      "loss": 3.2276,
      "theoretical_loss": 4.065557615888029,
      "tokens_seen": 355663872
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00045061183550651956,
      "loss": 2.972,
      "theoretical_loss": 4.06547215240855,
      "tokens_seen": 355729408
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00045060180541624874,
      "loss": 3.0475,
      "theoretical_loss": 4.065386709080206,
      "tokens_seen": 355794944
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004505917753259779,
      "loss": 3.3247,
      "theoretical_loss": 4.0653012858945345,
      "tokens_seen": 355860480
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004505817452357071,
      "loss": 3.0889,
      "theoretical_loss": 4.065215882843079,
      "tokens_seen": 355926016
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00045057171514543634,
      "loss": 3.4037,
      "theoretical_loss": 4.065130499917388,
      "tokens_seen": 355991552
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00045056168505516547,
      "loss": 3.3283,
      "theoretical_loss": 4.065045137109014,
      "tokens_seen": 356057088
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004505516549648947,
      "loss": 3.2851,
      "theoretical_loss": 4.064959794409518,
      "tokens_seen": 356122624
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004505416248746239,
      "loss": 2.8228,
      "theoretical_loss": 4.064874471810461,
      "tokens_seen": 356188160
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00045053159478435306,
      "loss": 3.0073,
      "theoretical_loss": 4.064789169303412,
      "tokens_seen": 356253696
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00045052156469408224,
      "loss": 3.1391,
      "theoretical_loss": 4.064703886879947,
      "tokens_seen": 356319232
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004505115346038114,
      "loss": 2.9671,
      "theoretical_loss": 4.064618624531644,
      "tokens_seen": 356384768
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004505015045135406,
      "loss": 3.1585,
      "theoretical_loss": 4.064533382250086,
      "tokens_seen": 356450304
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00045049147442326984,
      "loss": 3.1384,
      "theoretical_loss": 4.0644481600268625,
      "tokens_seen": 356515840
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00045048144433299897,
      "loss": 2.9681,
      "theoretical_loss": 4.064362957853569,
      "tokens_seen": 356581376
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004504714142427282,
      "loss": 3.3102,
      "theoretical_loss": 4.064277775721804,
      "tokens_seen": 356646912
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00045046138415245733,
      "loss": 3.1608,
      "theoretical_loss": 4.06419261362317,
      "tokens_seen": 356712448
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00045045135406218657,
      "loss": 3.1226,
      "theoretical_loss": 4.0641074715492795,
      "tokens_seen": 356777984
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004504413239719158,
      "loss": 2.9666,
      "theoretical_loss": 4.064022349491745,
      "tokens_seen": 356843520
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00045043129388164493,
      "loss": 2.8646,
      "theoretical_loss": 4.063937247442187,
      "tokens_seen": 356909056
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00045042126379137416,
      "loss": 3.1796,
      "theoretical_loss": 4.0638521653922295,
      "tokens_seen": 356974592
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00045041123370110334,
      "loss": 3.1634,
      "theoretical_loss": 4.063767103333502,
      "tokens_seen": 357040128
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004504012036108325,
      "loss": 3.3565,
      "theoretical_loss": 4.063682061257638,
      "tokens_seen": 357105664
    },
    {
      "epoch": 0.11,
      "objective/train/advantage_avg": 0.4949467182159424,
      "objective/train/docs_used": 214188,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.132028579711914,
      "objective/train/original_loss": 3.132028579711914,
      "objective/train/theoretical_loss": 4.06359703915628,
      "objective/train/tokens_used": 377631200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.245050311088562,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507404804229736,
      "objective/train/weighted_lm_loss": 3.290809392929077,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0417574644088745,
      "theoretical_loss": 4.06359703915628,
      "tokens_seen": 357171200
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004503911735205617,
      "loss": 3.05,
      "theoretical_loss": 4.06359703915628,
      "tokens_seen": 357171200
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004503811434302909,
      "loss": 3.2352,
      "theoretical_loss": 4.06351203702107,
      "tokens_seen": 357236736
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00045037111334002007,
      "loss": 3.1569,
      "theoretical_loss": 4.063427054843658,
      "tokens_seen": 357302272
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004503610832497493,
      "loss": 2.8131,
      "theoretical_loss": 4.0633420926157005,
      "tokens_seen": 357367808
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00045035105315947843,
      "loss": 2.9424,
      "theoretical_loss": 4.063257150328856,
      "tokens_seen": 357433344
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00045034102306920767,
      "loss": 3.1369,
      "theoretical_loss": 4.063172227974789,
      "tokens_seen": 357498880
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004503309929789368,
      "loss": 3.2599,
      "theoretical_loss": 4.06308732554517,
      "tokens_seen": 357564416
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00045032096288866603,
      "loss": 2.7792,
      "theoretical_loss": 4.063002443031674,
      "tokens_seen": 357629952
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004503109327983952,
      "loss": 3.357,
      "theoretical_loss": 4.062917580425979,
      "tokens_seen": 357695488
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004503009027081244,
      "loss": 3.088,
      "theoretical_loss": 4.062832737719772,
      "tokens_seen": 357761024
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00045029087261785357,
      "loss": 3.0131,
      "theoretical_loss": 4.062747914904741,
      "tokens_seen": 357826560
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00045028084252758275,
      "loss": 3.2038,
      "theoretical_loss": 4.062663111972582,
      "tokens_seen": 357892096
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00045027081243731193,
      "loss": 3.1027,
      "theoretical_loss": 4.062578328914994,
      "tokens_seen": 357957632
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00045026078234704117,
      "loss": 2.9872,
      "theoretical_loss": 4.062493565723683,
      "tokens_seen": 358023168
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004502507522567703,
      "loss": 2.9338,
      "theoretical_loss": 4.062408822390356,
      "tokens_seen": 358088704
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00045024072216649953,
      "loss": 3.0452,
      "theoretical_loss": 4.062324098906732,
      "tokens_seen": 358154240
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004502306920762287,
      "loss": 3.1721,
      "theoretical_loss": 4.062239395264526,
      "tokens_seen": 358219776
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004502206619859579,
      "loss": 3.2708,
      "theoretical_loss": 4.062154711455466,
      "tokens_seen": 358285312
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004502106318956871,
      "loss": 2.9784,
      "theoretical_loss": 4.06207004747128,
      "tokens_seen": 358350848
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00045020060180541626,
      "loss": 3.0647,
      "theoretical_loss": 4.061985403303702,
      "tokens_seen": 358416384
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00045019057171514544,
      "loss": 3.3044,
      "theoretical_loss": 4.061900778944473,
      "tokens_seen": 358481920
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004501805416248747,
      "loss": 3.2564,
      "theoretical_loss": 4.061816174385338,
      "tokens_seen": 358547456
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004501705115346038,
      "loss": 3.1902,
      "theoretical_loss": 4.061731589618044,
      "tokens_seen": 358612992
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00045016048144433304,
      "loss": 3.1048,
      "theoretical_loss": 4.061647024634347,
      "tokens_seen": 358678528
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00045015045135406216,
      "loss": 3.0262,
      "theoretical_loss": 4.061562479426007,
      "tokens_seen": 358744064
    },
    {
      "epoch": 0.11,
      "objective/train/advantage_avg": 0.492588073015213,
      "objective/train/docs_used": 215284,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.940509557723999,
      "objective/train/original_loss": 2.940509557723999,
      "objective/train/theoretical_loss": 4.061477953984786,
      "objective/train/tokens_used": 379269600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2427314668893814,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050492525100708,
      "objective/train/weighted_lm_loss": 3.0891168117523193,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0396549701690674,
      "theoretical_loss": 4.061477953984786,
      "tokens_seen": 358809600
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004501404212637914,
      "loss": 3.0785,
      "theoretical_loss": 4.061477953984786,
      "tokens_seen": 358809600
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004501303911735206,
      "loss": 3.1629,
      "theoretical_loss": 4.0613934483024545,
      "tokens_seen": 358875136
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00045012036108324976,
      "loss": 3.2103,
      "theoretical_loss": 4.061308962370787,
      "tokens_seen": 358940672
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00045011033099297894,
      "loss": 3.1049,
      "theoretical_loss": 4.061224496181562,
      "tokens_seen": 359006208
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004501003009027081,
      "loss": 3.0247,
      "theoretical_loss": 4.061140049726563,
      "tokens_seen": 359071744
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004500902708124373,
      "loss": 2.8361,
      "theoretical_loss": 4.06105562299758,
      "tokens_seen": 359137280
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00045008024072216654,
      "loss": 3.0787,
      "theoretical_loss": 4.060971215986406,
      "tokens_seen": 359202816
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00045007021063189567,
      "loss": 3.1773,
      "theoretical_loss": 4.06088682868484,
      "tokens_seen": 359268352
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004500601805416249,
      "loss": 3.2137,
      "theoretical_loss": 4.060802461084685,
      "tokens_seen": 359333888
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004500501504513541,
      "loss": 2.9826,
      "theoretical_loss": 4.06071811317775,
      "tokens_seen": 359399424
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00045004012036108326,
      "loss": 2.9534,
      "theoretical_loss": 4.060633784955849,
      "tokens_seen": 359464960
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00045003009027081244,
      "loss": 2.9578,
      "theoretical_loss": 4.060549476410799,
      "tokens_seen": 359530496
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004500200601805416,
      "loss": 3.0484,
      "theoretical_loss": 4.060465187534424,
      "tokens_seen": 359596032
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004500100300902708,
      "loss": 3.1404,
      "theoretical_loss": 4.060380918318552,
      "tokens_seen": 359661568
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00045000000000000004,
      "loss": 2.9234,
      "theoretical_loss": 4.060296668755016,
      "tokens_seen": 359727104
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044998996990972917,
      "loss": 3.0429,
      "theoretical_loss": 4.060212438835654,
      "tokens_seen": 359792640
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004499799398194584,
      "loss": 3.2781,
      "theoretical_loss": 4.0601282285523075,
      "tokens_seen": 359858176
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044996990972918753,
      "loss": 2.8392,
      "theoretical_loss": 4.060044037896826,
      "tokens_seen": 359923712
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044995987963891677,
      "loss": 3.2123,
      "theoretical_loss": 4.05995986686106,
      "tokens_seen": 359989248
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044994984954864595,
      "loss": 3.125,
      "theoretical_loss": 4.059875715436869,
      "tokens_seen": 360054784
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044993981945837513,
      "loss": 3.0193,
      "theoretical_loss": 4.0597915836161125,
      "tokens_seen": 360120320
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004499297893681043,
      "loss": 3.0537,
      "theoretical_loss": 4.059707471390659,
      "tokens_seen": 360185856
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044991975927783355,
      "loss": 3.2633,
      "theoretical_loss": 4.0596233787523825,
      "tokens_seen": 360251392
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044990972918756267,
      "loss": 3.3942,
      "theoretical_loss": 4.059539305693156,
      "tokens_seen": 360316928
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004498996990972919,
      "loss": 3.07,
      "theoretical_loss": 4.059455252204863,
      "tokens_seen": 360382464
    },
    {
      "epoch": 0.11,
      "objective/train/advantage_avg": 0.49309682846069336,
      "objective/train/docs_used": 215704,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.5583856105804443,
      "objective/train/original_loss": 3.5583858489990234,
      "objective/train/theoretical_loss": 4.05937121827939,
      "objective/train/tokens_used": 380908000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24325811862945557,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505461692810059,
      "objective/train/weighted_lm_loss": 3.7383341789245605,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0437887907028198,
      "theoretical_loss": 4.05937121827939,
      "tokens_seen": 360448000
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044988966900702103,
      "loss": 3.1955,
      "theoretical_loss": 4.05937121827939,
      "tokens_seen": 360448000
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044987963891675027,
      "loss": 3.2519,
      "theoretical_loss": 4.059287203908627,
      "tokens_seen": 360513536
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044986960882647945,
      "loss": 3.1312,
      "theoretical_loss": 4.0592032090844725,
      "tokens_seen": 360579072
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044985957873620863,
      "loss": 3.2457,
      "theoretical_loss": 4.059119233798825,
      "tokens_seen": 360644608
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004498495486459378,
      "loss": 2.9758,
      "theoretical_loss": 4.059035278043591,
      "tokens_seen": 360710144
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.000449839518555667,
      "loss": 3.2251,
      "theoretical_loss": 4.0589513418106815,
      "tokens_seen": 360775680
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004498294884653962,
      "loss": 3.2685,
      "theoretical_loss": 4.0588674250920125,
      "tokens_seen": 360841216
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004498194583751254,
      "loss": 3.031,
      "theoretical_loss": 4.058783527879502,
      "tokens_seen": 360906752
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044980942828485454,
      "loss": 3.304,
      "theoretical_loss": 4.058699650165078,
      "tokens_seen": 360972288
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044979939819458377,
      "loss": 3.042,
      "theoretical_loss": 4.058615791940667,
      "tokens_seen": 361037824
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004497893681043129,
      "loss": 3.1397,
      "theoretical_loss": 4.058531953198207,
      "tokens_seen": 361103360
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044977933801404213,
      "loss": 3.1697,
      "theoretical_loss": 4.058448133929635,
      "tokens_seen": 361168896
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004497693079237713,
      "loss": 3.0028,
      "theoretical_loss": 4.058364334126896,
      "tokens_seen": 361234432
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004497592778335005,
      "loss": 3.2926,
      "theoretical_loss": 4.058280553781939,
      "tokens_seen": 361299968
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004497492477432297,
      "loss": 3.2364,
      "theoretical_loss": 4.058196792886719,
      "tokens_seen": 361365504
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004497392176529589,
      "loss": 3.176,
      "theoretical_loss": 4.0581130514331925,
      "tokens_seen": 361431040
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044972918756268804,
      "loss": 3.1162,
      "theoretical_loss": 4.0580293294133245,
      "tokens_seen": 361496576
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004497191574724173,
      "loss": 3.0171,
      "theoretical_loss": 4.057945626819082,
      "tokens_seen": 361562112
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004497091273821464,
      "loss": 3.0273,
      "theoretical_loss": 4.05786194364244,
      "tokens_seen": 361627648
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044969909729187564,
      "loss": 3.0507,
      "theoretical_loss": 4.057778279875373,
      "tokens_seen": 361693184
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004496890672016049,
      "loss": 3.0262,
      "theoretical_loss": 4.057694635509866,
      "tokens_seen": 361758720
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.000449679037111334,
      "loss": 2.9146,
      "theoretical_loss": 4.057611010537905,
      "tokens_seen": 361824256
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044966900702106324,
      "loss": 3.2268,
      "theoretical_loss": 4.057527404951483,
      "tokens_seen": 361889792
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044965897693079236,
      "loss": 3.001,
      "theoretical_loss": 4.0574438187425965,
      "tokens_seen": 361955328
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004496489468405216,
      "loss": 3.1093,
      "theoretical_loss": 4.057360251903246,
      "tokens_seen": 362020864
    },
    {
      "epoch": 0.11,
      "objective/train/advantage_avg": 0.49268943071365356,
      "objective/train/docs_used": 216313,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.1062140464782715,
      "objective/train/original_loss": 3.1062142848968506,
      "objective/train/theoretical_loss": 4.057276704425439,
      "objective/train/tokens_used": 382546400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24284830689430237,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505034923553467,
      "objective/train/weighted_lm_loss": 3.263183832168579,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0428944826126099,
      "theoretical_loss": 4.057276704425439,
      "tokens_seen": 362086400
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004496389167502508,
      "loss": 3.0395,
      "theoretical_loss": 4.057276704425439,
      "tokens_seen": 362086400
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044962888665997996,
      "loss": 3.2342,
      "theoretical_loss": 4.057193176301185,
      "tokens_seen": 362151936
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044961885656970914,
      "loss": 3.1103,
      "theoretical_loss": 4.057109667522502,
      "tokens_seen": 362217472
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004496088264794383,
      "loss": 3.0291,
      "theoretical_loss": 4.057026178081408,
      "tokens_seen": 362283008
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004495987963891675,
      "loss": 2.9503,
      "theoretical_loss": 4.056942707969931,
      "tokens_seen": 362348544
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044958876629889674,
      "loss": 3.1083,
      "theoretical_loss": 4.056859257180098,
      "tokens_seen": 362414080
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044957873620862587,
      "loss": 3.0647,
      "theoretical_loss": 4.056775825703945,
      "tokens_seen": 362479616
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004495687061183551,
      "loss": 3.3605,
      "theoretical_loss": 4.056692413533512,
      "tokens_seen": 362545152
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004495586760280843,
      "loss": 3.4695,
      "theoretical_loss": 4.056609020660843,
      "tokens_seen": 362610688
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044954864593781346,
      "loss": 2.9266,
      "theoretical_loss": 4.0565256470779865,
      "tokens_seen": 362676224
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044953861584754264,
      "loss": 3.2974,
      "theoretical_loss": 4.056442292776994,
      "tokens_seen": 362741760
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004495285857572718,
      "loss": 3.0193,
      "theoretical_loss": 4.056358957749928,
      "tokens_seen": 362807296
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.000449518555667001,
      "loss": 2.9968,
      "theoretical_loss": 4.056275641988848,
      "tokens_seen": 362872832
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044950852557673024,
      "loss": 3.0395,
      "theoretical_loss": 4.056192345485823,
      "tokens_seen": 362938368
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044949849548645937,
      "loss": 3.1355,
      "theoretical_loss": 4.056109068232925,
      "tokens_seen": 363003904
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004494884653961886,
      "loss": 3.164,
      "theoretical_loss": 4.056025810222232,
      "tokens_seen": 363069440
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044947843530591773,
      "loss": 3.1857,
      "theoretical_loss": 4.055942571445824,
      "tokens_seen": 363134976
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044946840521564697,
      "loss": 3.2006,
      "theoretical_loss": 4.055859351895789,
      "tokens_seen": 363200512
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044945837512537615,
      "loss": 3.5483,
      "theoretical_loss": 4.055776151564217,
      "tokens_seen": 363266048
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044944834503510533,
      "loss": 3.0426,
      "theoretical_loss": 4.055692970443205,
      "tokens_seen": 363331584
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004494383149448345,
      "loss": 3.2582,
      "theoretical_loss": 4.055609808524853,
      "tokens_seen": 363397120
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044942828485456375,
      "loss": 3.2752,
      "theoretical_loss": 4.055526665801265,
      "tokens_seen": 363462656
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044941825476429287,
      "loss": 3.3584,
      "theoretical_loss": 4.055443542264552,
      "tokens_seen": 363528192
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004494082246740221,
      "loss": 3.2317,
      "theoretical_loss": 4.0553604379068275,
      "tokens_seen": 363593728
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044939819458375123,
      "loss": 3.1451,
      "theoretical_loss": 4.055277352720212,
      "tokens_seen": 363659264
    },
    {
      "epoch": 0.11,
      "objective/train/advantage_avg": 0.4946575462818146,
      "objective/train/docs_used": 217479,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.014084577560425,
      "objective/train/original_loss": 3.0140843391418457,
      "objective/train/theoretical_loss": 4.055194286696828,
      "objective/train/tokens_used": 384184800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24475136399269104,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050709843635559,
      "objective/train/weighted_lm_loss": 3.1669459342956543,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0338950157165527,
      "theoretical_loss": 4.055194286696828,
      "tokens_seen": 363724800
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044938816449348047,
      "loss": 3.0393,
      "theoretical_loss": 4.055194286696828,
      "tokens_seen": 363724800
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044937813440320965,
      "loss": 3.0401,
      "theoretical_loss": 4.055111239828806,
      "tokens_seen": 363790336
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044936810431293883,
      "loss": 3.0308,
      "theoretical_loss": 4.055028212108276,
      "tokens_seen": 363855872
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.000449358074222668,
      "loss": 3.1695,
      "theoretical_loss": 4.054945203527378,
      "tokens_seen": 363921408
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004493480441323972,
      "loss": 3.1716,
      "theoretical_loss": 4.054862214078254,
      "tokens_seen": 363986944
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004493380140421264,
      "loss": 3.113,
      "theoretical_loss": 4.0547792437530505,
      "tokens_seen": 364052480
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004493279839518556,
      "loss": 3.0238,
      "theoretical_loss": 4.05469629254392,
      "tokens_seen": 364118016
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044931795386158474,
      "loss": 3.2219,
      "theoretical_loss": 4.0546133604430175,
      "tokens_seen": 364183552
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044930792377131397,
      "loss": 3.2178,
      "theoretical_loss": 4.054530447442505,
      "tokens_seen": 364249088
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004492978936810431,
      "loss": 3.3459,
      "theoretical_loss": 4.054447553534548,
      "tokens_seen": 364314624
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044928786359077234,
      "loss": 3.1496,
      "theoretical_loss": 4.054364678711315,
      "tokens_seen": 364380160
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004492778335005015,
      "loss": 3.1532,
      "theoretical_loss": 4.054281822964984,
      "tokens_seen": 364445696
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004492678034102307,
      "loss": 3.2837,
      "theoretical_loss": 4.054198986287732,
      "tokens_seen": 364511232
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004492577733199599,
      "loss": 3.1619,
      "theoretical_loss": 4.054116168671743,
      "tokens_seen": 364576768
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004492477432296891,
      "loss": 3.119,
      "theoretical_loss": 4.054033370109206,
      "tokens_seen": 364642304
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044923771313941824,
      "loss": 3.1911,
      "theoretical_loss": 4.053950590592316,
      "tokens_seen": 364707840
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004492276830491475,
      "loss": 3.5094,
      "theoretical_loss": 4.053867830113268,
      "tokens_seen": 364773376
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004492176529588766,
      "loss": 3.2944,
      "theoretical_loss": 4.053785088664266,
      "tokens_seen": 364838912
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044920762286860584,
      "loss": 3.1059,
      "theoretical_loss": 4.053702366237517,
      "tokens_seen": 364904448
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.000449197592778335,
      "loss": 2.972,
      "theoretical_loss": 4.053619662825232,
      "tokens_seen": 364969984
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004491875626880642,
      "loss": 3.1268,
      "theoretical_loss": 4.053536978419627,
      "tokens_seen": 365035520
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004491775325977934,
      "loss": 3.1372,
      "theoretical_loss": 4.053454313012924,
      "tokens_seen": 365101056
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044916750250752256,
      "loss": 3.1413,
      "theoretical_loss": 4.053371666597348,
      "tokens_seen": 365166592
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044915747241725174,
      "loss": 3.3886,
      "theoretical_loss": 4.053289039165128,
      "tokens_seen": 365232128
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.000449147442326981,
      "loss": 3.2397,
      "theoretical_loss": 4.0532064307085,
      "tokens_seen": 365297664
    },
    {
      "epoch": 0.11,
      "objective/train/advantage_avg": 0.4940948486328125,
      "objective/train/docs_used": 218231,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.6605443954467773,
      "objective/train/original_loss": 2.6605443954467773,
      "objective/train/theoretical_loss": 4.053123841219701,
      "objective/train/tokens_used": 385823200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24421502649784088,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050650954246521,
      "objective/train/weighted_lm_loss": 2.794917106628418,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0304546356201172,
      "theoretical_loss": 4.053123841219701,
      "tokens_seen": 365363200
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004491374122367101,
      "loss": 3.0073,
      "theoretical_loss": 4.053123841219701,
      "tokens_seen": 365363200
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044912738214643934,
      "loss": 3.2791,
      "theoretical_loss": 4.053041270690977,
      "tokens_seen": 365428736
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044911735205616847,
      "loss": 3.2406,
      "theoretical_loss": 4.052958719114574,
      "tokens_seen": 365494272
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004491073219658977,
      "loss": 3.3176,
      "theoretical_loss": 4.052876186482747,
      "tokens_seen": 365559808
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004490972918756269,
      "loss": 3.124,
      "theoretical_loss": 4.052793672787751,
      "tokens_seen": 365625344
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044908726178535607,
      "loss": 3.3356,
      "theoretical_loss": 4.05271117802185,
      "tokens_seen": 365690880
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044907723169508525,
      "loss": 3.1921,
      "theoretical_loss": 4.052628702177309,
      "tokens_seen": 365756416
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004490672016048145,
      "loss": 3.27,
      "theoretical_loss": 4.052546245246399,
      "tokens_seen": 365821952
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004490571715145436,
      "loss": 2.8891,
      "theoretical_loss": 4.052463807221396,
      "tokens_seen": 365887488
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044904714142427284,
      "loss": 3.0004,
      "theoretical_loss": 4.05238138809458,
      "tokens_seen": 365953024
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044903711133400197,
      "loss": 3.1853,
      "theoretical_loss": 4.052298987858236,
      "tokens_seen": 366018560
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004490270812437312,
      "loss": 3.2174,
      "theoretical_loss": 4.052216606504652,
      "tokens_seen": 366084096
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004490170511534604,
      "loss": 3.166,
      "theoretical_loss": 4.0521342440261225,
      "tokens_seen": 366149632
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044900702106318957,
      "loss": 3.1148,
      "theoretical_loss": 4.052051900414945,
      "tokens_seen": 366215168
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044899699097291875,
      "loss": 3.0168,
      "theoretical_loss": 4.051969575663423,
      "tokens_seen": 366280704
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044898696088264793,
      "loss": 3.0834,
      "theoretical_loss": 4.0518872697638635,
      "tokens_seen": 366346240
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004489769307923771,
      "loss": 3.2566,
      "theoretical_loss": 4.051804982708577,
      "tokens_seen": 366411776
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044896690070210635,
      "loss": 3.1989,
      "theoretical_loss": 4.051722714489882,
      "tokens_seen": 366477312
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004489568706118355,
      "loss": 2.9968,
      "theoretical_loss": 4.051640465100097,
      "tokens_seen": 366542848
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004489468405215647,
      "loss": 3.0451,
      "theoretical_loss": 4.051558234531548,
      "tokens_seen": 366608384
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044893681043129395,
      "loss": 3.3156,
      "theoretical_loss": 4.051476022776566,
      "tokens_seen": 366673920
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044892678034102307,
      "loss": 3.4372,
      "theoretical_loss": 4.051393829827483,
      "tokens_seen": 366739456
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004489167502507523,
      "loss": 3.0035,
      "theoretical_loss": 4.051311655676639,
      "tokens_seen": 366804992
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044890672016048143,
      "loss": 2.9695,
      "theoretical_loss": 4.051229500316378,
      "tokens_seen": 366870528
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044889669007021067,
      "loss": 3.1096,
      "theoretical_loss": 4.051147363739045,
      "tokens_seen": 366936064
    },
    {
      "epoch": 0.11,
      "objective/train/advantage_avg": 0.49469780921936035,
      "objective/train/docs_used": 219467,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.986088752746582,
      "objective/train/original_loss": 2.9860892295837402,
      "objective/train/theoretical_loss": 4.051065245936996,
      "objective/train/tokens_used": 387461600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24478444457054138,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507140159606934,
      "objective/train/weighted_lm_loss": 3.1373984813690186,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 4.051065245936996,
      "tokens_seen": 367001600
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044888665997993985,
      "loss": 3.0496,
      "theoretical_loss": 4.051065245936996,
      "tokens_seen": 367001600
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044887662988966903,
      "loss": 3.1589,
      "theoretical_loss": 4.050983146902584,
      "tokens_seen": 367067136
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004488665997993982,
      "loss": 3.3101,
      "theoretical_loss": 4.050901066628172,
      "tokens_seen": 367132672
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004488565697091274,
      "loss": 3.1505,
      "theoretical_loss": 4.050819005106125,
      "tokens_seen": 367198208
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004488465396188566,
      "loss": 3.0701,
      "theoretical_loss": 4.050736962328814,
      "tokens_seen": 367263744
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004488365095285858,
      "loss": 3.084,
      "theoretical_loss": 4.050654938288613,
      "tokens_seen": 367329280
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044882647943831494,
      "loss": 2.8914,
      "theoretical_loss": 4.0505729329779,
      "tokens_seen": 367394816
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004488164493480442,
      "loss": 2.8875,
      "theoretical_loss": 4.0504909463890595,
      "tokens_seen": 367460352
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004488064192577733,
      "loss": 3.3176,
      "theoretical_loss": 4.050408978514479,
      "tokens_seen": 367525888
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044879638916750254,
      "loss": 3.3126,
      "theoretical_loss": 4.050327029346551,
      "tokens_seen": 367591424
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004487863590772317,
      "loss": 3.2164,
      "theoretical_loss": 4.0502450988776735,
      "tokens_seen": 367656960
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004487763289869609,
      "loss": 3.1233,
      "theoretical_loss": 4.050163187100246,
      "tokens_seen": 367722496
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004487662988966901,
      "loss": 3.1081,
      "theoretical_loss": 4.050081294006675,
      "tokens_seen": 367788032
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004487562688064193,
      "loss": 2.9287,
      "theoretical_loss": 4.04999941958937,
      "tokens_seen": 367853568
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044874623871614844,
      "loss": 3.3721,
      "theoretical_loss": 4.049917563840747,
      "tokens_seen": 367919104
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004487362086258777,
      "loss": 3.0565,
      "theoretical_loss": 4.049835726753223,
      "tokens_seen": 367984640
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004487261785356068,
      "loss": 3.0242,
      "theoretical_loss": 4.049753908319223,
      "tokens_seen": 368050176
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044871614844533604,
      "loss": 3.1503,
      "theoretical_loss": 4.049672108531174,
      "tokens_seen": 368115712
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004487061183550652,
      "loss": 3.1776,
      "theoretical_loss": 4.049590327381509,
      "tokens_seen": 368181248
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004486960882647944,
      "loss": 3.12,
      "theoretical_loss": 4.049508564862664,
      "tokens_seen": 368246784
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004486860581745236,
      "loss": 3.2111,
      "theoretical_loss": 4.04942682096708,
      "tokens_seen": 368312320
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044867602808425276,
      "loss": 3.2836,
      "theoretical_loss": 4.049345095687203,
      "tokens_seen": 368377856
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044866599799398194,
      "loss": 3.0954,
      "theoretical_loss": 4.049263389015484,
      "tokens_seen": 368443392
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004486559679037112,
      "loss": 3.2793,
      "theoretical_loss": 4.0491817009443745,
      "tokens_seen": 368508928
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004486459378134403,
      "loss": 2.8804,
      "theoretical_loss": 4.0491000314663355,
      "tokens_seen": 368574464
    },
    {
      "epoch": 0.11,
      "objective/train/advantage_avg": 0.49493730068206787,
      "objective/train/docs_used": 220098,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.2176449298858643,
      "objective/train/original_loss": 3.2176449298858643,
      "objective/train/theoretical_loss": 4.04901838057383,
      "objective/train/tokens_used": 389100000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24503342807292938,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507392883300781,
      "objective/train/weighted_lm_loss": 3.3809900283813477,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0338950157165527,
      "theoretical_loss": 4.04901838057383,
      "tokens_seen": 368640000
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044863590772316954,
      "loss": 3.2652,
      "theoretical_loss": 4.04901838057383,
      "tokens_seen": 368640000
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044862587763289867,
      "loss": 3.1112,
      "theoretical_loss": 4.048936748259324,
      "tokens_seen": 368705536
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004486158475426279,
      "loss": 3.0866,
      "theoretical_loss": 4.048855134515291,
      "tokens_seen": 368771072
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004486058174523571,
      "loss": 3.2157,
      "theoretical_loss": 4.048773539334207,
      "tokens_seen": 368836608
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044859578736208627,
      "loss": 3.3665,
      "theoretical_loss": 4.048691962708553,
      "tokens_seen": 368902144
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044858575727181545,
      "loss": 3.3508,
      "theoretical_loss": 4.0486104046308125,
      "tokens_seen": 368967680
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004485757271815447,
      "loss": 3.2534,
      "theoretical_loss": 4.048528865093476,
      "tokens_seen": 369033216
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004485656970912738,
      "loss": 3.0411,
      "theoretical_loss": 4.048447344089038,
      "tokens_seen": 369098752
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044855566700100304,
      "loss": 3.0961,
      "theoretical_loss": 4.048365841609995,
      "tokens_seen": 369164288
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044854563691073217,
      "loss": 3.3174,
      "theoretical_loss": 4.04828435764885,
      "tokens_seen": 369229824
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004485356068204614,
      "loss": 2.9119,
      "theoretical_loss": 4.048202892198112,
      "tokens_seen": 369295360
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004485255767301906,
      "loss": 3.2446,
      "theoretical_loss": 4.048121445250289,
      "tokens_seen": 369360896
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044851554663991977,
      "loss": 3.2006,
      "theoretical_loss": 4.048040016797899,
      "tokens_seen": 369426432
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044850551654964895,
      "loss": 3.2438,
      "theoretical_loss": 4.047958606833462,
      "tokens_seen": 369491968
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044849548645937813,
      "loss": 3.1137,
      "theoretical_loss": 4.0478772153495015,
      "tokens_seen": 369557504
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004484854563691073,
      "loss": 3.1083,
      "theoretical_loss": 4.047795842338545,
      "tokens_seen": 369623040
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044847542627883655,
      "loss": 3.2162,
      "theoretical_loss": 4.047714487793129,
      "tokens_seen": 369688576
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004484653961885657,
      "loss": 3.0872,
      "theoretical_loss": 4.047633151705787,
      "tokens_seen": 369754112
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004484553660982949,
      "loss": 3.161,
      "theoretical_loss": 4.047551834069064,
      "tokens_seen": 369819648
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044844533600802404,
      "loss": 3.4215,
      "theoretical_loss": 4.047470534875504,
      "tokens_seen": 369885184
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044843530591775327,
      "loss": 3.0655,
      "theoretical_loss": 4.047389254117657,
      "tokens_seen": 369950720
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044842527582748245,
      "loss": 3.1695,
      "theoretical_loss": 4.047307991788079,
      "tokens_seen": 370016256
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044841524573721163,
      "loss": 3.0527,
      "theoretical_loss": 4.047226747879329,
      "tokens_seen": 370081792
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004484052156469408,
      "loss": 3.1638,
      "theoretical_loss": 4.04714552238397,
      "tokens_seen": 370147328
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044839518555667005,
      "loss": 3.1099,
      "theoretical_loss": 4.04706431529457,
      "tokens_seen": 370212864
    },
    {
      "epoch": 0.11,
      "objective/train/advantage_avg": 0.49376192688941956,
      "objective/train/docs_used": 221122,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.7026710510253906,
      "objective/train/original_loss": 2.7026712894439697,
      "objective/train/theoretical_loss": 4.0469831266037,
      "objective/train/tokens_used": 390738400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24386447668075562,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050615906715393,
      "objective/train/weighted_lm_loss": 2.8396804332733154,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0408108234405518,
      "theoretical_loss": 4.0469831266037,
      "tokens_seen": 370278400
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004483851554663992,
      "loss": 3.1989,
      "theoretical_loss": 4.0469831266037,
      "tokens_seen": 370278400
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004483751253761284,
      "loss": 3.1756,
      "theoretical_loss": 4.0469019563039375,
      "tokens_seen": 370343936
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044836509528585754,
      "loss": 3.1289,
      "theoretical_loss": 4.046820804387862,
      "tokens_seen": 370409472
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004483550651955868,
      "loss": 3.1171,
      "theoretical_loss": 4.046739670848059,
      "tokens_seen": 370475008
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044834503510531596,
      "loss": 3.2333,
      "theoretical_loss": 4.046658555677118,
      "tokens_seen": 370540544
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044833500501504514,
      "loss": 3.0616,
      "theoretical_loss": 4.046577458867633,
      "tokens_seen": 370606080
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004483249749247743,
      "loss": 3.2757,
      "theoretical_loss": 4.046496380412201,
      "tokens_seen": 370671616
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004483149448345035,
      "loss": 3.0806,
      "theoretical_loss": 4.0464153203034225,
      "tokens_seen": 370737152
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004483049147442327,
      "loss": 2.9592,
      "theoretical_loss": 4.046334278533907,
      "tokens_seen": 370802688
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004482948846539619,
      "loss": 3.3001,
      "theoretical_loss": 4.046253255096263,
      "tokens_seen": 370868224
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044828485456369104,
      "loss": 3.0894,
      "theoretical_loss": 4.0461722499831065,
      "tokens_seen": 370933760
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004482748244734203,
      "loss": 3.0832,
      "theoretical_loss": 4.046091263187057,
      "tokens_seen": 370999296
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004482647943831494,
      "loss": 3.0082,
      "theoretical_loss": 4.046010294700737,
      "tokens_seen": 371064832
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044825476429287864,
      "loss": 2.9631,
      "theoretical_loss": 4.045929344516774,
      "tokens_seen": 371130368
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004482447342026078,
      "loss": 3.3603,
      "theoretical_loss": 4.045848412627802,
      "tokens_seen": 371195904
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.000448234704112337,
      "loss": 3.0351,
      "theoretical_loss": 4.045767499026455,
      "tokens_seen": 371261440
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004482246740220662,
      "loss": 3.0471,
      "theoretical_loss": 4.045686603705375,
      "tokens_seen": 371326976
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004482146439317954,
      "loss": 3.0304,
      "theoretical_loss": 4.045605726657207,
      "tokens_seen": 371392512
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044820461384152455,
      "loss": 3.277,
      "theoretical_loss": 4.045524867874599,
      "tokens_seen": 371458048
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004481945837512538,
      "loss": 3.4506,
      "theoretical_loss": 4.0454440273502055,
      "tokens_seen": 371523584
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044818455366098296,
      "loss": 3.1895,
      "theoretical_loss": 4.045363205076683,
      "tokens_seen": 371589120
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044817452357071214,
      "loss": 3.3565,
      "theoretical_loss": 4.045282401046695,
      "tokens_seen": 371654656
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004481644934804414,
      "loss": 3.2092,
      "theoretical_loss": 4.045201615252906,
      "tokens_seen": 371720192
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004481544633901705,
      "loss": 3.056,
      "theoretical_loss": 4.045120847687986,
      "tokens_seen": 371785728
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044814443329989974,
      "loss": 3.3509,
      "theoretical_loss": 4.045040098344612,
      "tokens_seen": 371851264
    },
    {
      "epoch": 0.11,
      "objective/train/advantage_avg": 0.4953921139240265,
      "objective/train/docs_used": 221664,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.069113254547119,
      "objective/train/original_loss": 3.0691137313842773,
      "objective/train/theoretical_loss": 4.04495936721546,
      "objective/train/tokens_used": 392376800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24548228085041046,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050787091255188,
      "objective/train/weighted_lm_loss": 3.2249245643615723,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0408108234405518,
      "theoretical_loss": 4.04495936721546,
      "tokens_seen": 371916800
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044813440320962887,
      "loss": 3.1786,
      "theoretical_loss": 4.04495936721546,
      "tokens_seen": 371916800
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004481243731193581,
      "loss": 3.063,
      "theoretical_loss": 4.044878654293215,
      "tokens_seen": 371982336
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004481143430290873,
      "loss": 3.3317,
      "theoretical_loss": 4.044797959570563,
      "tokens_seen": 372047872
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044810431293881647,
      "loss": 3.224,
      "theoretical_loss": 4.044717283040197,
      "tokens_seen": 372113408
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044809428284854565,
      "loss": 3.1819,
      "theoretical_loss": 4.044636624694812,
      "tokens_seen": 372178944
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004480842527582749,
      "loss": 3.2481,
      "theoretical_loss": 4.044555984527107,
      "tokens_seen": 372244480
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.000448074222668004,
      "loss": 3.3322,
      "theoretical_loss": 4.044475362529786,
      "tokens_seen": 372310016
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044806419257773324,
      "loss": 3.1924,
      "theoretical_loss": 4.044394758695559,
      "tokens_seen": 372375552
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044805416248746237,
      "loss": 3.0771,
      "theoretical_loss": 4.044314173017138,
      "tokens_seen": 372441088
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004480441323971916,
      "loss": 3.2102,
      "theoretical_loss": 4.04423360548724,
      "tokens_seen": 372506624
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004480341023069208,
      "loss": 3.1808,
      "theoretical_loss": 4.044153056098585,
      "tokens_seen": 372572160
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044802407221664997,
      "loss": 3.0944,
      "theoretical_loss": 4.044072524843899,
      "tokens_seen": 372637696
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044801404212637915,
      "loss": 3.178,
      "theoretical_loss": 4.04399201171591,
      "tokens_seen": 372703232
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044800401203610833,
      "loss": 3.2075,
      "theoretical_loss": 4.043911516707354,
      "tokens_seen": 372768768
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004479939819458375,
      "loss": 3.2215,
      "theoretical_loss": 4.043831039810967,
      "tokens_seen": 372834304
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044798395185556675,
      "loss": 3.0415,
      "theoretical_loss": 4.043750581019491,
      "tokens_seen": 372899840
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004479739217652959,
      "loss": 2.9836,
      "theoretical_loss": 4.043670140325673,
      "tokens_seen": 372965376
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004479638916750251,
      "loss": 3.1518,
      "theoretical_loss": 4.043589717722263,
      "tokens_seen": 373030912
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044795386158475424,
      "loss": 3.3139,
      "theoretical_loss": 4.043509313202014,
      "tokens_seen": 373096448
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044794383149448347,
      "loss": 3.0307,
      "theoretical_loss": 4.043428926757687,
      "tokens_seen": 373161984
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044793380140421265,
      "loss": 3.1838,
      "theoretical_loss": 4.043348558382044,
      "tokens_seen": 373227520
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044792377131394183,
      "loss": 3.2109,
      "theoretical_loss": 4.04326820806785,
      "tokens_seen": 373293056
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.000447913741223671,
      "loss": 3.1841,
      "theoretical_loss": 4.043187875807879,
      "tokens_seen": 373358592
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044790371113340025,
      "loss": 2.9671,
      "theoretical_loss": 4.043107561594906,
      "tokens_seen": 373424128
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004478936810431294,
      "loss": 2.75,
      "theoretical_loss": 4.043027265421709,
      "tokens_seen": 373489664
    },
    {
      "epoch": 0.11,
      "objective/train/advantage_avg": 0.49529731273651123,
      "objective/train/docs_used": 222906,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.7403268814086914,
      "objective/train/original_loss": 2.7403271198272705,
      "objective/train/theoretical_loss": 4.042946987281072,
      "objective/train/tokens_used": 394015200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24538378417491913,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507771968841553,
      "objective/train/weighted_lm_loss": 2.8793253898620605,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0417574644088745,
      "theoretical_loss": 4.042946987281072,
      "tokens_seen": 373555200
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004478836509528586,
      "loss": 3.147,
      "theoretical_loss": 4.042946987281072,
      "tokens_seen": 373555200
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044787362086258774,
      "loss": 3.1466,
      "theoretical_loss": 4.042866727165784,
      "tokens_seen": 373620736
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.000447863590772317,
      "loss": 2.9486,
      "theoretical_loss": 4.0427864850686355,
      "tokens_seen": 373686272
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044785356068204616,
      "loss": 3.0861,
      "theoretical_loss": 4.042706260982422,
      "tokens_seen": 373751808
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044784353059177534,
      "loss": 2.8711,
      "theoretical_loss": 4.042626054899946,
      "tokens_seen": 373817344
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004478335005015045,
      "loss": 3.1448,
      "theoretical_loss": 4.042545866814011,
      "tokens_seen": 373882880
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004478234704112337,
      "loss": 2.868,
      "theoretical_loss": 4.042465696717425,
      "tokens_seen": 373948416
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004478134403209629,
      "loss": 3.0301,
      "theoretical_loss": 4.042385544603,
      "tokens_seen": 374013952
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004478034102306921,
      "loss": 3.0293,
      "theoretical_loss": 4.0423054104635545,
      "tokens_seen": 374079488
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044779338014042124,
      "loss": 3.3707,
      "theoretical_loss": 4.042225294291908,
      "tokens_seen": 374145024
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004477833500501505,
      "loss": 3.0842,
      "theoretical_loss": 4.042145196080887,
      "tokens_seen": 374210560
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004477733199598796,
      "loss": 3.3066,
      "theoretical_loss": 4.042065115823319,
      "tokens_seen": 374276096
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044776328986960884,
      "loss": 3.0456,
      "theoretical_loss": 4.041985053512038,
      "tokens_seen": 374341632
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.000447753259779338,
      "loss": 3.1203,
      "theoretical_loss": 4.041905009139881,
      "tokens_seen": 374407168
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004477432296890672,
      "loss": 3.2772,
      "theoretical_loss": 4.04182498269969,
      "tokens_seen": 374472704
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004477331995987964,
      "loss": 2.9045,
      "theoretical_loss": 4.04174497418431,
      "tokens_seen": 374538240
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004477231695085256,
      "loss": 3.1371,
      "theoretical_loss": 4.041664983586591,
      "tokens_seen": 374603776
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044771313941825475,
      "loss": 3.2061,
      "theoretical_loss": 4.041585010899388,
      "tokens_seen": 374669312
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.000447703109327984,
      "loss": 2.9929,
      "theoretical_loss": 4.041505056115556,
      "tokens_seen": 374734848
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004476930792377131,
      "loss": 3.2631,
      "theoretical_loss": 4.041425119227959,
      "tokens_seen": 374800384
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044768304914744234,
      "loss": 3.3154,
      "theoretical_loss": 4.041345200229464,
      "tokens_seen": 374865920
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004476730190571715,
      "loss": 3.0333,
      "theoretical_loss": 4.041265299112939,
      "tokens_seen": 374931456
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004476629889669007,
      "loss": 3.1663,
      "theoretical_loss": 4.04118541587126,
      "tokens_seen": 374996992
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004476529588766299,
      "loss": 3.0911,
      "theoretical_loss": 4.041105550497304,
      "tokens_seen": 375062528
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044764292878635907,
      "loss": 3.2468,
      "theoretical_loss": 4.041025702983954,
      "tokens_seen": 375128064
    },
    {
      "epoch": 0.11,
      "objective/train/advantage_avg": 0.4930410087108612,
      "objective/train/docs_used": 223164,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.100318431854248,
      "objective/train/original_loss": 3.100318670272827,
      "objective/train/theoretical_loss": 4.040945873324096,
      "objective/train/tokens_used": 395653600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24316418170928955,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505400896072388,
      "objective/train/weighted_lm_loss": 3.2571089267730713,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 4.040945873324096,
      "tokens_seen": 375193600
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044763289869608825,
      "loss": 3.2968,
      "theoretical_loss": 4.040945873324096,
      "tokens_seen": 375193600
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004476228686058175,
      "loss": 3.2593,
      "theoretical_loss": 4.040866061510623,
      "tokens_seen": 375259136
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004476128385155466,
      "loss": 2.9845,
      "theoretical_loss": 4.040786267536426,
      "tokens_seen": 375324672
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044760280842527585,
      "loss": 3.2305,
      "theoretical_loss": 4.040706491394406,
      "tokens_seen": 375390208
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044759277833500503,
      "loss": 3.265,
      "theoretical_loss": 4.040626733077466,
      "tokens_seen": 375455744
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004475827482447342,
      "loss": 3.186,
      "theoretical_loss": 4.040546992578511,
      "tokens_seen": 375521280
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004475727181544634,
      "loss": 3.1243,
      "theoretical_loss": 4.040467269890454,
      "tokens_seen": 375586816
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044756268806419257,
      "loss": 3.1105,
      "theoretical_loss": 4.040387565006208,
      "tokens_seen": 375652352
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044755265797392175,
      "loss": 3.3381,
      "theoretical_loss": 4.0403078779186945,
      "tokens_seen": 375717888
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.000447542627883651,
      "loss": 3.1435,
      "theoretical_loss": 4.040228208620833,
      "tokens_seen": 375783424
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004475325977933801,
      "loss": 3.1444,
      "theoretical_loss": 4.040148557105555,
      "tokens_seen": 375848960
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044752256770310935,
      "loss": 3.2452,
      "theoretical_loss": 4.040068923365788,
      "tokens_seen": 375914496
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004475125376128385,
      "loss": 2.939,
      "theoretical_loss": 4.039989307394469,
      "tokens_seen": 375980032
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004475025075225677,
      "loss": 3.1292,
      "theoretical_loss": 4.0399097091845375,
      "tokens_seen": 376045568
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004474924774322969,
      "loss": 3.1096,
      "theoretical_loss": 4.039830128728935,
      "tokens_seen": 376111104
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004474824473420261,
      "loss": 3.2522,
      "theoretical_loss": 4.039750566020611,
      "tokens_seen": 376176640
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044747241725175526,
      "loss": 3.0426,
      "theoretical_loss": 4.039671021052515,
      "tokens_seen": 376242176
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044746238716148444,
      "loss": 3.1976,
      "theoretical_loss": 4.039591493817603,
      "tokens_seen": 376307712
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004474523570712136,
      "loss": 2.9204,
      "theoretical_loss": 4.039511984308835,
      "tokens_seen": 376373248
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044744232698094285,
      "loss": 3.2294,
      "theoretical_loss": 4.0394324925191745,
      "tokens_seen": 376438784
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044743229689067204,
      "loss": 3.0463,
      "theoretical_loss": 4.039353018441588,
      "tokens_seen": 376504320
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004474222668004012,
      "loss": 3.3719,
      "theoretical_loss": 4.039273562069047,
      "tokens_seen": 376569856
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044741223671013045,
      "loss": 2.9801,
      "theoretical_loss": 4.0391941233945285,
      "tokens_seen": 376635392
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004474022066198596,
      "loss": 3.1127,
      "theoretical_loss": 4.039114702411009,
      "tokens_seen": 376700928
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004473921765295888,
      "loss": 3.2905,
      "theoretical_loss": 4.039035299111475,
      "tokens_seen": 376766464
    },
    {
      "epoch": 0.11,
      "objective/train/advantage_avg": 0.4941399097442627,
      "objective/train/docs_used": 224842,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.0917468070983887,
      "objective/train/original_loss": 3.0917468070983887,
      "objective/train/theoretical_loss": 4.038955913488913,
      "objective/train/tokens_used": 397292000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24422840774059296,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050655484199524,
      "objective/train/weighted_lm_loss": 3.2481155395507812,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0457526445388794,
      "theoretical_loss": 4.038955913488913,
      "tokens_seen": 376832000
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044738214643931794,
      "loss": 3.2992,
      "theoretical_loss": 4.038955913488913,
      "tokens_seen": 376832000
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004473721163490472,
      "loss": 3.4247,
      "theoretical_loss": 4.038876545536313,
      "tokens_seen": 376897536
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044736208625877636,
      "loss": 3.0137,
      "theoretical_loss": 4.038797195246673,
      "tokens_seen": 376963072
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044735205616850554,
      "loss": 3.1196,
      "theoretical_loss": 4.03871786261299,
      "tokens_seen": 377028608
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004473420260782347,
      "loss": 3.1075,
      "theoretical_loss": 4.03863854762827,
      "tokens_seen": 377094144
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004473319959879639,
      "loss": 2.9575,
      "theoretical_loss": 4.038559250285518,
      "tokens_seen": 377159680
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004473219658976931,
      "loss": 3.0808,
      "theoretical_loss": 4.038479970577747,
      "tokens_seen": 377225216
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004473119358074223,
      "loss": 3.2052,
      "theoretical_loss": 4.038400708497971,
      "tokens_seen": 377290752
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044730190571715144,
      "loss": 3.2206,
      "theoretical_loss": 4.03832146403921,
      "tokens_seen": 377356288
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004472918756268807,
      "loss": 3.1935,
      "theoretical_loss": 4.03824223719449,
      "tokens_seen": 377421824
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004472818455366098,
      "loss": 3.089,
      "theoretical_loss": 4.038163027956834,
      "tokens_seen": 377487360
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044727181544633904,
      "loss": 3.2364,
      "theoretical_loss": 4.038083836319276,
      "tokens_seen": 377552896
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004472617853560682,
      "loss": 3.2504,
      "theoretical_loss": 4.038004662274851,
      "tokens_seen": 377618432
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004472517552657974,
      "loss": 3.1627,
      "theoretical_loss": 4.037925505816597,
      "tokens_seen": 377683968
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004472417251755266,
      "loss": 3.0523,
      "theoretical_loss": 4.037846366937559,
      "tokens_seen": 377749504
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004472316950852558,
      "loss": 3.3277,
      "theoretical_loss": 4.037767245630782,
      "tokens_seen": 377815040
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044722166499498495,
      "loss": 3.1849,
      "theoretical_loss": 4.037688141889319,
      "tokens_seen": 377880576
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004472116349047142,
      "loss": 3.0116,
      "theoretical_loss": 4.037609055706224,
      "tokens_seen": 377946112
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004472016048144433,
      "loss": 3.2877,
      "theoretical_loss": 4.0375299870745565,
      "tokens_seen": 378011648
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044719157472417254,
      "loss": 3.0326,
      "theoretical_loss": 4.03745093598738,
      "tokens_seen": 378077184
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004471815446339017,
      "loss": 2.9925,
      "theoretical_loss": 4.03737190243776,
      "tokens_seen": 378142720
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004471715145436309,
      "loss": 2.9216,
      "theoretical_loss": 4.0372928864187685,
      "tokens_seen": 378208256
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004471614844533601,
      "loss": 3.2722,
      "theoretical_loss": 4.03721388792348,
      "tokens_seen": 378273792
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044715145436308927,
      "loss": 3.1814,
      "theoretical_loss": 4.037134906944973,
      "tokens_seen": 378339328
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044714142427281845,
      "loss": 3.1336,
      "theoretical_loss": 4.0370559434763305,
      "tokens_seen": 378404864
    },
    {
      "epoch": 0.11,
      "objective/train/advantage_avg": 0.4949733018875122,
      "objective/train/docs_used": 225570,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.2313029766082764,
      "objective/train/original_loss": 3.2313032150268555,
      "objective/train/theoretical_loss": 4.036976997510639,
      "objective/train/tokens_used": 398930400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24504797160625458,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507431030273438,
      "objective/train/weighted_lm_loss": 3.3952205181121826,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0451053380966187,
      "theoretical_loss": 4.036976997510639,
      "tokens_seen": 378470400
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004471313941825477,
      "loss": 3.1548,
      "theoretical_loss": 4.036976997510639,
      "tokens_seen": 378470400
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004471213640922768,
      "loss": 3.0206,
      "theoretical_loss": 4.036898069040989,
      "tokens_seen": 378535936
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044711133400200605,
      "loss": 3.1495,
      "theoretical_loss": 4.036819158060474,
      "tokens_seen": 378601472
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044710130391173523,
      "loss": 3.2097,
      "theoretical_loss": 4.0367402645621935,
      "tokens_seen": 378667008
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004470912738214644,
      "loss": 3.1262,
      "theoretical_loss": 4.036661388539249,
      "tokens_seen": 378732544
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004470812437311936,
      "loss": 3.4597,
      "theoretical_loss": 4.036582529984749,
      "tokens_seen": 378798080
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044707121364092277,
      "loss": 3.1975,
      "theoretical_loss": 4.0365036888918,
      "tokens_seen": 378863616
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044706118355065195,
      "loss": 3.141,
      "theoretical_loss": 4.036424865253518,
      "tokens_seen": 378929152
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004470511534603812,
      "loss": 3.0957,
      "theoretical_loss": 4.036346059063021,
      "tokens_seen": 378994688
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004470411233701103,
      "loss": 3.385,
      "theoretical_loss": 4.036267270313431,
      "tokens_seen": 379060224
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044703109327983955,
      "loss": 3.2665,
      "theoretical_loss": 4.036188498997872,
      "tokens_seen": 379125760
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004470210631895687,
      "loss": 3.3705,
      "theoretical_loss": 4.036109745109475,
      "tokens_seen": 379191296
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004470110330992979,
      "loss": 3.1582,
      "theoretical_loss": 4.036031008641373,
      "tokens_seen": 379256832
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004470010030090271,
      "loss": 3.5033,
      "theoretical_loss": 4.035952289586705,
      "tokens_seen": 379322368
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.0004469909729187563,
      "loss": 3.1816,
      "theoretical_loss": 4.03587358793861,
      "tokens_seen": 379387904
    },
    {
      "epoch": 0.11,
      "learning_rate": 0.00044698094282848546,
      "loss": 3.1576,
      "theoretical_loss": 4.035794903690235,
      "tokens_seen": 379453440
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044697091273821464,
      "loss": 3.0592,
      "theoretical_loss": 4.035716236834727,
      "tokens_seen": 379518976
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004469608826479438,
      "loss": 3.1981,
      "theoretical_loss": 4.03563758736524,
      "tokens_seen": 379584512
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044695085255767305,
      "loss": 3.3857,
      "theoretical_loss": 4.035558955274933,
      "tokens_seen": 379650048
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004469408224674022,
      "loss": 2.9632,
      "theoretical_loss": 4.035480340556963,
      "tokens_seen": 379715584
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004469307923771314,
      "loss": 3.0685,
      "theoretical_loss": 4.035401743204496,
      "tokens_seen": 379781120
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004469207622868606,
      "loss": 3.3355,
      "theoretical_loss": 4.035323163210702,
      "tokens_seen": 379846656
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004469107321965898,
      "loss": 3.0968,
      "theoretical_loss": 4.035244600568751,
      "tokens_seen": 379912192
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044690070210631896,
      "loss": 3.0622,
      "theoretical_loss": 4.035166055271819,
      "tokens_seen": 379977728
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044689067201604814,
      "loss": 2.9948,
      "theoretical_loss": 4.035087527313088,
      "tokens_seen": 380043264
    },
    {
      "epoch": 0.12,
      "objective/train/advantage_avg": 0.4945727586746216,
      "objective/train/docs_used": 226798,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.2410969734191895,
      "objective/train/original_loss": 3.2410967350006104,
      "objective/train/theoretical_loss": 4.035009016685741,
      "objective/train/tokens_used": 400568800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2446741908788681,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507010221481323,
      "objective/train/weighted_lm_loss": 3.4052376747131348,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0408108234405518,
      "theoretical_loss": 4.035009016685741,
      "tokens_seen": 380108800
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004468806419257773,
      "loss": 3.5391,
      "theoretical_loss": 4.035009016685741,
      "tokens_seen": 380108800
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044687061183550656,
      "loss": 3.2123,
      "theoretical_loss": 4.034930523382966,
      "tokens_seen": 380174336
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004468605817452357,
      "loss": 3.2911,
      "theoretical_loss": 4.034852047397955,
      "tokens_seen": 380239872
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004468505516549649,
      "loss": 3.0175,
      "theoretical_loss": 4.034773588723901,
      "tokens_seen": 380305408
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044684052156469405,
      "loss": 3.1868,
      "theoretical_loss": 4.034695147354006,
      "tokens_seen": 380370944
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004468304914744233,
      "loss": 3.1505,
      "theoretical_loss": 4.034616723281472,
      "tokens_seen": 380436480
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044682046138415246,
      "loss": 3.2534,
      "theoretical_loss": 4.034538316499505,
      "tokens_seen": 380502016
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044681043129388164,
      "loss": 3.1003,
      "theoretical_loss": 4.0344599270013175,
      "tokens_seen": 380567552
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004468004012036108,
      "loss": 3.0355,
      "theoretical_loss": 4.034381554780124,
      "tokens_seen": 380633088
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044679037111334,
      "loss": 3.1652,
      "theoretical_loss": 4.034303199829141,
      "tokens_seen": 380698624
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004467803410230692,
      "loss": 3.036,
      "theoretical_loss": 4.034224862141594,
      "tokens_seen": 380764160
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004467703109327984,
      "loss": 3.1407,
      "theoretical_loss": 4.034146541710705,
      "tokens_seen": 380829696
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044676028084252755,
      "loss": 3.095,
      "theoretical_loss": 4.0340682385297075,
      "tokens_seen": 380895232
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004467502507522568,
      "loss": 3.1386,
      "theoretical_loss": 4.033989952591835,
      "tokens_seen": 380960768
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044674022066198597,
      "loss": 2.8035,
      "theoretical_loss": 4.033911683890322,
      "tokens_seen": 381026304
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044673019057171515,
      "loss": 3.0198,
      "theoretical_loss": 4.0338334324184135,
      "tokens_seen": 381091840
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044672016048144433,
      "loss": 2.8874,
      "theoretical_loss": 4.033755198169352,
      "tokens_seen": 381157376
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004467101303911735,
      "loss": 3.0263,
      "theoretical_loss": 4.033676981136388,
      "tokens_seen": 381222912
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004467001003009027,
      "loss": 3.1028,
      "theoretical_loss": 4.033598781312774,
      "tokens_seen": 381288448
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004466900702106319,
      "loss": 3.2673,
      "theoretical_loss": 4.033520598691767,
      "tokens_seen": 381353984
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004466800401203611,
      "loss": 2.9594,
      "theoretical_loss": 4.033442433266627,
      "tokens_seen": 381419520
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004466700100300903,
      "loss": 2.9119,
      "theoretical_loss": 4.0333642850306175,
      "tokens_seen": 381485056
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044665997993981947,
      "loss": 3.1771,
      "theoretical_loss": 4.0332861539770075,
      "tokens_seen": 381550592
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044664994984954865,
      "loss": 3.1667,
      "theoretical_loss": 4.033208040099068,
      "tokens_seen": 381616128
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004466399197592779,
      "loss": 3.0067,
      "theoretical_loss": 4.033129943390076,
      "tokens_seen": 381681664
    },
    {
      "epoch": 0.12,
      "objective/train/advantage_avg": 0.49497804045677185,
      "objective/train/docs_used": 227456,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.530057668685913,
      "objective/train/original_loss": 3.530057907104492,
      "objective/train/theoretical_loss": 4.033051863843309,
      "objective/train/tokens_used": 402207200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24506641924381256,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050743579864502,
      "objective/train/weighted_lm_loss": 3.709152936935425,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0363596677780151,
      "theoretical_loss": 4.033051863843309,
      "tokens_seen": 381747200
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.000446629889669007,
      "loss": 3.2583,
      "theoretical_loss": 4.033051863843309,
      "tokens_seen": 381747200
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044661985957873625,
      "loss": 3.2522,
      "theoretical_loss": 4.032973801452052,
      "tokens_seen": 381812736
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044660982948846543,
      "loss": 3.1232,
      "theoretical_loss": 4.032895756209591,
      "tokens_seen": 381878272
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004465997993981946,
      "loss": 3.162,
      "theoretical_loss": 4.032817728109217,
      "tokens_seen": 381943808
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004465897693079238,
      "loss": 3.3204,
      "theoretical_loss": 4.032739717144223,
      "tokens_seen": 382009344
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044657973921765297,
      "loss": 3.1032,
      "theoretical_loss": 4.03266172330791,
      "tokens_seen": 382074880
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044656970912738215,
      "loss": 3.366,
      "theoretical_loss": 4.032583746593578,
      "tokens_seen": 382140416
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004465596790371114,
      "loss": 3.0726,
      "theoretical_loss": 4.032505786994534,
      "tokens_seen": 382205952
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004465496489468405,
      "loss": 2.9753,
      "theoretical_loss": 4.032427844504086,
      "tokens_seen": 382271488
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044653961885656975,
      "loss": 3.0117,
      "theoretical_loss": 4.032349919115549,
      "tokens_seen": 382337024
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004465295887662989,
      "loss": 3.4777,
      "theoretical_loss": 4.032272010822239,
      "tokens_seen": 382402560
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004465195586760281,
      "loss": 3.0809,
      "theoretical_loss": 4.032194119617477,
      "tokens_seen": 382468096
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004465095285857573,
      "loss": 3.2633,
      "theoretical_loss": 4.032116245494589,
      "tokens_seen": 382533632
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004464994984954865,
      "loss": 3.205,
      "theoretical_loss": 4.032038388446901,
      "tokens_seen": 382599168
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044648946840521566,
      "loss": 3.0535,
      "theoretical_loss": 4.031960548467747,
      "tokens_seen": 382664704
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044647943831494484,
      "loss": 3.1826,
      "theoretical_loss": 4.031882725550463,
      "tokens_seen": 382730240
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.000446469408224674,
      "loss": 3.2434,
      "theoretical_loss": 4.031804919688387,
      "tokens_seen": 382795776
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044645937813440325,
      "loss": 3.0489,
      "theoretical_loss": 4.031727130874863,
      "tokens_seen": 382861312
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004464493480441324,
      "loss": 3.0439,
      "theoretical_loss": 4.031649359103238,
      "tokens_seen": 382926848
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004464393179538616,
      "loss": 3.3107,
      "theoretical_loss": 4.031571604366864,
      "tokens_seen": 382992384
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004464292878635908,
      "loss": 2.9523,
      "theoretical_loss": 4.031493866659092,
      "tokens_seen": 383057920
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044641925777332,
      "loss": 3.1759,
      "theoretical_loss": 4.031416145973285,
      "tokens_seen": 383123456
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044640922768304916,
      "loss": 3.2387,
      "theoretical_loss": 4.031338442302802,
      "tokens_seen": 383188992
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044639919759277834,
      "loss": 3.2281,
      "theoretical_loss": 4.03126075564101,
      "tokens_seen": 383254528
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004463891675025075,
      "loss": 3.1399,
      "theoretical_loss": 4.031183085981278,
      "tokens_seen": 383320064
    },
    {
      "epoch": 0.12,
      "objective/train/advantage_avg": 0.4941565692424774,
      "objective/train/docs_used": 228876,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.1912314891815186,
      "objective/train/original_loss": 3.1912317276000977,
      "objective/train/theoretical_loss": 4.031105433316977,
      "objective/train/tokens_used": 403845600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24424496293067932,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506571531295776,
      "objective/train/weighted_lm_loss": 3.352841377258301,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 4.031105433316977,
      "tokens_seen": 383385600
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044637913741223676,
      "loss": 3.215,
      "theoretical_loss": 4.031105433316977,
      "tokens_seen": 383385600
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004463691073219659,
      "loss": 3.1703,
      "theoretical_loss": 4.031027797641488,
      "tokens_seen": 383451136
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004463590772316951,
      "loss": 3.1677,
      "theoretical_loss": 4.0309501789481885,
      "tokens_seen": 383516672
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044634904714142425,
      "loss": 3.2495,
      "theoretical_loss": 4.030872577230465,
      "tokens_seen": 383582208
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004463390170511535,
      "loss": 3.3269,
      "theoretical_loss": 4.030794992481702,
      "tokens_seen": 383647744
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044632898696088266,
      "loss": 3.0878,
      "theoretical_loss": 4.030717424695295,
      "tokens_seen": 383713280
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044631895687061184,
      "loss": 3.193,
      "theoretical_loss": 4.030639873864638,
      "tokens_seen": 383778816
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.000446308926780341,
      "loss": 3.2037,
      "theoretical_loss": 4.030562339983129,
      "tokens_seen": 383844352
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004462988966900702,
      "loss": 3.2436,
      "theoretical_loss": 4.030484823044173,
      "tokens_seen": 383909888
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004462888665997994,
      "loss": 3.1171,
      "theoretical_loss": 4.0304073230411745,
      "tokens_seen": 383975424
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004462788365095286,
      "loss": 2.9076,
      "theoretical_loss": 4.0303298399675445,
      "tokens_seen": 384040960
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044626880641925775,
      "loss": 2.9725,
      "theoretical_loss": 4.030252373816697,
      "tokens_seen": 384106496
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.000446258776328987,
      "loss": 3.081,
      "theoretical_loss": 4.030174924582049,
      "tokens_seen": 384172032
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044624874623871617,
      "loss": 3.207,
      "theoretical_loss": 4.030097492257021,
      "tokens_seen": 384237568
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044623871614844535,
      "loss": 2.952,
      "theoretical_loss": 4.03002007683504,
      "tokens_seen": 384303104
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044622868605817453,
      "loss": 3.1611,
      "theoretical_loss": 4.029942678309534,
      "tokens_seen": 384368640
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004462186559679037,
      "loss": 3.229,
      "theoretical_loss": 4.029865296673934,
      "tokens_seen": 384434176
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004462086258776329,
      "loss": 3.263,
      "theoretical_loss": 4.029787931921677,
      "tokens_seen": 384499712
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004461985957873621,
      "loss": 3.2964,
      "theoretical_loss": 4.029710584046203,
      "tokens_seen": 384565248
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044618856569709125,
      "loss": 3.2939,
      "theoretical_loss": 4.029633253040954,
      "tokens_seen": 384630784
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004461785356068205,
      "loss": 3.1907,
      "theoretical_loss": 4.029555938899378,
      "tokens_seen": 384696320
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004461685055165496,
      "loss": 3.215,
      "theoretical_loss": 4.029478641614924,
      "tokens_seen": 384761856
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044615847542627885,
      "loss": 2.906,
      "theoretical_loss": 4.029401361181049,
      "tokens_seen": 384827392
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044614844533600803,
      "loss": 2.9716,
      "theoretical_loss": 4.029324097591209,
      "tokens_seen": 384892928
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004461384152457372,
      "loss": 3.2958,
      "theoretical_loss": 4.029246850838865,
      "tokens_seen": 384958464
    },
    {
      "epoch": 0.12,
      "objective/train/advantage_avg": 0.49539193511009216,
      "objective/train/docs_used": 229454,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.632185220718384,
      "objective/train/original_loss": 2.632185697555542,
      "objective/train/theoretical_loss": 4.029169620917484,
      "objective/train/tokens_used": 405484000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24551180005073547,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507872104644775,
      "objective/train/weighted_lm_loss": 2.765669345855713,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0253151655197144,
      "theoretical_loss": 4.029169620917484,
      "tokens_seen": 385024000
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004461283851554664,
      "loss": 3.2851,
      "theoretical_loss": 4.029169620917484,
      "tokens_seen": 385024000
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044611835506519563,
      "loss": 3.2871,
      "theoretical_loss": 4.029092407820533,
      "tokens_seen": 385089536
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044610832497492476,
      "loss": 3.1318,
      "theoretical_loss": 4.029015211541487,
      "tokens_seen": 385155072
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.000446098294884654,
      "loss": 2.9811,
      "theoretical_loss": 4.028938032073819,
      "tokens_seen": 385220608
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004460882647943831,
      "loss": 3.1994,
      "theoretical_loss": 4.0288608694110115,
      "tokens_seen": 385286144
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044607823470411235,
      "loss": 3.2021,
      "theoretical_loss": 4.028783723546547,
      "tokens_seen": 385351680
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044606820461384153,
      "loss": 3.2525,
      "theoretical_loss": 4.028706594473912,
      "tokens_seen": 385417216
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004460581745235707,
      "loss": 3.2536,
      "theoretical_loss": 4.028629482186598,
      "tokens_seen": 385482752
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004460481444332999,
      "loss": 3.0302,
      "theoretical_loss": 4.028552386678099,
      "tokens_seen": 385548288
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004460381143430291,
      "loss": 3.1321,
      "theoretical_loss": 4.028475307941912,
      "tokens_seen": 385613824
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044602808425275826,
      "loss": 3.0327,
      "theoretical_loss": 4.028398245971541,
      "tokens_seen": 385679360
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004460180541624875,
      "loss": 2.9637,
      "theoretical_loss": 4.028321200760489,
      "tokens_seen": 385744896
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004460080240722166,
      "loss": 2.8939,
      "theoretical_loss": 4.028244172302266,
      "tokens_seen": 385810432
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044599799398194586,
      "loss": 3.2439,
      "theoretical_loss": 4.028167160590383,
      "tokens_seen": 385875968
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.000445987963891675,
      "loss": 3.039,
      "theoretical_loss": 4.028090165618357,
      "tokens_seen": 385941504
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004459779338014042,
      "loss": 3.0951,
      "theoretical_loss": 4.028013187379709,
      "tokens_seen": 386007040
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004459679037111334,
      "loss": 3.1399,
      "theoretical_loss": 4.027936225867959,
      "tokens_seen": 386072576
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004459578736208626,
      "loss": 3.2086,
      "theoretical_loss": 4.027859281076638,
      "tokens_seen": 386138112
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044594784353059176,
      "loss": 2.9644,
      "theoretical_loss": 4.027782352999273,
      "tokens_seen": 386203648
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.000445937813440321,
      "loss": 3.4373,
      "theoretical_loss": 4.027705441629399,
      "tokens_seen": 386269184
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004459277833500502,
      "loss": 2.8851,
      "theoretical_loss": 4.027628546960555,
      "tokens_seen": 386334720
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044591775325977936,
      "loss": 3.1847,
      "theoretical_loss": 4.0275516689862805,
      "tokens_seen": 386400256
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044590772316950854,
      "loss": 2.9011,
      "theoretical_loss": 4.027474807700122,
      "tokens_seen": 386465792
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004458976930792377,
      "loss": 3.2432,
      "theoretical_loss": 4.027397963095627,
      "tokens_seen": 386531328
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044588766298896696,
      "loss": 3.0394,
      "theoretical_loss": 4.027321135166347,
      "tokens_seen": 386596864
    },
    {
      "epoch": 0.12,
      "objective/train/advantage_avg": 0.4952486753463745,
      "objective/train/docs_used": 230121,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.985079526901245,
      "objective/train/original_loss": 2.985078811645508,
      "objective/train/theoretical_loss": 4.027244323905839,
      "objective/train/tokens_used": 407122400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24532413482666016,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050771951675415,
      "objective/train/weighted_lm_loss": 3.136719226837158,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 4.027244323905839,
      "tokens_seen": 386662400
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004458776328986961,
      "loss": 3.1163,
      "theoretical_loss": 4.027244323905839,
      "tokens_seen": 386662400
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004458676028084253,
      "loss": 3.2005,
      "theoretical_loss": 4.027167529307661,
      "tokens_seen": 386727936
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044585757271815445,
      "loss": 3.3476,
      "theoretical_loss": 4.027090751365376,
      "tokens_seen": 386793472
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004458475426278837,
      "loss": 2.9529,
      "theoretical_loss": 4.027013990072551,
      "tokens_seen": 386859008
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044583751253761286,
      "loss": 3.2125,
      "theoretical_loss": 4.026937245422756,
      "tokens_seen": 386924544
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044582748244734204,
      "loss": 3.2574,
      "theoretical_loss": 4.026860517409563,
      "tokens_seen": 386990080
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004458174523570712,
      "loss": 2.9845,
      "theoretical_loss": 4.026783806026552,
      "tokens_seen": 387055616
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004458074222668004,
      "loss": 3.2016,
      "theoretical_loss": 4.0267071112673,
      "tokens_seen": 387121152
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004457973921765296,
      "loss": 3.0851,
      "theoretical_loss": 4.026630433125394,
      "tokens_seen": 387186688
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004457873620862588,
      "loss": 3.1264,
      "theoretical_loss": 4.026553771594421,
      "tokens_seen": 387252224
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044577733199598795,
      "loss": 3.0453,
      "theoretical_loss": 4.026477126667972,
      "tokens_seen": 387317760
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004457673019057172,
      "loss": 3.2042,
      "theoretical_loss": 4.026400498339642,
      "tokens_seen": 387383296
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044575727181544637,
      "loss": 3.0288,
      "theoretical_loss": 4.026323886603029,
      "tokens_seen": 387448832
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044574724172517555,
      "loss": 3.0314,
      "theoretical_loss": 4.026247291451735,
      "tokens_seen": 387514368
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044573721163490473,
      "loss": 2.9828,
      "theoretical_loss": 4.026170712879368,
      "tokens_seen": 387579904
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004457271815446339,
      "loss": 2.956,
      "theoretical_loss": 4.026094150879533,
      "tokens_seen": 387645440
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004457171514543631,
      "loss": 3.295,
      "theoretical_loss": 4.026017605445846,
      "tokens_seen": 387710976
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004457071213640923,
      "loss": 3.2004,
      "theoretical_loss": 4.025941076571922,
      "tokens_seen": 387776512
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044569709127382145,
      "loss": 3.0723,
      "theoretical_loss": 4.025864564251379,
      "tokens_seen": 387842048
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004456870611835507,
      "loss": 2.9591,
      "theoretical_loss": 4.025788068477843,
      "tokens_seen": 387907584
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004456770310932798,
      "loss": 3.1178,
      "theoretical_loss": 4.025711589244939,
      "tokens_seen": 387973120
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044566700100300905,
      "loss": 3.1414,
      "theoretical_loss": 4.025635126546299,
      "tokens_seen": 388038656
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044565697091273823,
      "loss": 3.0103,
      "theoretical_loss": 4.025558680375555,
      "tokens_seen": 388104192
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004456469408224674,
      "loss": 3.1519,
      "theoretical_loss": 4.025482250726346,
      "tokens_seen": 388169728
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004456369107321966,
      "loss": 2.9792,
      "theoretical_loss": 4.025405837592311,
      "tokens_seen": 388235264
    },
    {
      "epoch": 0.12,
      "objective/train/advantage_avg": 0.4947003126144409,
      "objective/train/docs_used": 231531,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.14778995513916,
      "objective/train/original_loss": 3.147789716720581,
      "objective/train/theoretical_loss": 4.025329440967096,
      "objective/train/tokens_used": 408760800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2448006123304367,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050714373588562,
      "objective/train/weighted_lm_loss": 3.3072593212127686,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0451053380966187,
      "theoretical_loss": 4.025329440967096,
      "tokens_seen": 388300800
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044562688064192583,
      "loss": 3.1125,
      "theoretical_loss": 4.025329440967096,
      "tokens_seen": 388300800
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044561685055165496,
      "loss": 2.985,
      "theoretical_loss": 4.0252530608443475,
      "tokens_seen": 388366336
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004456068204613842,
      "loss": 3.3009,
      "theoretical_loss": 4.025176697217718,
      "tokens_seen": 388431872
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004455967903711133,
      "loss": 3.0362,
      "theoretical_loss": 4.025100350080864,
      "tokens_seen": 388497408
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044558676028084255,
      "loss": 3.179,
      "theoretical_loss": 4.02502401942744,
      "tokens_seen": 388562944
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044557673019057173,
      "loss": 3.4096,
      "theoretical_loss": 4.024947705251112,
      "tokens_seen": 388628480
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004455667001003009,
      "loss": 3.0972,
      "theoretical_loss": 4.024871407545542,
      "tokens_seen": 388694016
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004455566700100301,
      "loss": 3.1309,
      "theoretical_loss": 4.024795126304402,
      "tokens_seen": 388759552
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004455466399197593,
      "loss": 2.9567,
      "theoretical_loss": 4.024718861521362,
      "tokens_seen": 388825088
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044553660982948846,
      "loss": 3.0094,
      "theoretical_loss": 4.0246426131901,
      "tokens_seen": 388890624
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004455265797392177,
      "loss": 3.1179,
      "theoretical_loss": 4.024566381304294,
      "tokens_seen": 388956160
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004455165496489468,
      "loss": 2.9652,
      "theoretical_loss": 4.024490165857627,
      "tokens_seen": 389021696
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044550651955867606,
      "loss": 3.0807,
      "theoretical_loss": 4.024413966843787,
      "tokens_seen": 389087232
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004454964894684052,
      "loss": 3.1547,
      "theoretical_loss": 4.024337784256463,
      "tokens_seen": 389152768
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004454864593781344,
      "loss": 3.0722,
      "theoretical_loss": 4.024261618089348,
      "tokens_seen": 389218304
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004454764292878636,
      "loss": 3.1096,
      "theoretical_loss": 4.024185468336139,
      "tokens_seen": 389283840
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004454663991975928,
      "loss": 2.9247,
      "theoretical_loss": 4.024109334990538,
      "tokens_seen": 389349376
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044545636910732196,
      "loss": 3.1893,
      "theoretical_loss": 4.024033218046245,
      "tokens_seen": 389414912
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004454463390170512,
      "loss": 3.0628,
      "theoretical_loss": 4.023957117496972,
      "tokens_seen": 389480448
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004454363089267803,
      "loss": 3.0764,
      "theoretical_loss": 4.023881033336427,
      "tokens_seen": 389545984
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044542627883650956,
      "loss": 3.3999,
      "theoretical_loss": 4.023804965558325,
      "tokens_seen": 389611520
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004454162487462387,
      "loss": 3.1786,
      "theoretical_loss": 4.023728914156385,
      "tokens_seen": 389677056
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004454062186559679,
      "loss": 3.27,
      "theoretical_loss": 4.023652879124325,
      "tokens_seen": 389742592
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004453961885656971,
      "loss": 3.0981,
      "theoretical_loss": 4.023576860455872,
      "tokens_seen": 389808128
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004453861584754263,
      "loss": 3.1762,
      "theoretical_loss": 4.023500858144755,
      "tokens_seen": 389873664
    },
    {
      "epoch": 0.12,
      "objective/train/advantage_avg": 0.4959947168827057,
      "objective/train/docs_used": 232298,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.812929630279541,
      "objective/train/original_loss": 2.812929391860962,
      "objective/train/theoretical_loss": 4.0234248721847035,
      "objective/train/tokens_used": 410399200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.246047243475914,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.05085027217865,
      "objective/train/weighted_lm_loss": 2.955744981765747,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0454468727111816,
      "theoretical_loss": 4.0234248721847035,
      "tokens_seen": 389939200
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044537612838515547,
      "loss": 3.3229,
      "theoretical_loss": 4.0234248721847035,
      "tokens_seen": 389939200
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044536609829488465,
      "loss": 3.0622,
      "theoretical_loss": 4.023348902569453,
      "tokens_seen": 390004736
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044535606820461383,
      "loss": 3.2226,
      "theoretical_loss": 4.023272949292743,
      "tokens_seen": 390070272
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044534603811434306,
      "loss": 2.9751,
      "theoretical_loss": 4.023197012348315,
      "tokens_seen": 390135808
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004453360080240722,
      "loss": 3.185,
      "theoretical_loss": 4.023121091729914,
      "tokens_seen": 390201344
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004453259779338014,
      "loss": 3.2516,
      "theoretical_loss": 4.02304518743129,
      "tokens_seen": 390266880
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044531594784353055,
      "loss": 3.0652,
      "theoretical_loss": 4.022969299446194,
      "tokens_seen": 390332416
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004453059177532598,
      "loss": 3.3911,
      "theoretical_loss": 4.022893427768381,
      "tokens_seen": 390397952
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044529588766298897,
      "loss": 3.2018,
      "theoretical_loss": 4.022817572391613,
      "tokens_seen": 390463488
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044528585757271815,
      "loss": 2.8983,
      "theoretical_loss": 4.02274173330965,
      "tokens_seen": 390529024
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044527582748244733,
      "loss": 3.0102,
      "theoretical_loss": 4.02266591051626,
      "tokens_seen": 390594560
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044526579739217657,
      "loss": 3.3354,
      "theoretical_loss": 4.022590104005211,
      "tokens_seen": 390660096
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004452557673019057,
      "loss": 3.0675,
      "theoretical_loss": 4.022514313770276,
      "tokens_seen": 390725632
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044524573721163493,
      "loss": 2.9984,
      "theoretical_loss": 4.022438539805231,
      "tokens_seen": 390791168
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044523570712136406,
      "loss": 3.0306,
      "theoretical_loss": 4.0223627821038574,
      "tokens_seen": 390856704
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004452256770310933,
      "loss": 3.2791,
      "theoretical_loss": 4.022287040659938,
      "tokens_seen": 390922240
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044521564694082247,
      "loss": 3.0098,
      "theoretical_loss": 4.022211315467258,
      "tokens_seen": 390987776
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044520561685055165,
      "loss": 3.2095,
      "theoretical_loss": 4.022135606519608,
      "tokens_seen": 391053312
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044519558676028083,
      "loss": 3.1438,
      "theoretical_loss": 4.022059913810782,
      "tokens_seen": 391118848
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044518555667001,
      "loss": 3.0851,
      "theoretical_loss": 4.021984237334577,
      "tokens_seen": 391184384
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044517552657973925,
      "loss": 2.9927,
      "theoretical_loss": 4.021908577084791,
      "tokens_seen": 391249920
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044516549648946843,
      "loss": 3.037,
      "theoretical_loss": 4.021832933055231,
      "tokens_seen": 391315456
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004451554663991976,
      "loss": 3.0246,
      "theoretical_loss": 4.021757305239702,
      "tokens_seen": 391380992
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004451454363089268,
      "loss": 3.028,
      "theoretical_loss": 4.021681693632015,
      "tokens_seen": 391446528
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044513540621865603,
      "loss": 2.8715,
      "theoretical_loss": 4.021606098225982,
      "tokens_seen": 391512064
    },
    {
      "epoch": 0.12,
      "objective/train/advantage_avg": 0.49438780546188354,
      "objective/train/docs_used": 233436,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.9921693801879883,
      "objective/train/original_loss": 2.9921693801879883,
      "objective/train/theoretical_loss": 4.021530519015424,
      "objective/train/tokens_used": 412037600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24447834491729736,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050681471824646,
      "objective/train/weighted_lm_loss": 3.143890142440796,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0432155132293701,
      "theoretical_loss": 4.021530519015424,
      "tokens_seen": 391577600
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044512537612838516,
      "loss": 3.1175,
      "theoretical_loss": 4.021530519015424,
      "tokens_seen": 391577600
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004451153460381144,
      "loss": 2.9554,
      "theoretical_loss": 4.021454955994159,
      "tokens_seen": 391643136
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004451053159478435,
      "loss": 3.0077,
      "theoretical_loss": 4.02137940915601,
      "tokens_seen": 391708672
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044509528585757275,
      "loss": 3.1965,
      "theoretical_loss": 4.021303878494807,
      "tokens_seen": 391774208
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044508525576730194,
      "loss": 2.9657,
      "theoretical_loss": 4.02122836400438,
      "tokens_seen": 391839744
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004450752256770311,
      "loss": 3.0201,
      "theoretical_loss": 4.021152865678563,
      "tokens_seen": 391905280
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004450651955867603,
      "loss": 3.2405,
      "theoretical_loss": 4.0210773835111935,
      "tokens_seen": 391970816
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004450551654964895,
      "loss": 3.2132,
      "theoretical_loss": 4.021001917496112,
      "tokens_seen": 392036352
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044504513540621866,
      "loss": 2.8934,
      "theoretical_loss": 4.020926467627164,
      "tokens_seen": 392101888
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004450351053159479,
      "loss": 2.9649,
      "theoretical_loss": 4.020851033898196,
      "tokens_seen": 392167424
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.000445025075225677,
      "loss": 3.2253,
      "theoretical_loss": 4.020775616303059,
      "tokens_seen": 392232960
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044501504513540626,
      "loss": 3.1816,
      "theoretical_loss": 4.020700214835609,
      "tokens_seen": 392298496
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004450050150451354,
      "loss": 3.2582,
      "theoretical_loss": 4.020624829489702,
      "tokens_seen": 392364032
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004449949849548646,
      "loss": 3.237,
      "theoretical_loss": 4.0205494602592,
      "tokens_seen": 392429568
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004449849548645938,
      "loss": 2.8602,
      "theoretical_loss": 4.020474107137967,
      "tokens_seen": 392495104
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.000444974924774323,
      "loss": 2.7178,
      "theoretical_loss": 4.020398770119872,
      "tokens_seen": 392560640
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044496489468405216,
      "loss": 3.213,
      "theoretical_loss": 4.020323449198785,
      "tokens_seen": 392626176
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004449548645937814,
      "loss": 3.057,
      "theoretical_loss": 4.020248144368582,
      "tokens_seen": 392691712
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004449448345035105,
      "loss": 3.2086,
      "theoretical_loss": 4.020172855623139,
      "tokens_seen": 392757248
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044493480441323976,
      "loss": 3.0973,
      "theoretical_loss": 4.02009758295634,
      "tokens_seen": 392822784
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004449247743229689,
      "loss": 3.1001,
      "theoretical_loss": 4.020022326362067,
      "tokens_seen": 392888320
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004449147442326981,
      "loss": 2.8606,
      "theoretical_loss": 4.019947085834209,
      "tokens_seen": 392953856
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004449047141424273,
      "loss": 3.3349,
      "theoretical_loss": 4.019871861366658,
      "tokens_seen": 393019392
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004448946840521565,
      "loss": 3.0303,
      "theoretical_loss": 4.019796652953309,
      "tokens_seen": 393084928
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044488465396188567,
      "loss": 3.0786,
      "theoretical_loss": 4.019721460588057,
      "tokens_seen": 393150464
    },
    {
      "epoch": 0.12,
      "objective/train/advantage_avg": 0.4940994083881378,
      "objective/train/docs_used": 234056,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.9761195182800293,
      "objective/train/original_loss": 2.97611927986145,
      "objective/train/theoretical_loss": 4.019646284264807,
      "objective/train/tokens_used": 413676000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24418498575687408,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506511926651,
      "objective/train/weighted_lm_loss": 3.1266849040985107,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0451053380966187,
      "theoretical_loss": 4.019646284264807,
      "tokens_seen": 393216000
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044487462387161485,
      "loss": 3.108,
      "theoretical_loss": 4.019646284264807,
      "tokens_seen": 393216000
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044486459378134403,
      "loss": 3.1025,
      "theoretical_loss": 4.019571123977462,
      "tokens_seen": 393281536
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044485456369107326,
      "loss": 2.8188,
      "theoretical_loss": 4.01949597971993,
      "tokens_seen": 393347072
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004448445336008024,
      "loss": 3.0893,
      "theoretical_loss": 4.019420851486123,
      "tokens_seen": 393412608
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004448345035105316,
      "loss": 2.9829,
      "theoretical_loss": 4.019345739269955,
      "tokens_seen": 393478144
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044482447342026075,
      "loss": 2.9409,
      "theoretical_loss": 4.019270643065344,
      "tokens_seen": 393543680
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044481444332999,
      "loss": 3.0281,
      "theoretical_loss": 4.019195562866211,
      "tokens_seen": 393609216
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044480441323971917,
      "loss": 3.2024,
      "theoretical_loss": 4.0191204986664815,
      "tokens_seen": 393674752
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044479438314944835,
      "loss": 3.0459,
      "theoretical_loss": 4.019045450460084,
      "tokens_seen": 393740288
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044478435305917753,
      "loss": 3.0642,
      "theoretical_loss": 4.018970418240949,
      "tokens_seen": 393805824
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044477432296890677,
      "loss": 2.8814,
      "theoretical_loss": 4.01889540200301,
      "tokens_seen": 393871360
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004447642928786359,
      "loss": 3.0009,
      "theoretical_loss": 4.018820401740206,
      "tokens_seen": 393936896
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044475426278836513,
      "loss": 3.2329,
      "theoretical_loss": 4.018745417446478,
      "tokens_seen": 394002432
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044474423269809426,
      "loss": 2.9658,
      "theoretical_loss": 4.018670449115772,
      "tokens_seen": 394067968
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004447342026078235,
      "loss": 3.0116,
      "theoretical_loss": 4.018595496742034,
      "tokens_seen": 394133504
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044472417251755267,
      "loss": 3.0666,
      "theoretical_loss": 4.018520560319215,
      "tokens_seen": 394199040
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044471414242728185,
      "loss": 3.0647,
      "theoretical_loss": 4.01844563984127,
      "tokens_seen": 394264576
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044470411233701103,
      "loss": 3.3439,
      "theoretical_loss": 4.018370735302156,
      "tokens_seen": 394330112
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004446940822467402,
      "loss": 3.0702,
      "theoretical_loss": 4.018295846695835,
      "tokens_seen": 394395648
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004446840521564694,
      "loss": 3.271,
      "theoretical_loss": 4.0182209740162715,
      "tokens_seen": 394461184
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044467402206619863,
      "loss": 3.2658,
      "theoretical_loss": 4.018146117257433,
      "tokens_seen": 394526720
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044466399197592776,
      "loss": 3.0501,
      "theoretical_loss": 4.0180712764132895,
      "tokens_seen": 394592256
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.000444653961885657,
      "loss": 3.1232,
      "theoretical_loss": 4.017996451477815,
      "tokens_seen": 394657792
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004446439317953861,
      "loss": 3.1932,
      "theoretical_loss": 4.017921642444987,
      "tokens_seen": 394723328
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044463390170511536,
      "loss": 2.9553,
      "theoretical_loss": 4.017846849308788,
      "tokens_seen": 394788864
    },
    {
      "epoch": 0.12,
      "objective/train/advantage_avg": 0.49314919114112854,
      "objective/train/docs_used": 235431,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.150310516357422,
      "objective/train/original_loss": 3.150310516357422,
      "objective/train/theoretical_loss": 4.017772072063201,
      "objective/train/tokens_used": 415314400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24325567483901978,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050551414489746,
      "objective/train/weighted_lm_loss": 3.3096566200256348,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0430042743682861,
      "theoretical_loss": 4.017772072063201,
      "tokens_seen": 394854400
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044462387161484454,
      "loss": 3.1103,
      "theoretical_loss": 4.017772072063201,
      "tokens_seen": 394854400
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004446138415245737,
      "loss": 3.1133,
      "theoretical_loss": 4.017697310702211,
      "tokens_seen": 394919936
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004446038114343029,
      "loss": 2.8195,
      "theoretical_loss": 4.017622565219812,
      "tokens_seen": 394985472
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044459378134403214,
      "loss": 3.0114,
      "theoretical_loss": 4.017547835609996,
      "tokens_seen": 395051008
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044458375125376126,
      "loss": 3.0815,
      "theoretical_loss": 4.017473121866759,
      "tokens_seen": 395116544
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004445737211634905,
      "loss": 3.0354,
      "theoretical_loss": 4.017398423984105,
      "tokens_seen": 395182080
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004445636910732196,
      "loss": 2.8032,
      "theoretical_loss": 4.017323741956034,
      "tokens_seen": 395247616
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044455366098294886,
      "loss": 2.9871,
      "theoretical_loss": 4.0172490757765535,
      "tokens_seen": 395313152
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044454363089267804,
      "loss": 3.3387,
      "theoretical_loss": 4.017174425439675,
      "tokens_seen": 395378688
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004445336008024072,
      "loss": 2.8858,
      "theoretical_loss": 4.017099790939412,
      "tokens_seen": 395444224
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004445235707121364,
      "loss": 3.1696,
      "theoretical_loss": 4.017025172269779,
      "tokens_seen": 395509760
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004445135406218656,
      "loss": 3.3217,
      "theoretical_loss": 4.016950569424798,
      "tokens_seen": 395575296
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044450351053159477,
      "loss": 3.1956,
      "theoretical_loss": 4.01687598239849,
      "tokens_seen": 395640832
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.000444493480441324,
      "loss": 3.0031,
      "theoretical_loss": 4.016801411184884,
      "tokens_seen": 395706368
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044448345035105313,
      "loss": 2.9959,
      "theoretical_loss": 4.016726855778008,
      "tokens_seen": 395771904
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044447342026078236,
      "loss": 2.9162,
      "theoretical_loss": 4.0166523161718946,
      "tokens_seen": 395837440
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004444633901705115,
      "loss": 2.8054,
      "theoretical_loss": 4.0165777923605805,
      "tokens_seen": 395902976
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004444533600802407,
      "loss": 3.0545,
      "theoretical_loss": 4.016503284338105,
      "tokens_seen": 395968512
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044444332998996996,
      "loss": 3.0282,
      "theoretical_loss": 4.016428792098512,
      "tokens_seen": 396034048
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004444332998996991,
      "loss": 2.8985,
      "theoretical_loss": 4.016354315635844,
      "tokens_seen": 396099584
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004444232698094283,
      "loss": 2.6332,
      "theoretical_loss": 4.016279854944154,
      "tokens_seen": 396165120
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004444132397191575,
      "loss": 2.9978,
      "theoretical_loss": 4.016205410017491,
      "tokens_seen": 396230656
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004444032096288867,
      "loss": 3.1173,
      "theoretical_loss": 4.016130980849912,
      "tokens_seen": 396296192
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044439317953861587,
      "loss": 2.9293,
      "theoretical_loss": 4.016056567435475,
      "tokens_seen": 396361728
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044438314944834505,
      "loss": 3.0219,
      "theoretical_loss": 4.015982169768244,
      "tokens_seen": 396427264
    },
    {
      "epoch": 0.12,
      "objective/train/advantage_avg": 0.49297916889190674,
      "objective/train/docs_used": 235840,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.5858728885650635,
      "objective/train/original_loss": 2.5858728885650635,
      "objective/train/theoretical_loss": 4.0159077878422815,
      "objective/train/tokens_used": 416952800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24317675828933716,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505340099334717,
      "objective/train/weighted_lm_loss": 2.7163543701171875,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0304546356201172,
      "theoretical_loss": 4.0159077878422815,
      "tokens_seen": 396492800
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044437311935807423,
      "loss": 2.8351,
      "theoretical_loss": 4.0159077878422815,
      "tokens_seen": 396492800
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044436308926780346,
      "loss": 3.0958,
      "theoretical_loss": 4.015833421651657,
      "tokens_seen": 396558336
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004443530591775326,
      "loss": 2.8946,
      "theoretical_loss": 4.015759071190443,
      "tokens_seen": 396623872
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004443430290872618,
      "loss": 3.0527,
      "theoretical_loss": 4.015684736452712,
      "tokens_seen": 396689408
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044433299899699095,
      "loss": 3.3742,
      "theoretical_loss": 4.015610417432544,
      "tokens_seen": 396754944
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004443229689067202,
      "loss": 2.8225,
      "theoretical_loss": 4.0155361141240205,
      "tokens_seen": 396820480
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044431293881644937,
      "loss": 2.9552,
      "theoretical_loss": 4.015461826521225,
      "tokens_seen": 396886016
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044430290872617855,
      "loss": 3.2712,
      "theoretical_loss": 4.015387554618244,
      "tokens_seen": 396951552
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044429287863590773,
      "loss": 3.0589,
      "theoretical_loss": 4.01531329840917,
      "tokens_seen": 397017088
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044428284854563697,
      "loss": 3.1769,
      "theoretical_loss": 4.015239057888097,
      "tokens_seen": 397082624
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004442728184553661,
      "loss": 3.2234,
      "theoretical_loss": 4.015164833049122,
      "tokens_seen": 397148160
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044426278836509533,
      "loss": 3.0318,
      "theoretical_loss": 4.015090623886345,
      "tokens_seen": 397213696
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044425275827482446,
      "loss": 2.8318,
      "theoretical_loss": 4.01501643039387,
      "tokens_seen": 397279232
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004442427281845537,
      "loss": 3.0799,
      "theoretical_loss": 4.014942252565803,
      "tokens_seen": 397344768
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044423269809428287,
      "loss": 3.2281,
      "theoretical_loss": 4.014868090396256,
      "tokens_seen": 397410304
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044422266800401205,
      "loss": 2.946,
      "theoretical_loss": 4.01479394387934,
      "tokens_seen": 397475840
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044421263791374123,
      "loss": 3.0879,
      "theoretical_loss": 4.014719813009172,
      "tokens_seen": 397541376
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004442026078234704,
      "loss": 3.0367,
      "theoretical_loss": 4.014645697779872,
      "tokens_seen": 397606912
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004441925777331996,
      "loss": 3.0768,
      "theoretical_loss": 4.014571598185563,
      "tokens_seen": 397672448
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044418254764292883,
      "loss": 2.967,
      "theoretical_loss": 4.01449751422037,
      "tokens_seen": 397737984
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044417251755265796,
      "loss": 2.8913,
      "theoretical_loss": 4.014423445878423,
      "tokens_seen": 397803520
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004441624874623872,
      "loss": 3.1157,
      "theoretical_loss": 4.014349393153853,
      "tokens_seen": 397869056
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004441524573721163,
      "loss": 2.9808,
      "theoretical_loss": 4.014275356040796,
      "tokens_seen": 397934592
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044414242728184556,
      "loss": 3.164,
      "theoretical_loss": 4.014201334533391,
      "tokens_seen": 398000128
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044413239719157474,
      "loss": 2.9506,
      "theoretical_loss": 4.01412732862578,
      "tokens_seen": 398065664
    },
    {
      "epoch": 0.12,
      "objective/train/advantage_avg": 0.4954759180545807,
      "objective/train/docs_used": 236468,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.685086965560913,
      "objective/train/original_loss": 2.685086250305176,
      "objective/train/theoretical_loss": 4.014053338312108,
      "objective/train/tokens_used": 418591200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24554702639579773,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507957935333252,
      "objective/train/weighted_lm_loss": 2.8212814331054688,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0451053380966187,
      "theoretical_loss": 4.014053338312108,
      "tokens_seen": 398131200
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004441223671013039,
      "loss": 3.062,
      "theoretical_loss": 4.014053338312108,
      "tokens_seen": 398131200
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004441123370110331,
      "loss": 2.8257,
      "theoretical_loss": 4.013979363586522,
      "tokens_seen": 398196736
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044410230692076234,
      "loss": 3.0745,
      "theoretical_loss": 4.013905404443173,
      "tokens_seen": 398262272
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044409227683049146,
      "loss": 3.1875,
      "theoretical_loss": 4.013831460876216,
      "tokens_seen": 398327808
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004440822467402207,
      "loss": 2.955,
      "theoretical_loss": 4.01375753287981,
      "tokens_seen": 398393344
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004440722166499498,
      "loss": 3.2996,
      "theoretical_loss": 4.013683620448113,
      "tokens_seen": 398458880
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044406218655967906,
      "loss": 3.1041,
      "theoretical_loss": 4.013609723575291,
      "tokens_seen": 398524416
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044405215646940824,
      "loss": 3.2029,
      "theoretical_loss": 4.01353584225551,
      "tokens_seen": 398589952
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004440421263791374,
      "loss": 3.2532,
      "theoretical_loss": 4.01346197648294,
      "tokens_seen": 398655488
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004440320962888666,
      "loss": 2.9971,
      "theoretical_loss": 4.013388126251755,
      "tokens_seen": 398721024
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004440220661985958,
      "loss": 3.0983,
      "theoretical_loss": 4.013314291556131,
      "tokens_seen": 398786560
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044401203610832497,
      "loss": 3.3063,
      "theoretical_loss": 4.0132404723902475,
      "tokens_seen": 398852096
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004440020060180542,
      "loss": 3.2002,
      "theoretical_loss": 4.013166668748288,
      "tokens_seen": 398917632
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044399197592778333,
      "loss": 3.151,
      "theoretical_loss": 4.013092880624437,
      "tokens_seen": 398983168
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044398194583751256,
      "loss": 3.0523,
      "theoretical_loss": 4.013019108012884,
      "tokens_seen": 399048704
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004439719157472417,
      "loss": 3.2369,
      "theoretical_loss": 4.012945350907822,
      "tokens_seen": 399114240
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004439618856569709,
      "loss": 3.2228,
      "theoretical_loss": 4.012871609303445,
      "tokens_seen": 399179776
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004439518555667001,
      "loss": 3.0588,
      "theoretical_loss": 4.012797883193952,
      "tokens_seen": 399245312
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004439418254764293,
      "loss": 3.2139,
      "theoretical_loss": 4.012724172573543,
      "tokens_seen": 399310848
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044393179538615847,
      "loss": 3.2144,
      "theoretical_loss": 4.012650477436425,
      "tokens_seen": 399376384
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004439217652958877,
      "loss": 3.1901,
      "theoretical_loss": 4.012576797776805,
      "tokens_seen": 399441920
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044391173520561683,
      "loss": 3.2001,
      "theoretical_loss": 4.0125031335888925,
      "tokens_seen": 399507456
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044390170511534607,
      "loss": 3.1418,
      "theoretical_loss": 4.012429484866903,
      "tokens_seen": 399572992
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004438916750250752,
      "loss": 3.1132,
      "theoretical_loss": 4.012355851605053,
      "tokens_seen": 399638528
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044388164493480443,
      "loss": 3.1453,
      "theoretical_loss": 4.012282233797562,
      "tokens_seen": 399704064
    },
    {
      "epoch": 0.12,
      "objective/train/advantage_avg": 0.49458757042884827,
      "objective/train/docs_used": 237759,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.6130058765411377,
      "objective/train/original_loss": 2.6130053997039795,
      "objective/train/theoretical_loss": 4.0122086314386545,
      "objective/train/tokens_used": 420229600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2446780800819397,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050702452659607,
      "objective/train/weighted_lm_loss": 2.7455286979675293,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0402631759643555,
      "theoretical_loss": 4.0122086314386545,
      "tokens_seen": 399769600
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004438716148445336,
      "loss": 3.0978,
      "theoretical_loss": 4.0122086314386545,
      "tokens_seen": 399769600
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004438615847542628,
      "loss": 3.0312,
      "theoretical_loss": 4.012135044522555,
      "tokens_seen": 399835136
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044385155466399197,
      "loss": 3.0551,
      "theoretical_loss": 4.012061473043495,
      "tokens_seen": 399900672
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044384152457372115,
      "loss": 2.921,
      "theoretical_loss": 4.0119879169957064,
      "tokens_seen": 399966208
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044383149448345033,
      "loss": 3.0326,
      "theoretical_loss": 4.011914376373426,
      "tokens_seen": 400031744
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044382146439317957,
      "loss": 3.1473,
      "theoretical_loss": 4.01184085117089,
      "tokens_seen": 400097280
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004438114343029087,
      "loss": 3.1315,
      "theoretical_loss": 4.011767341382341,
      "tokens_seen": 400162816
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044380140421263793,
      "loss": 3.0605,
      "theoretical_loss": 4.011693847002025,
      "tokens_seen": 400228352
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004437913741223671,
      "loss": 3.1508,
      "theoretical_loss": 4.011620368024191,
      "tokens_seen": 400293888
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004437813440320963,
      "loss": 2.8331,
      "theoretical_loss": 4.011546904443088,
      "tokens_seen": 400359424
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004437713139418255,
      "loss": 2.9113,
      "theoretical_loss": 4.011473456252971,
      "tokens_seen": 400424960
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044376128385155466,
      "loss": 2.9137,
      "theoretical_loss": 4.011400023448097,
      "tokens_seen": 400490496
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044375125376128384,
      "loss": 2.9356,
      "theoretical_loss": 4.0113266060227275,
      "tokens_seen": 400556032
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044374122367101307,
      "loss": 2.9422,
      "theoretical_loss": 4.011253203971124,
      "tokens_seen": 400621568
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004437311935807422,
      "loss": 3.0348,
      "theoretical_loss": 4.011179817287556,
      "tokens_seen": 400687104
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044372116349047143,
      "loss": 2.9257,
      "theoretical_loss": 4.011106445966291,
      "tokens_seen": 400752640
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044371113340020056,
      "loss": 3.1406,
      "theoretical_loss": 4.011033090001602,
      "tokens_seen": 400818176
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004437011033099298,
      "loss": 3.1664,
      "theoretical_loss": 4.010959749387765,
      "tokens_seen": 400883712
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044369107321965903,
      "loss": 3.2752,
      "theoretical_loss": 4.010886424119059,
      "tokens_seen": 400949248
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044368104312938816,
      "loss": 2.994,
      "theoretical_loss": 4.010813114189765,
      "tokens_seen": 401014784
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004436710130391174,
      "loss": 2.6771,
      "theoretical_loss": 4.0107398195941695,
      "tokens_seen": 401080320
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004436609829488465,
      "loss": 3.1621,
      "theoretical_loss": 4.010666540326559,
      "tokens_seen": 401145856
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044365095285857576,
      "loss": 3.1048,
      "theoretical_loss": 4.010593276381226,
      "tokens_seen": 401211392
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044364092276830494,
      "loss": 3.0339,
      "theoretical_loss": 4.010520027752464,
      "tokens_seen": 401276928
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004436308926780341,
      "loss": 3.2648,
      "theoretical_loss": 4.01044679443457,
      "tokens_seen": 401342464
    },
    {
      "epoch": 0.12,
      "objective/train/advantage_avg": 0.4960605502128601,
      "objective/train/docs_used": 238664,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.6084442138671875,
      "objective/train/original_loss": 2.6084442138671875,
      "objective/train/theoretical_loss": 4.010373576421845,
      "objective/train/tokens_used": 421868000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24610836803913116,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0508571863174438,
      "objective/train/weighted_lm_loss": 2.741203546524048,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.044721245765686,
      "theoretical_loss": 4.010373576421845,
      "tokens_seen": 401408000
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004436208625877633,
      "loss": 3.079,
      "theoretical_loss": 4.010373576421845,
      "tokens_seen": 401408000
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044361083249749254,
      "loss": 3.2382,
      "theoretical_loss": 4.010300373708591,
      "tokens_seen": 401473536
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044360080240722166,
      "loss": 3.0485,
      "theoretical_loss": 4.010227186289116,
      "tokens_seen": 401539072
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004435907723169509,
      "loss": 2.7664,
      "theoretical_loss": 4.010154014157727,
      "tokens_seen": 401604608
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044358074222668,
      "loss": 3.0878,
      "theoretical_loss": 4.010080857308741,
      "tokens_seen": 401670144
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044357071213640926,
      "loss": 3.1958,
      "theoretical_loss": 4.010007715736468,
      "tokens_seen": 401735680
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044356068204613844,
      "loss": 2.9678,
      "theoretical_loss": 4.009934589435231,
      "tokens_seen": 401801216
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004435506519558676,
      "loss": 3.0909,
      "theoretical_loss": 4.009861478399348,
      "tokens_seen": 401866752
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004435406218655968,
      "loss": 3.0651,
      "theoretical_loss": 4.009788382623146,
      "tokens_seen": 401932288
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.000443530591775326,
      "loss": 2.91,
      "theoretical_loss": 4.009715302100952,
      "tokens_seen": 401997824
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044352056168505517,
      "loss": 2.7633,
      "theoretical_loss": 4.009642236827098,
      "tokens_seen": 402063360
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004435105315947844,
      "loss": 3.132,
      "theoretical_loss": 4.009569186795915,
      "tokens_seen": 402128896
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044350050150451353,
      "loss": 3.3941,
      "theoretical_loss": 4.009496152001741,
      "tokens_seen": 402194432
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044349047141424276,
      "loss": 2.857,
      "theoretical_loss": 4.009423132438918,
      "tokens_seen": 402259968
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004434804413239719,
      "loss": 3.1942,
      "theoretical_loss": 4.009350128101785,
      "tokens_seen": 402325504
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004434704112337011,
      "loss": 3.0485,
      "theoretical_loss": 4.00927713898469,
      "tokens_seen": 402391040
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004434603811434303,
      "loss": 3.2654,
      "theoretical_loss": 4.009204165081981,
      "tokens_seen": 402456576
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004434503510531595,
      "loss": 3.1701,
      "theoretical_loss": 4.009131206388012,
      "tokens_seen": 402522112
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044344032096288867,
      "loss": 3.1346,
      "theoretical_loss": 4.009058262897135,
      "tokens_seen": 402587648
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004434302908726179,
      "loss": 3.1479,
      "theoretical_loss": 4.008985334603709,
      "tokens_seen": 402653184
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044342026078234703,
      "loss": 3.1037,
      "theoretical_loss": 4.008912421502095,
      "tokens_seen": 402718720
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044341023069207627,
      "loss": 2.9289,
      "theoretical_loss": 4.008839523586658,
      "tokens_seen": 402784256
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004434002006018054,
      "loss": 3.1108,
      "theoretical_loss": 4.008766640851763,
      "tokens_seen": 402849792
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044339017051153463,
      "loss": 2.9849,
      "theoretical_loss": 4.008693773291781,
      "tokens_seen": 402915328
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004433801404212638,
      "loss": 2.9243,
      "theoretical_loss": 4.008620920901085,
      "tokens_seen": 402980864
    },
    {
      "epoch": 0.12,
      "objective/train/advantage_avg": 0.49536532163619995,
      "objective/train/docs_used": 240054,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.5082550048828125,
      "objective/train/original_loss": 3.5082545280456543,
      "objective/train/theoretical_loss": 4.00854808367405,
      "objective/train/tokens_used": 423506400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24545978009700775,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507843494415283,
      "objective/train/weighted_lm_loss": 3.6864986419677734,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0304546356201172,
      "theoretical_loss": 4.00854808367405,
      "tokens_seen": 403046400
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.000443370110330993,
      "loss": 3.153,
      "theoretical_loss": 4.00854808367405,
      "tokens_seen": 403046400
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044336008024072217,
      "loss": 2.9397,
      "theoretical_loss": 4.0084752616050565,
      "tokens_seen": 403111936
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044335005015045135,
      "loss": 2.8052,
      "theoretical_loss": 4.008402454688485,
      "tokens_seen": 403177472
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044334002006018053,
      "loss": 3.2022,
      "theoretical_loss": 4.008329662918721,
      "tokens_seen": 403243008
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044332998996990977,
      "loss": 3.2911,
      "theoretical_loss": 4.008256886290153,
      "tokens_seen": 403308544
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004433199598796389,
      "loss": 3.1362,
      "theoretical_loss": 4.00818412479717,
      "tokens_seen": 403374080
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044330992978936813,
      "loss": 3.098,
      "theoretical_loss": 4.008111378434169,
      "tokens_seen": 403439616
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004432998996990973,
      "loss": 2.9772,
      "theoretical_loss": 4.008038647195543,
      "tokens_seen": 403505152
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004432898696088265,
      "loss": 2.9132,
      "theoretical_loss": 4.007965931075695,
      "tokens_seen": 403570688
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004432798395185557,
      "loss": 3.0321,
      "theoretical_loss": 4.007893230069027,
      "tokens_seen": 403636224
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044326980942828486,
      "loss": 3.1677,
      "theoretical_loss": 4.007820544169944,
      "tokens_seen": 403701760
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044325977933801404,
      "loss": 3.2688,
      "theoretical_loss": 4.007747873372857,
      "tokens_seen": 403767296
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044324974924774327,
      "loss": 3.1068,
      "theoretical_loss": 4.007675217672174,
      "tokens_seen": 403832832
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004432397191574724,
      "loss": 3.0687,
      "theoretical_loss": 4.007602577062314,
      "tokens_seen": 403898368
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044322968906720163,
      "loss": 3.288,
      "theoretical_loss": 4.007529951537692,
      "tokens_seen": 403963904
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044321965897693076,
      "loss": 2.8574,
      "theoretical_loss": 4.00745734109273,
      "tokens_seen": 404029440
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044320962888666,
      "loss": 2.9454,
      "theoretical_loss": 4.0073847457218505,
      "tokens_seen": 404094976
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004431995987963892,
      "loss": 3.1665,
      "theoretical_loss": 4.0073121654194805,
      "tokens_seen": 404160512
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044318956870611836,
      "loss": 3.0799,
      "theoretical_loss": 4.00723960018005,
      "tokens_seen": 404226048
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044317953861584754,
      "loss": 3.1384,
      "theoretical_loss": 4.007167049997992,
      "tokens_seen": 404291584
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004431695085255767,
      "loss": 2.9752,
      "theoretical_loss": 4.007094514867741,
      "tokens_seen": 404357120
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004431594784353059,
      "loss": 2.985,
      "theoretical_loss": 4.007021994783736,
      "tokens_seen": 404422656
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044314944834503514,
      "loss": 3.1768,
      "theoretical_loss": 4.0069494897404185,
      "tokens_seen": 404488192
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044313941825476427,
      "loss": 2.6251,
      "theoretical_loss": 4.006876999732232,
      "tokens_seen": 404553728
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004431293881644935,
      "loss": 3.2001,
      "theoretical_loss": 4.006804524753626,
      "tokens_seen": 404619264
    },
    {
      "epoch": 0.12,
      "objective/train/advantage_avg": 0.49555790424346924,
      "objective/train/docs_used": 240788,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.656939744949341,
      "objective/train/original_loss": 2.65693998336792,
      "objective/train/theoretical_loss": 4.006732064799048,
      "objective/train/tokens_used": 425144800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24564804136753082,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050804615020752,
      "objective/train/weighted_lm_loss": 2.791757106781006,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0425469875335693,
      "theoretical_loss": 4.006732064799048,
      "tokens_seen": 404684800
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004431193580742227,
      "loss": 3.0714,
      "theoretical_loss": 4.006732064799048,
      "tokens_seen": 404684800
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044310932798395186,
      "loss": 3.1978,
      "theoretical_loss": 4.006659619862954,
      "tokens_seen": 404750336
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044309929789368104,
      "loss": 2.9539,
      "theoretical_loss": 4.006587189939798,
      "tokens_seen": 404815872
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004430892678034102,
      "loss": 3.0397,
      "theoretical_loss": 4.00651477502404,
      "tokens_seen": 404881408
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004430792377131394,
      "loss": 2.7681,
      "theoretical_loss": 4.006442375110142,
      "tokens_seen": 404946944
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044306920762286864,
      "loss": 2.9498,
      "theoretical_loss": 4.006369990192569,
      "tokens_seen": 405012480
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044305917753259777,
      "loss": 2.95,
      "theoretical_loss": 4.00629762026579,
      "tokens_seen": 405078016
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.000443049147442327,
      "loss": 3.1909,
      "theoretical_loss": 4.006225265324274,
      "tokens_seen": 405143552
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044303911735205613,
      "loss": 3.0607,
      "theoretical_loss": 4.006152925362496,
      "tokens_seen": 405209088
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044302908726178537,
      "loss": 3.1247,
      "theoretical_loss": 4.006080600374932,
      "tokens_seen": 405274624
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044301905717151455,
      "loss": 3.1435,
      "theoretical_loss": 4.006008290356062,
      "tokens_seen": 405340160
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044300902708124373,
      "loss": 3.0826,
      "theoretical_loss": 4.005935995300369,
      "tokens_seen": 405405696
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004429989969909729,
      "loss": 3.0564,
      "theoretical_loss": 4.005863715202339,
      "tokens_seen": 405471232
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004429889669007021,
      "loss": 3.3518,
      "theoretical_loss": 4.005791450056459,
      "tokens_seen": 405536768
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044297893681043127,
      "loss": 3.0993,
      "theoretical_loss": 4.0057191998572215,
      "tokens_seen": 405602304
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004429689067201605,
      "loss": 3.0658,
      "theoretical_loss": 4.005646964599119,
      "tokens_seen": 405667840
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044295887662988963,
      "loss": 3.2428,
      "theoretical_loss": 4.005574744276652,
      "tokens_seen": 405733376
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044294884653961887,
      "loss": 2.9911,
      "theoretical_loss": 4.0055025388843175,
      "tokens_seen": 405798912
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004429388164493481,
      "loss": 3.0609,
      "theoretical_loss": 4.0054303484166205,
      "tokens_seen": 405864448
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044292878635907723,
      "loss": 3.0099,
      "theoretical_loss": 4.005358172868066,
      "tokens_seen": 405929984
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044291875626880647,
      "loss": 2.9321,
      "theoretical_loss": 4.005286012233162,
      "tokens_seen": 405995520
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004429087261785356,
      "loss": 3.3901,
      "theoretical_loss": 4.0052138665064225,
      "tokens_seen": 406061056
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044289869608826483,
      "loss": 3.1055,
      "theoretical_loss": 4.00514173568236,
      "tokens_seen": 406126592
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.000442888665997994,
      "loss": 3.0439,
      "theoretical_loss": 4.005069619755494,
      "tokens_seen": 406192128
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004428786359077232,
      "loss": 2.7945,
      "theoretical_loss": 4.004997518720344,
      "tokens_seen": 406257664
    },
    {
      "epoch": 0.12,
      "objective/train/advantage_avg": 0.49491509795188904,
      "objective/train/docs_used": 241796,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.063617467880249,
      "objective/train/original_loss": 3.063617467880249,
      "objective/train/theoretical_loss": 4.004925432571433,
      "objective/train/tokens_used": 426783200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2449880689382553,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050736904144287,
      "objective/train/weighted_lm_loss": 3.219026803970337,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0304546356201172,
      "theoretical_loss": 4.004925432571433,
      "tokens_seen": 406323200
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044286860581745237,
      "loss": 2.8935,
      "theoretical_loss": 4.004925432571433,
      "tokens_seen": 406323200
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044285857572718155,
      "loss": 2.756,
      "theoretical_loss": 4.004853361303288,
      "tokens_seen": 406388736
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044284854563691073,
      "loss": 2.9063,
      "theoretical_loss": 4.004781304910438,
      "tokens_seen": 406454272
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044283851554663997,
      "loss": 2.8685,
      "theoretical_loss": 4.004709263387414,
      "tokens_seen": 406519808
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004428284854563691,
      "loss": 2.9412,
      "theoretical_loss": 4.004637236728753,
      "tokens_seen": 406585344
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044281845536609833,
      "loss": 3.1052,
      "theoretical_loss": 4.0045652249289905,
      "tokens_seen": 406650880
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004428084252758275,
      "loss": 2.8869,
      "theoretical_loss": 4.004493227982669,
      "tokens_seen": 406716416
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004427983951855567,
      "loss": 3.1975,
      "theoretical_loss": 4.004421245884332,
      "tokens_seen": 406781952
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004427883650952859,
      "loss": 3.0962,
      "theoretical_loss": 4.004349278628525,
      "tokens_seen": 406847488
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044277833500501506,
      "loss": 3.1986,
      "theoretical_loss": 4.004277326209799,
      "tokens_seen": 406913024
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044276830491474424,
      "loss": 3.3461,
      "theoretical_loss": 4.004205388622704,
      "tokens_seen": 406978560
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044275827482447347,
      "loss": 3.1528,
      "theoretical_loss": 4.004133465861796,
      "tokens_seen": 407044096
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004427482447342026,
      "loss": 3.148,
      "theoretical_loss": 4.004061557921634,
      "tokens_seen": 407109632
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044273821464393184,
      "loss": 3.1944,
      "theoretical_loss": 4.003989664796777,
      "tokens_seen": 407175168
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044272818455366096,
      "loss": 3.1934,
      "theoretical_loss": 4.003917786481791,
      "tokens_seen": 407240704
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004427181544633902,
      "loss": 2.9297,
      "theoretical_loss": 4.003845922971243,
      "tokens_seen": 407306240
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004427081243731194,
      "loss": 3.1094,
      "theoretical_loss": 4.003774074259699,
      "tokens_seen": 407371776
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044269809428284856,
      "loss": 3.314,
      "theoretical_loss": 4.003702240341733,
      "tokens_seen": 407437312
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044268806419257774,
      "loss": 2.9917,
      "theoretical_loss": 4.003630421211922,
      "tokens_seen": 407502848
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004426780341023069,
      "loss": 3.0218,
      "theoretical_loss": 4.003558616864842,
      "tokens_seen": 407568384
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004426680040120361,
      "loss": 3.1118,
      "theoretical_loss": 4.003486827295075,
      "tokens_seen": 407633920
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044265797392176534,
      "loss": 3.1382,
      "theoretical_loss": 4.003415052497203,
      "tokens_seen": 407699456
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044264794383149447,
      "loss": 3.1697,
      "theoretical_loss": 4.003343292465816,
      "tokens_seen": 407764992
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004426379137412237,
      "loss": 3.129,
      "theoretical_loss": 4.0032715471955,
      "tokens_seen": 407830528
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004426278836509529,
      "loss": 3.035,
      "theoretical_loss": 4.00319981668085,
      "tokens_seen": 407896064
    },
    {
      "epoch": 0.12,
      "objective/train/advantage_avg": 0.4938761293888092,
      "objective/train/docs_used": 242638,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.8538668155670166,
      "objective/train/original_loss": 2.8538668155670166,
      "objective/train/theoretical_loss": 4.003128100916459,
      "objective/train/tokens_used": 428421600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24399639666080475,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506279468536377,
      "objective/train/weighted_lm_loss": 2.9983699321746826,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.038975715637207,
      "theoretical_loss": 4.003128100916459,
      "tokens_seen": 407961600
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044261785356068206,
      "loss": 3.0172,
      "theoretical_loss": 4.003128100916459,
      "tokens_seen": 407961600
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044260782347041124,
      "loss": 3.0586,
      "theoretical_loss": 4.003056399896928,
      "tokens_seen": 408027136
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004425977933801404,
      "loss": 3.05,
      "theoretical_loss": 4.002984713616855,
      "tokens_seen": 408092672
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004425877632898696,
      "loss": 2.9993,
      "theoretical_loss": 4.002913042070844,
      "tokens_seen": 408158208
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044257773319959884,
      "loss": 3.1172,
      "theoretical_loss": 4.002841385253504,
      "tokens_seen": 408223744
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044256770310932797,
      "loss": 3.2282,
      "theoretical_loss": 4.002769743159441,
      "tokens_seen": 408289280
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004425576730190572,
      "loss": 3.0955,
      "theoretical_loss": 4.00269811578327,
      "tokens_seen": 408354816
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044254764292878633,
      "loss": 3.0958,
      "theoretical_loss": 4.002626503119606,
      "tokens_seen": 408420352
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044253761283851557,
      "loss": 2.8759,
      "theoretical_loss": 4.0025549051630644,
      "tokens_seen": 408485888
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044252758274824475,
      "loss": 2.9177,
      "theoretical_loss": 4.002483321908269,
      "tokens_seen": 408551424
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044251755265797393,
      "loss": 3.4419,
      "theoretical_loss": 4.002411753349842,
      "tokens_seen": 408616960
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004425075225677031,
      "loss": 3.2204,
      "theoretical_loss": 4.002340199482409,
      "tokens_seen": 408682496
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004424974924774323,
      "loss": 2.919,
      "theoretical_loss": 4.0022686603006,
      "tokens_seen": 408748032
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044248746238716147,
      "loss": 3.2114,
      "theoretical_loss": 4.002197135799047,
      "tokens_seen": 408813568
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004424774322968907,
      "loss": 3.0911,
      "theoretical_loss": 4.002125625972386,
      "tokens_seen": 408879104
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044246740220661983,
      "loss": 3.0664,
      "theoretical_loss": 4.002054130815253,
      "tokens_seen": 408944640
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044245737211634907,
      "loss": 3.2188,
      "theoretical_loss": 4.00198265032229,
      "tokens_seen": 409010176
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044244734202607825,
      "loss": 3.2673,
      "theoretical_loss": 4.001911184488139,
      "tokens_seen": 409075712
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044243731193580743,
      "loss": 3.0349,
      "theoretical_loss": 4.001839733307447,
      "tokens_seen": 409141248
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004424272818455366,
      "loss": 3.0974,
      "theoretical_loss": 4.001768296774863,
      "tokens_seen": 409206784
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004424172517552658,
      "loss": 3.0551,
      "theoretical_loss": 4.001696874885038,
      "tokens_seen": 409272320
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.000442407221664995,
      "loss": 2.9824,
      "theoretical_loss": 4.001625467632628,
      "tokens_seen": 409337856
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004423971915747242,
      "loss": 3.118,
      "theoretical_loss": 4.001554075012289,
      "tokens_seen": 409403392
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044238716148445334,
      "loss": 2.9416,
      "theoretical_loss": 4.001482697018682,
      "tokens_seen": 409468928
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044237713139418257,
      "loss": 2.9903,
      "theoretical_loss": 4.00141133364647,
      "tokens_seen": 409534464
    },
    {
      "epoch": 0.12,
      "objective/train/advantage_avg": 0.4962788224220276,
      "objective/train/docs_used": 244086,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.6330807209014893,
      "objective/train/original_loss": 2.63308048248291,
      "objective/train/theoretical_loss": 4.0013399848903175,
      "objective/train/tokens_used": 430060000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24632515013217926,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0508800745010376,
      "objective/train/weighted_lm_loss": 2.767012357711792,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0417574644088745,
      "theoretical_loss": 4.0013399848903175,
      "tokens_seen": 409600000
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004423671013039117,
      "loss": 3.1013,
      "theoretical_loss": 4.0013399848903175,
      "tokens_seen": 409600000
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044235707121364093,
      "loss": 3.1948,
      "theoretical_loss": 4.001268650744895,
      "tokens_seen": 409665536
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004423470411233701,
      "loss": 3.0818,
      "theoretical_loss": 4.001197331204873,
      "tokens_seen": 409731072
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004423370110330993,
      "loss": 3.093,
      "theoretical_loss": 4.0011260262649255,
      "tokens_seen": 409796608
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004423269809428285,
      "loss": 3.1766,
      "theoretical_loss": 4.00105473591973,
      "tokens_seen": 409862144
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004423169508525577,
      "loss": 3.2,
      "theoretical_loss": 4.000983460163966,
      "tokens_seen": 409927680
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044230692076228684,
      "loss": 3.1784,
      "theoretical_loss": 4.000912198992316,
      "tokens_seen": 409993216
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004422968906720161,
      "loss": 3.1989,
      "theoretical_loss": 4.000840952399466,
      "tokens_seen": 410058752
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004422868605817452,
      "loss": 3.1383,
      "theoretical_loss": 4.000769720380103,
      "tokens_seen": 410124288
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044227683049147444,
      "loss": 3.1819,
      "theoretical_loss": 4.000698502928919,
      "tokens_seen": 410189824
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004422668004012036,
      "loss": 3.2219,
      "theoretical_loss": 4.000627300040607,
      "tokens_seen": 410255360
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004422567703109328,
      "loss": 3.2272,
      "theoretical_loss": 4.000556111709864,
      "tokens_seen": 410320896
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.000442246740220662,
      "loss": 3.2176,
      "theoretical_loss": 4.00048493793139,
      "tokens_seen": 410386432
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044223671013039116,
      "loss": 3.4008,
      "theoretical_loss": 4.000413778699884,
      "tokens_seen": 410451968
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044222668004012034,
      "loss": 3.2545,
      "theoretical_loss": 4.000342634010054,
      "tokens_seen": 410517504
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004422166499498496,
      "loss": 3.3524,
      "theoretical_loss": 4.000271503856607,
      "tokens_seen": 410583040
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004422066198595787,
      "loss": 3.1591,
      "theoretical_loss": 4.000200388234251,
      "tokens_seen": 410648576
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044219658976930794,
      "loss": 3.0124,
      "theoretical_loss": 4.000129287137703,
      "tokens_seen": 410714112
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004421865596790371,
      "loss": 2.898,
      "theoretical_loss": 4.000058200561677,
      "tokens_seen": 410779648
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004421765295887663,
      "loss": 2.9574,
      "theoretical_loss": 3.99998712850089,
      "tokens_seen": 410845184
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044216649949849554,
      "loss": 3.2538,
      "theoretical_loss": 3.999916070950066,
      "tokens_seen": 410910720
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044215646940822467,
      "loss": 2.9578,
      "theoretical_loss": 3.9998450279039277,
      "tokens_seen": 410976256
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004421464393179539,
      "loss": 3.2483,
      "theoretical_loss": 3.9997739993572035,
      "tokens_seen": 411041792
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004421364092276831,
      "loss": 3.1647,
      "theoretical_loss": 3.999702985304621,
      "tokens_seen": 411107328
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044212637913741226,
      "loss": 3.1793,
      "theoretical_loss": 3.999631985740914,
      "tokens_seen": 411172864
    },
    {
      "epoch": 0.12,
      "objective/train/advantage_avg": 0.4963129460811615,
      "objective/train/docs_used": 244630,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.2444276809692383,
      "objective/train/original_loss": 3.2444279193878174,
      "objective/train/theoretical_loss": 3.999561000660818,
      "objective/train/tokens_used": 431698400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24636799097061157,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0508836507797241,
      "objective/train/weighted_lm_loss": 3.4094364643096924,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0458241701126099,
      "theoretical_loss": 3.999561000660818,
      "tokens_seen": 411238400
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044211634904714144,
      "loss": 2.9974,
      "theoretical_loss": 3.999561000660818,
      "tokens_seen": 411238400
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004421063189568706,
      "loss": 3.0105,
      "theoretical_loss": 3.99949003005907,
      "tokens_seen": 411303936
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004420962888665998,
      "loss": 3.1063,
      "theoretical_loss": 3.9994190739304116,
      "tokens_seen": 411369472
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044208625877632904,
      "loss": 3.0242,
      "theoretical_loss": 3.999348132269586,
      "tokens_seen": 411435008
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044207622868605817,
      "loss": 3.1455,
      "theoretical_loss": 3.9992772050713388,
      "tokens_seen": 411500544
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004420661985957874,
      "loss": 3.2103,
      "theoretical_loss": 3.9992062923304195,
      "tokens_seen": 411566080
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044205616850551653,
      "loss": 2.9401,
      "theoretical_loss": 3.9991353940415797,
      "tokens_seen": 411631616
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044204613841524577,
      "loss": 2.7354,
      "theoretical_loss": 3.999064510199574,
      "tokens_seen": 411697152
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044203610832497495,
      "loss": 3.0261,
      "theoretical_loss": 3.998993640799159,
      "tokens_seen": 411762688
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044202607823470413,
      "loss": 2.9114,
      "theoretical_loss": 3.9989227858350955,
      "tokens_seen": 411828224
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004420160481444333,
      "loss": 3.1174,
      "theoretical_loss": 3.998851945302146,
      "tokens_seen": 411893760
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004420060180541625,
      "loss": 3.4254,
      "theoretical_loss": 3.998781119195075,
      "tokens_seen": 411959296
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044199598796389167,
      "loss": 3.0274,
      "theoretical_loss": 3.998710307508651,
      "tokens_seen": 412024832
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004419859578736209,
      "loss": 2.8714,
      "theoretical_loss": 3.9986395102376453,
      "tokens_seen": 412090368
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044197592778335003,
      "loss": 3.0866,
      "theoretical_loss": 3.9985687273768313,
      "tokens_seen": 412155904
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044196589769307927,
      "loss": 3.1875,
      "theoretical_loss": 3.998497958920985,
      "tokens_seen": 412221440
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044195586760280845,
      "loss": 2.9685,
      "theoretical_loss": 3.9984272048648855,
      "tokens_seen": 412286976
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.00044194583751253763,
      "loss": 2.874,
      "theoretical_loss": 3.9983564652033152,
      "tokens_seen": 412352512
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.0004419358074222668,
      "loss": 3.1941,
      "theoretical_loss": 3.998285739931058,
      "tokens_seen": 412418048
    },
    {
      "epoch": 0.12,
      "learning_rate": 0.000441925777331996,
      "loss": 3.1524,
      "theoretical_loss": 3.9982150290429,
      "tokens_seen": 412483584
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004419157472417252,
      "loss": 2.8739,
      "theoretical_loss": 3.998144332533633,
      "tokens_seen": 412549120
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004419057171514544,
      "loss": 3.1844,
      "theoretical_loss": 3.9980736503980485,
      "tokens_seen": 412614656
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00044189568706118354,
      "loss": 3.0202,
      "theoretical_loss": 3.9980029826309424,
      "tokens_seen": 412680192
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00044188565697091277,
      "loss": 3.1393,
      "theoretical_loss": 3.9979323292271123,
      "tokens_seen": 412745728
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004418756268806419,
      "loss": 2.7904,
      "theoretical_loss": 3.9978616901813586,
      "tokens_seen": 412811264
    },
    {
      "epoch": 0.13,
      "objective/train/advantage_avg": 0.4955955743789673,
      "objective/train/docs_used": 245807,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.322376251220703,
      "objective/train/original_loss": 2.322376251220703,
      "objective/train/theoretical_loss": 3.997791065488486,
      "objective/train/tokens_used": 433336800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24565617740154266,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050808310508728,
      "objective/train/weighted_lm_loss": 2.4402048587799072,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0417574644088745,
      "theoretical_loss": 3.997791065488486,
      "tokens_seen": 412876800
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00044186559679037113,
      "loss": 2.7607,
      "theoretical_loss": 3.997791065488486,
      "tokens_seen": 412876800
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004418555667001003,
      "loss": 2.9997,
      "theoretical_loss": 3.997720455143299,
      "tokens_seen": 412942336
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004418455366098295,
      "loss": 3.0902,
      "theoretical_loss": 3.9976498591406076,
      "tokens_seen": 413007872
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004418355065195587,
      "loss": 3.1544,
      "theoretical_loss": 3.9975792774752232,
      "tokens_seen": 413073408
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004418254764292879,
      "loss": 2.952,
      "theoretical_loss": 3.99750871014196,
      "tokens_seen": 413138944
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00044181544633901704,
      "loss": 3.1051,
      "theoretical_loss": 3.997438157135635,
      "tokens_seen": 413204480
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004418054162487463,
      "loss": 3.0242,
      "theoretical_loss": 3.997367618451068,
      "tokens_seen": 413270016
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004417953861584754,
      "loss": 3.2218,
      "theoretical_loss": 3.99729709408308,
      "tokens_seen": 413335552
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00044178535606820464,
      "loss": 3.2915,
      "theoretical_loss": 3.9972265840264978,
      "tokens_seen": 413401088
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004417753259779338,
      "loss": 3.1157,
      "theoretical_loss": 3.9971560882761485,
      "tokens_seen": 413466624
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.000441765295887663,
      "loss": 3.1911,
      "theoretical_loss": 3.9970856068268623,
      "tokens_seen": 413532160
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004417552657973922,
      "loss": 3.1022,
      "theoretical_loss": 3.9970151396734726,
      "tokens_seen": 413597696
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00044174523570712136,
      "loss": 3.2268,
      "theoretical_loss": 3.9969446868108145,
      "tokens_seen": 413663232
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00044173520561685054,
      "loss": 3.011,
      "theoretical_loss": 3.996874248233728,
      "tokens_seen": 413728768
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004417251755265798,
      "loss": 3.0634,
      "theoretical_loss": 3.9968038239370527,
      "tokens_seen": 413794304
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004417151454363089,
      "loss": 3.1866,
      "theoretical_loss": 3.9967334139156327,
      "tokens_seen": 413859840
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00044170511534603814,
      "loss": 3.1213,
      "theoretical_loss": 3.9966630181643152,
      "tokens_seen": 413925376
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00044169508525576727,
      "loss": 3.1426,
      "theoretical_loss": 3.996592636677949,
      "tokens_seen": 413990912
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004416850551654965,
      "loss": 3.1909,
      "theoretical_loss": 3.9965222694513862,
      "tokens_seen": 414056448
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004416750250752257,
      "loss": 2.7597,
      "theoretical_loss": 3.9964519164794803,
      "tokens_seen": 414121984
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00044166499498495487,
      "loss": 3.1036,
      "theoretical_loss": 3.9963815777570897,
      "tokens_seen": 414187520
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00044165496489468405,
      "loss": 3.2796,
      "theoretical_loss": 3.996311253279074,
      "tokens_seen": 414253056
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004416449348044133,
      "loss": 2.6744,
      "theoretical_loss": 3.996240943040296,
      "tokens_seen": 414318592
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004416349047141424,
      "loss": 3.2298,
      "theoretical_loss": 3.99617064703562,
      "tokens_seen": 414384128
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00044162487462387164,
      "loss": 3.0025,
      "theoretical_loss": 3.9961003652599145,
      "tokens_seen": 414449664
    },
    {
      "epoch": 0.13,
      "objective/train/advantage_avg": 0.4956885278224945,
      "objective/train/docs_used": 246540,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.890080451965332,
      "objective/train/original_loss": 2.890080690383911,
      "objective/train/theoretical_loss": 3.99603009770805,
      "objective/train/tokens_used": 434975200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24575375020503998,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0508180856704712,
      "objective/train/weighted_lm_loss": 3.036958932876587,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0363596677780151,
      "theoretical_loss": 3.99603009770805,
      "tokens_seen": 414515200
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00044161484453360077,
      "loss": 3.0825,
      "theoretical_loss": 3.99603009770805,
      "tokens_seen": 414515200
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00044160481444333,
      "loss": 3.1426,
      "theoretical_loss": 3.9959598443748994,
      "tokens_seen": 414580736
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004415947843530592,
      "loss": 3.3152,
      "theoretical_loss": 3.9958896052553383,
      "tokens_seen": 414646272
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00044158475426278837,
      "loss": 3.147,
      "theoretical_loss": 3.9958193803442468,
      "tokens_seen": 414711808
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00044157472417251755,
      "loss": 3.3641,
      "theoretical_loss": 3.995749169636504,
      "tokens_seen": 414777344
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00044156469408224673,
      "loss": 2.9563,
      "theoretical_loss": 3.995678973126995,
      "tokens_seen": 414842880
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004415546639919759,
      "loss": 2.9787,
      "theoretical_loss": 3.995608790810606,
      "tokens_seen": 414908416
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00044154463390170515,
      "loss": 2.9425,
      "theoretical_loss": 3.995538622682226,
      "tokens_seen": 414973952
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004415346038114343,
      "loss": 2.9967,
      "theoretical_loss": 3.9954684687367465,
      "tokens_seen": 415039488
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004415245737211635,
      "loss": 3.1602,
      "theoretical_loss": 3.9953983289690624,
      "tokens_seen": 415105024
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00044151454363089264,
      "loss": 3.2486,
      "theoretical_loss": 3.9953282033740702,
      "tokens_seen": 415170560
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00044150451354062187,
      "loss": 2.8959,
      "theoretical_loss": 3.99525809194667,
      "tokens_seen": 415236096
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00044149448345035105,
      "loss": 2.9426,
      "theoretical_loss": 3.995187994681764,
      "tokens_seen": 415301632
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00044148445336008023,
      "loss": 2.976,
      "theoretical_loss": 3.9951179115742574,
      "tokens_seen": 415367168
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004414744232698094,
      "loss": 3.1256,
      "theoretical_loss": 3.995047842619057,
      "tokens_seen": 415432704
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00044146439317953865,
      "loss": 3.093,
      "theoretical_loss": 3.9949777878110746,
      "tokens_seen": 415498240
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004414543630892678,
      "loss": 3.1932,
      "theoretical_loss": 3.9949077471452212,
      "tokens_seen": 415563776
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.000441444332998997,
      "loss": 3.0572,
      "theoretical_loss": 3.994837720616414,
      "tokens_seen": 415629312
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004414343029087262,
      "loss": 3.0709,
      "theoretical_loss": 3.9947677082195696,
      "tokens_seen": 415694848
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004414242728184554,
      "loss": 3.1767,
      "theoretical_loss": 3.9946977099496097,
      "tokens_seen": 415760384
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004414142427281846,
      "loss": 3.2634,
      "theoretical_loss": 3.9946277258014575,
      "tokens_seen": 415825920
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00044140421263791374,
      "loss": 3.0897,
      "theoretical_loss": 3.9945577557700394,
      "tokens_seen": 415891456
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00044139418254764297,
      "loss": 2.9566,
      "theoretical_loss": 3.9944877998502832,
      "tokens_seen": 415956992
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004413841524573721,
      "loss": 3.0077,
      "theoretical_loss": 3.994417858037121,
      "tokens_seen": 416022528
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00044137412236710133,
      "loss": 3.0153,
      "theoretical_loss": 3.9943479303254854,
      "tokens_seen": 416088064
    },
    {
      "epoch": 0.13,
      "objective/train/advantage_avg": 0.4955214858055115,
      "objective/train/docs_used": 247824,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.858980178833008,
      "objective/train/original_loss": 2.858980178833008,
      "objective/train/theoretical_loss": 3.9942780167103145,
      "objective/train/tokens_used": 436613600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24560469388961792,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0508006811141968,
      "objective/train/weighted_lm_loss": 3.0040857791900635,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 3.9942780167103145,
      "tokens_seen": 416153600
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004413640922768305,
      "loss": 3.1063,
      "theoretical_loss": 3.9942780167103145,
      "tokens_seen": 416153600
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004413540621865597,
      "loss": 3.2445,
      "theoretical_loss": 3.9942081171865462,
      "tokens_seen": 416219136
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004413440320962889,
      "loss": 3.3067,
      "theoretical_loss": 3.9941382317491225,
      "tokens_seen": 416284672
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004413340020060181,
      "loss": 2.9425,
      "theoretical_loss": 3.9940683603929883,
      "tokens_seen": 416350208
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00044132397191574724,
      "loss": 3.0644,
      "theoretical_loss": 3.9939985031130902,
      "tokens_seen": 416415744
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004413139418254765,
      "loss": 2.9884,
      "theoretical_loss": 3.993928659904377,
      "tokens_seen": 416481280
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004413039117352056,
      "loss": 3.1827,
      "theoretical_loss": 3.993858830761802,
      "tokens_seen": 416546816
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00044129388164493484,
      "loss": 3.0689,
      "theoretical_loss": 3.9937890156803197,
      "tokens_seen": 416612352
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.000441283851554664,
      "loss": 3.2405,
      "theoretical_loss": 3.9937192146548863,
      "tokens_seen": 416677888
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004412738214643932,
      "loss": 3.0654,
      "theoretical_loss": 3.993649427680464,
      "tokens_seen": 416743424
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004412637913741224,
      "loss": 3.1655,
      "theoretical_loss": 3.993579654752013,
      "tokens_seen": 416808960
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00044125376128385156,
      "loss": 3.1423,
      "theoretical_loss": 3.9935098958645,
      "tokens_seen": 416874496
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00044124373119358074,
      "loss": 3.1217,
      "theoretical_loss": 3.993440151012892,
      "tokens_seen": 416940032
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00044123370110331,
      "loss": 3.0258,
      "theoretical_loss": 3.99337042019216,
      "tokens_seen": 417005568
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004412236710130391,
      "loss": 3.1566,
      "theoretical_loss": 3.993300703397276,
      "tokens_seen": 417071104
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00044121364092276834,
      "loss": 3.0796,
      "theoretical_loss": 3.9932310006232163,
      "tokens_seen": 417136640
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00044120361083249747,
      "loss": 3.1591,
      "theoretical_loss": 3.993161311864959,
      "tokens_seen": 417202176
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004411935807422267,
      "loss": 3.0125,
      "theoretical_loss": 3.9930916371174843,
      "tokens_seen": 417267712
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004411835506519559,
      "loss": 3.3043,
      "theoretical_loss": 3.9930219763757755,
      "tokens_seen": 417333248
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00044117352056168507,
      "loss": 3.1564,
      "theoretical_loss": 3.9929523296348197,
      "tokens_seen": 417398784
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00044116349047141425,
      "loss": 2.936,
      "theoretical_loss": 3.9928826968896036,
      "tokens_seen": 417464320
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004411534603811435,
      "loss": 3.1144,
      "theoretical_loss": 3.9928130781351197,
      "tokens_seen": 417529856
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004411434302908726,
      "loss": 3.1986,
      "theoretical_loss": 3.9927434733663603,
      "tokens_seen": 417595392
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00044113340020060184,
      "loss": 3.2017,
      "theoretical_loss": 3.992673882578323,
      "tokens_seen": 417660928
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00044112337011033097,
      "loss": 2.8774,
      "theoretical_loss": 3.9926043057660054,
      "tokens_seen": 417726464
    },
    {
      "epoch": 0.13,
      "objective/train/advantage_avg": 0.49592125415802,
      "objective/train/docs_used": 248559,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.895479917526245,
      "objective/train/original_loss": 2.895479679107666,
      "objective/train/theoretical_loss": 3.992534742924409,
      "objective/train/tokens_used": 438252000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24598200619220734,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050842523574829,
      "objective/train/weighted_lm_loss": 3.042731761932373,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0417574644088745,
      "theoretical_loss": 3.992534742924409,
      "tokens_seen": 417792000
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004411133400200602,
      "loss": 2.814,
      "theoretical_loss": 3.992534742924409,
      "tokens_seen": 417792000
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004411033099297894,
      "loss": 2.9935,
      "theoretical_loss": 3.9924651940485387,
      "tokens_seen": 417857536
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00044109327983951857,
      "loss": 3.0036,
      "theoretical_loss": 3.9923956591334004,
      "tokens_seen": 417923072
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00044108324974924775,
      "loss": 3.0207,
      "theoretical_loss": 3.9923261381740023,
      "tokens_seen": 417988608
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00044107321965897693,
      "loss": 3.0023,
      "theoretical_loss": 3.9922566311653576,
      "tokens_seen": 418054144
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004410631895687061,
      "loss": 2.9132,
      "theoretical_loss": 3.9921871381024787,
      "tokens_seen": 418119680
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00044105315947843535,
      "loss": 2.9794,
      "theoretical_loss": 3.9921176589803835,
      "tokens_seen": 418185216
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004410431293881645,
      "loss": 2.8316,
      "theoretical_loss": 3.9920481937940915,
      "tokens_seen": 418250752
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004410330992978937,
      "loss": 3.0781,
      "theoretical_loss": 3.991978742538624,
      "tokens_seen": 418316288
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00044102306920762284,
      "loss": 3.1599,
      "theoretical_loss": 3.9919093052090058,
      "tokens_seen": 418381824
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00044101303911735207,
      "loss": 3.3056,
      "theoretical_loss": 3.9918398818002636,
      "tokens_seen": 418447360
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00044100300902708125,
      "loss": 3.6267,
      "theoretical_loss": 3.9917704723074268,
      "tokens_seen": 418512896
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00044099297893681043,
      "loss": 3.0784,
      "theoretical_loss": 3.9917010767255277,
      "tokens_seen": 418578432
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004409829488465396,
      "loss": 3.2958,
      "theoretical_loss": 3.991631695049601,
      "tokens_seen": 418643968
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00044097291875626885,
      "loss": 2.9607,
      "theoretical_loss": 3.991562327274684,
      "tokens_seen": 418709504
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.000440962888665998,
      "loss": 3.0898,
      "theoretical_loss": 3.9914929733958164,
      "tokens_seen": 418775040
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004409528585757272,
      "loss": 3.0197,
      "theoretical_loss": 3.99142363340804,
      "tokens_seen": 418840576
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00044094282848545634,
      "loss": 3.0013,
      "theoretical_loss": 3.9913543073063997,
      "tokens_seen": 418906112
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004409327983951856,
      "loss": 3.1527,
      "theoretical_loss": 3.9912849950859437,
      "tokens_seen": 418971648
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00044092276830491476,
      "loss": 3.155,
      "theoretical_loss": 3.9912156967417207,
      "tokens_seen": 419037184
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00044091273821464394,
      "loss": 2.9126,
      "theoretical_loss": 3.9911464122687845,
      "tokens_seen": 419102720
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004409027081243731,
      "loss": 2.8499,
      "theoretical_loss": 3.9910771416621893,
      "tokens_seen": 419168256
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004408926780341023,
      "loss": 3.2425,
      "theoretical_loss": 3.9910078849169923,
      "tokens_seen": 419233792
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004408826479438315,
      "loss": 3.0799,
      "theoretical_loss": 3.9909386420282544,
      "tokens_seen": 419299328
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004408726178535607,
      "loss": 2.7905,
      "theoretical_loss": 3.990869412991037,
      "tokens_seen": 419364864
    },
    {
      "epoch": 0.13,
      "objective/train/advantage_avg": 0.49430254101753235,
      "objective/train/docs_used": 249785,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.6033706665039062,
      "objective/train/original_loss": 2.6033709049224854,
      "objective/train/theoretical_loss": 3.9908001978004064,
      "objective/train/tokens_used": 439890400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.244392991065979,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506726503372192,
      "objective/train/weighted_lm_loss": 2.7352826595306396,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0396549701690674,
      "theoretical_loss": 3.9908001978004064,
      "tokens_seen": 419430400
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00044086258776328984,
      "loss": 2.9494,
      "theoretical_loss": 3.9908001978004064,
      "tokens_seen": 419430400
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004408525576730191,
      "loss": 3.0923,
      "theoretical_loss": 3.99073099645143,
      "tokens_seen": 419495936
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004408425275827482,
      "loss": 3.1187,
      "theoretical_loss": 3.990661808939178,
      "tokens_seen": 419561472
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00044083249749247744,
      "loss": 2.9432,
      "theoretical_loss": 3.9905926352587224,
      "tokens_seen": 419627008
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004408224674022066,
      "loss": 3.0005,
      "theoretical_loss": 3.990523475405139,
      "tokens_seen": 419692544
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004408124373119358,
      "loss": 2.9218,
      "theoretical_loss": 3.990454329373505,
      "tokens_seen": 419758080
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.000440802407221665,
      "loss": 3.0147,
      "theoretical_loss": 3.9903851971589015,
      "tokens_seen": 419823616
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004407923771313942,
      "loss": 3.2123,
      "theoretical_loss": 3.990316078756411,
      "tokens_seen": 419889152
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00044078234704112335,
      "loss": 3.0942,
      "theoretical_loss": 3.990246974161119,
      "tokens_seen": 419954688
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004407723169508526,
      "loss": 3.0924,
      "theoretical_loss": 3.9901778833681125,
      "tokens_seen": 420020224
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004407622868605817,
      "loss": 3.0059,
      "theoretical_loss": 3.9901088063724828,
      "tokens_seen": 420085760
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00044075225677031094,
      "loss": 3.0602,
      "theoretical_loss": 3.990039743169322,
      "tokens_seen": 420151296
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004407422266800401,
      "loss": 3.0024,
      "theoretical_loss": 3.989970693753726,
      "tokens_seen": 420216832
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004407321965897693,
      "loss": 3.1898,
      "theoretical_loss": 3.989901658120792,
      "tokens_seen": 420282368
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004407221664994985,
      "loss": 3.1334,
      "theoretical_loss": 3.9898326362656205,
      "tokens_seen": 420347904
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00044071213640922767,
      "loss": 3.1496,
      "theoretical_loss": 3.9897636281833155,
      "tokens_seen": 420413440
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00044070210631895685,
      "loss": 2.9993,
      "theoretical_loss": 3.989694633868981,
      "tokens_seen": 420478976
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004406920762286861,
      "loss": 3.093,
      "theoretical_loss": 3.9896256533177255,
      "tokens_seen": 420544512
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00044068204613841527,
      "loss": 3.3582,
      "theoretical_loss": 3.9895566865246597,
      "tokens_seen": 420610048
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00044067201604814445,
      "loss": 3.2009,
      "theoretical_loss": 3.9894877334848955,
      "tokens_seen": 420675584
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004406619859578737,
      "loss": 2.8202,
      "theoretical_loss": 3.989418794193549,
      "tokens_seen": 420741120
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004406519558676028,
      "loss": 3.1593,
      "theoretical_loss": 3.9893498686457383,
      "tokens_seen": 420806656
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00044064192577733204,
      "loss": 2.9707,
      "theoretical_loss": 3.9892809568365823,
      "tokens_seen": 420872192
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00044063189568706117,
      "loss": 3.1377,
      "theoretical_loss": 3.989212058761206,
      "tokens_seen": 420937728
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004406218655967904,
      "loss": 3.1204,
      "theoretical_loss": 3.989143174414733,
      "tokens_seen": 421003264
    },
    {
      "epoch": 0.13,
      "objective/train/advantage_avg": 0.4918384850025177,
      "objective/train/docs_used": 250392,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.895622491836548,
      "objective/train/original_loss": 2.895622491836548,
      "objective/train/theoretical_loss": 3.989074303792292,
      "objective/train/tokens_used": 441528800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24205686151981354,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0504142045974731,
      "objective/train/weighted_lm_loss": 3.041661500930786,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0338950157165527,
      "theoretical_loss": 3.989074303792292,
      "tokens_seen": 421068800
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004406118355065196,
      "loss": 3.0906,
      "theoretical_loss": 3.989074303792292,
      "tokens_seen": 421068800
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00044060180541624877,
      "loss": 3.0202,
      "theoretical_loss": 3.9890054468890135,
      "tokens_seen": 421134336
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00044059177532597795,
      "loss": 3.0755,
      "theoretical_loss": 3.9889366037000302,
      "tokens_seen": 421199872
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00044058174523570713,
      "loss": 3.2417,
      "theoretical_loss": 3.9888677742204766,
      "tokens_seen": 421265408
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004405717151454363,
      "loss": 3.0942,
      "theoretical_loss": 3.988798958445491,
      "tokens_seen": 421330944
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00044056168505516555,
      "loss": 3.0753,
      "theoretical_loss": 3.988730156370214,
      "tokens_seen": 421396480
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004405516549648947,
      "loss": 3.0451,
      "theoretical_loss": 3.9886613679897875,
      "tokens_seen": 421462016
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004405416248746239,
      "loss": 3.2236,
      "theoretical_loss": 3.988592593299358,
      "tokens_seen": 421527552
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00044053159478435304,
      "loss": 3.0018,
      "theoretical_loss": 3.9885238322940717,
      "tokens_seen": 421593088
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00044052156469408227,
      "loss": 2.9749,
      "theoretical_loss": 3.9884550849690794,
      "tokens_seen": 421658624
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00044051153460381145,
      "loss": 3.2285,
      "theoretical_loss": 3.988386351319534,
      "tokens_seen": 421724160
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00044050150451354063,
      "loss": 2.839,
      "theoretical_loss": 3.9883176313405904,
      "tokens_seen": 421789696
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004404914744232698,
      "loss": 3.219,
      "theoretical_loss": 3.9882489250274062,
      "tokens_seen": 421855232
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00044048144433299905,
      "loss": 3.108,
      "theoretical_loss": 3.9881802323751416,
      "tokens_seen": 421920768
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004404714142427282,
      "loss": 3.1439,
      "theoretical_loss": 3.9881115533789586,
      "tokens_seen": 421986304
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004404613841524574,
      "loss": 2.9826,
      "theoretical_loss": 3.9880428880340224,
      "tokens_seen": 422051840
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00044045135406218654,
      "loss": 3.0418,
      "theoretical_loss": 3.987974236335501,
      "tokens_seen": 422117376
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004404413239719158,
      "loss": 3.492,
      "theoretical_loss": 3.987905598278563,
      "tokens_seen": 422182912
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00044043129388164496,
      "loss": 3.2112,
      "theoretical_loss": 3.987836973858382,
      "tokens_seen": 422248448
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00044042126379137414,
      "loss": 3.1736,
      "theoretical_loss": 3.9877683630701326,
      "tokens_seen": 422313984
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004404112337011033,
      "loss": 2.898,
      "theoretical_loss": 3.9876997659089914,
      "tokens_seen": 422379520
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004404012036108325,
      "loss": 3.0438,
      "theoretical_loss": 3.987631182370139,
      "tokens_seen": 422445056
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004403911735205617,
      "loss": 2.9741,
      "theoretical_loss": 3.987562612448756,
      "tokens_seen": 422510592
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004403811434302909,
      "loss": 3.0418,
      "theoretical_loss": 3.9874940561400294,
      "tokens_seen": 422576128
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00044037111334002004,
      "loss": 3.226,
      "theoretical_loss": 3.9874255134391445,
      "tokens_seen": 422641664
    },
    {
      "epoch": 0.13,
      "objective/train/advantage_avg": 0.4946867823600769,
      "objective/train/docs_used": 251199,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.423170804977417,
      "objective/train/original_loss": 3.423170804977417,
      "objective/train/theoretical_loss": 3.9873569843412913,
      "objective/train/tokens_used": 443167200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24480226635932922,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050713062286377,
      "objective/train/weighted_lm_loss": 3.596906900405884,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0392107963562012,
      "theoretical_loss": 3.9873569843412913,
      "tokens_seen": 422707200
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004403610832497493,
      "loss": 3.3326,
      "theoretical_loss": 3.9873569843412913,
      "tokens_seen": 422707200
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004403510531594784,
      "loss": 3.0037,
      "theoretical_loss": 3.987288468841662,
      "tokens_seen": 422772736
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00044034102306920764,
      "loss": 3.2687,
      "theoretical_loss": 3.9872199669354504,
      "tokens_seen": 422838272
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004403309929789368,
      "loss": 2.9454,
      "theoretical_loss": 3.987151478617854,
      "tokens_seen": 422903808
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.000440320962888666,
      "loss": 3.219,
      "theoretical_loss": 3.987083003884072,
      "tokens_seen": 422969344
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004403109327983952,
      "loss": 2.7903,
      "theoretical_loss": 3.9870145427293053,
      "tokens_seen": 423034880
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004403009027081244,
      "loss": 3.277,
      "theoretical_loss": 3.9869460951487588,
      "tokens_seen": 423100416
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00044029087261785355,
      "loss": 3.2059,
      "theoretical_loss": 3.9868776611376395,
      "tokens_seen": 423165952
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004402808425275828,
      "loss": 2.7076,
      "theoretical_loss": 3.986809240691156,
      "tokens_seen": 423231488
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004402708124373119,
      "loss": 3.0313,
      "theoretical_loss": 3.9867408338045194,
      "tokens_seen": 423297024
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00044026078234704114,
      "loss": 2.9942,
      "theoretical_loss": 3.9866724404729443,
      "tokens_seen": 423362560
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004402507522567703,
      "loss": 2.8058,
      "theoretical_loss": 3.986604060691646,
      "tokens_seen": 423428096
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004402407221664995,
      "loss": 3.0146,
      "theoretical_loss": 3.9865356944558448,
      "tokens_seen": 423493632
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004402306920762287,
      "loss": 3.1647,
      "theoretical_loss": 3.9864673417607603,
      "tokens_seen": 423559168
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00044022066198595787,
      "loss": 2.9839,
      "theoretical_loss": 3.986399002601617,
      "tokens_seen": 423624704
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00044021063189568705,
      "loss": 3.1077,
      "theoretical_loss": 3.9863306769736413,
      "tokens_seen": 423690240
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004402006018054163,
      "loss": 2.9934,
      "theoretical_loss": 3.9862623648720605,
      "tokens_seen": 423755776
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004401905717151454,
      "loss": 3.0975,
      "theoretical_loss": 3.986194066292107,
      "tokens_seen": 423821312
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00044018054162487465,
      "loss": 3.0779,
      "theoretical_loss": 3.9861257812290125,
      "tokens_seen": 423886848
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00044017051153460383,
      "loss": 3.1165,
      "theoretical_loss": 3.9860575096780133,
      "tokens_seen": 423952384
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.000440160481444333,
      "loss": 3.1884,
      "theoretical_loss": 3.985989251634348,
      "tokens_seen": 424017920
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004401504513540622,
      "loss": 3.0944,
      "theoretical_loss": 3.985921007093257,
      "tokens_seen": 424083456
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00044014042126379137,
      "loss": 3.0184,
      "theoretical_loss": 3.985852776049983,
      "tokens_seen": 424148992
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00044013039117352055,
      "loss": 3.1558,
      "theoretical_loss": 3.985784558499771,
      "tokens_seen": 424214528
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004401203610832498,
      "loss": 2.7903,
      "theoretical_loss": 3.9857163544378693,
      "tokens_seen": 424280064
    },
    {
      "epoch": 0.13,
      "objective/train/advantage_avg": 0.4936714172363281,
      "objective/train/docs_used": 252247,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.6715826988220215,
      "objective/train/original_loss": 2.6715826988220215,
      "objective/train/theoretical_loss": 3.985648163859529,
      "objective/train/tokens_used": 444805600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24375616014003754,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506062507629395,
      "objective/train/weighted_lm_loss": 2.806828737258911,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0432155132293701,
      "theoretical_loss": 3.985648163859529,
      "tokens_seen": 424345600
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004401103309929789,
      "loss": 2.7691,
      "theoretical_loss": 3.985648163859529,
      "tokens_seen": 424345600
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00044010030090270815,
      "loss": 3.0329,
      "theoretical_loss": 3.9855799867600004,
      "tokens_seen": 424411136
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004400902708124373,
      "loss": 2.7966,
      "theoretical_loss": 3.9855118231345403,
      "tokens_seen": 424476672
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004400802407221665,
      "loss": 3.1198,
      "theoretical_loss": 3.9854436729784055,
      "tokens_seen": 424542208
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004400702106318957,
      "loss": 3.2688,
      "theoretical_loss": 3.985375536286856,
      "tokens_seen": 424607744
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004400601805416249,
      "loss": 2.7086,
      "theoretical_loss": 3.9853074130551542,
      "tokens_seen": 424673280
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00044005015045135406,
      "loss": 2.9555,
      "theoretical_loss": 3.9852393032785636,
      "tokens_seen": 424738816
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00044004012036108324,
      "loss": 3.0627,
      "theoretical_loss": 3.9851712069523524,
      "tokens_seen": 424804352
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004400300902708124,
      "loss": 2.9885,
      "theoretical_loss": 3.98510312407179,
      "tokens_seen": 424869888
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00044002006018054165,
      "loss": 3.0241,
      "theoretical_loss": 3.9850350546321467,
      "tokens_seen": 424935424
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004400100300902708,
      "loss": 2.9212,
      "theoretical_loss": 3.9849669986286984,
      "tokens_seen": 425000960
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00044,
      "loss": 2.873,
      "theoretical_loss": 3.9848989560567207,
      "tokens_seen": 425066496
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004399899699097292,
      "loss": 3.2274,
      "theoretical_loss": 3.984830926911493,
      "tokens_seen": 425132032
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004399799398194584,
      "loss": 2.7638,
      "theoretical_loss": 3.9847629111882963,
      "tokens_seen": 425197568
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043996990972918756,
      "loss": 3.0912,
      "theoretical_loss": 3.984694908882415,
      "tokens_seen": 425263104
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043995987963891674,
      "loss": 3.0061,
      "theoretical_loss": 3.984626919989134,
      "tokens_seen": 425328640
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004399498495486459,
      "loss": 3.2703,
      "theoretical_loss": 3.9845589445037426,
      "tokens_seen": 425394176
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043993981945837516,
      "loss": 2.8943,
      "theoretical_loss": 3.9844909824215318,
      "tokens_seen": 425459712
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043992978936810434,
      "loss": 2.8873,
      "theoretical_loss": 3.9844230337377944,
      "tokens_seen": 425525248
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004399197592778335,
      "loss": 3.21,
      "theoretical_loss": 3.984355098447826,
      "tokens_seen": 425590784
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004399097291875627,
      "loss": 2.8011,
      "theoretical_loss": 3.9842871765469248,
      "tokens_seen": 425656320
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004398996990972919,
      "loss": 2.908,
      "theoretical_loss": 3.984219268030392,
      "tokens_seen": 425721856
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004398896690070211,
      "loss": 2.9683,
      "theoretical_loss": 3.9841513728935287,
      "tokens_seen": 425787392
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043987963891675024,
      "loss": 3.0896,
      "theoretical_loss": 3.984083491131641,
      "tokens_seen": 425852928
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004398696088264795,
      "loss": 3.0776,
      "theoretical_loss": 3.9840156227400367,
      "tokens_seen": 425918464
    },
    {
      "epoch": 0.13,
      "objective/train/advantage_avg": 0.49436643719673157,
      "objective/train/docs_used": 253004,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.330651044845581,
      "objective/train/original_loss": 3.330651044845581,
      "objective/train/theoretical_loss": 3.9839477677140245,
      "objective/train/tokens_used": 446444000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2444731891155243,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050679326057434,
      "objective/train/weighted_lm_loss": 3.4994890689849854,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0417574644088745,
      "theoretical_loss": 3.9839477677140245,
      "tokens_seen": 425984000
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004398595787362086,
      "loss": 3.3169,
      "theoretical_loss": 3.9839477677140245,
      "tokens_seen": 425984000
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043984954864593784,
      "loss": 2.9016,
      "theoretical_loss": 3.9838799260489175,
      "tokens_seen": 426049536
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.000439839518555667,
      "loss": 3.3431,
      "theoretical_loss": 3.9838120977400306,
      "tokens_seen": 426115072
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004398294884653962,
      "loss": 3.2162,
      "theoretical_loss": 3.9837442827826797,
      "tokens_seen": 426180608
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004398194583751254,
      "loss": 3.049,
      "theoretical_loss": 3.9836764811721848,
      "tokens_seen": 426246144
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004398094282848546,
      "loss": 3.2197,
      "theoretical_loss": 3.9836086929038674,
      "tokens_seen": 426311680
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043979939819458375,
      "loss": 3.1393,
      "theoretical_loss": 3.983540917973052,
      "tokens_seen": 426377216
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.000439789368104313,
      "loss": 2.8645,
      "theoretical_loss": 3.983473156375064,
      "tokens_seen": 426442752
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004397793380140421,
      "loss": 3.0324,
      "theoretical_loss": 3.983405408105233,
      "tokens_seen": 426508288
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043976930792377134,
      "loss": 3.0824,
      "theoretical_loss": 3.98333767315889,
      "tokens_seen": 426573824
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004397592778335005,
      "loss": 3.0342,
      "theoretical_loss": 3.983269951531368,
      "tokens_seen": 426639360
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004397492477432297,
      "loss": 3.1957,
      "theoretical_loss": 3.9832022432180025,
      "tokens_seen": 426704896
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004397392176529589,
      "loss": 3.0283,
      "theoretical_loss": 3.983134548214133,
      "tokens_seen": 426770432
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043972918756268807,
      "loss": 3.0996,
      "theoretical_loss": 3.983066866515099,
      "tokens_seen": 426835968
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043971915747241725,
      "loss": 3.2127,
      "theoretical_loss": 3.982999198116244,
      "tokens_seen": 426901504
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004397091273821465,
      "loss": 3.1076,
      "theoretical_loss": 3.9829315430129117,
      "tokens_seen": 426967040
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004396990972918756,
      "loss": 3.0768,
      "theoretical_loss": 3.9828639012004516,
      "tokens_seen": 427032576
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043968906720160485,
      "loss": 3.101,
      "theoretical_loss": 3.9827962726742117,
      "tokens_seen": 427098112
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043967903711133403,
      "loss": 3.2511,
      "theoretical_loss": 3.982728657429546,
      "tokens_seen": 427163648
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004396690070210632,
      "loss": 3.1731,
      "theoretical_loss": 3.9826610554618083,
      "tokens_seen": 427229184
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004396589769307924,
      "loss": 3.2126,
      "theoretical_loss": 3.982593466766355,
      "tokens_seen": 427294720
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043964894684052157,
      "loss": 3.135,
      "theoretical_loss": 3.982525891338546,
      "tokens_seen": 427360256
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043963891675025075,
      "loss": 3.0124,
      "theoretical_loss": 3.9824583291737423,
      "tokens_seen": 427425792
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043962888665998,
      "loss": 2.9563,
      "theoretical_loss": 3.982390780267308,
      "tokens_seen": 427491328
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004396188565697091,
      "loss": 3.309,
      "theoretical_loss": 3.98232324461461,
      "tokens_seen": 427556864
    },
    {
      "epoch": 0.13,
      "objective/train/advantage_avg": 0.4931120276451111,
      "objective/train/docs_used": 254260,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.9755115509033203,
      "objective/train/original_loss": 2.9755117893218994,
      "objective/train/theoretical_loss": 3.982255722211016,
      "objective/train/tokens_used": 448082400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2432476431131363,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505475997924805,
      "objective/train/weighted_lm_loss": 3.125749349594116,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0408108234405518,
      "theoretical_loss": 3.982255722211016,
      "tokens_seen": 427622400
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043960882647943835,
      "loss": 3.1749,
      "theoretical_loss": 3.982255722211016,
      "tokens_seen": 427622400
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004395987963891675,
      "loss": 3.1814,
      "theoretical_loss": 3.982188213051897,
      "tokens_seen": 427687936
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004395887662988967,
      "loss": 3.1542,
      "theoretical_loss": 3.9821207171326267,
      "tokens_seen": 427753472
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004395787362086259,
      "loss": 3.1534,
      "theoretical_loss": 3.98205323444858,
      "tokens_seen": 427819008
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004395687061183551,
      "loss": 3.0027,
      "theoretical_loss": 3.9819857649951347,
      "tokens_seen": 427884544
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043955867602808426,
      "loss": 3.0427,
      "theoretical_loss": 3.981918308767672,
      "tokens_seen": 427950080
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043954864593781344,
      "loss": 2.9556,
      "theoretical_loss": 3.9818508657615737,
      "tokens_seen": 428015616
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004395386158475426,
      "loss": 2.8758,
      "theoretical_loss": 3.9817834359722246,
      "tokens_seen": 428081152
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043952858575727185,
      "loss": 3.3135,
      "theoretical_loss": 3.981716019395012,
      "tokens_seen": 428146688
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.000439518555667001,
      "loss": 3.1624,
      "theoretical_loss": 3.981648616025325,
      "tokens_seen": 428212224
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004395085255767302,
      "loss": 2.9985,
      "theoretical_loss": 3.9815812258585552,
      "tokens_seen": 428277760
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004394984954864594,
      "loss": 3.0428,
      "theoretical_loss": 3.9815138488900974,
      "tokens_seen": 428343296
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004394884653961886,
      "loss": 3.0017,
      "theoretical_loss": 3.981446485115348,
      "tokens_seen": 428408832
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043947843530591776,
      "loss": 2.918,
      "theoretical_loss": 3.981379134529705,
      "tokens_seen": 428474368
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043946840521564694,
      "loss": 3.0671,
      "theoretical_loss": 3.98131179712857,
      "tokens_seen": 428539904
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004394583751253761,
      "loss": 3.1102,
      "theoretical_loss": 3.981244472907346,
      "tokens_seen": 428605440
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043944834503510536,
      "loss": 2.8949,
      "theoretical_loss": 3.9811771618614387,
      "tokens_seen": 428670976
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004394383149448345,
      "loss": 3.2506,
      "theoretical_loss": 3.981109863986256,
      "tokens_seen": 428736512
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004394282848545637,
      "loss": 2.9902,
      "theoretical_loss": 3.981042579277208,
      "tokens_seen": 428802048
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043941825476429285,
      "loss": 2.9398,
      "theoretical_loss": 3.9809753077297074,
      "tokens_seen": 428867584
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004394082246740221,
      "loss": 2.8982,
      "theoretical_loss": 3.980908049339169,
      "tokens_seen": 428933120
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043939819458375126,
      "loss": 2.9993,
      "theoretical_loss": 3.9808408041010104,
      "tokens_seen": 428998656
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043938816449348044,
      "loss": 3.0545,
      "theoretical_loss": 3.98077357201065,
      "tokens_seen": 429064192
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004393781344032096,
      "loss": 3.2601,
      "theoretical_loss": 3.98070635306351,
      "tokens_seen": 429129728
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004393681043129388,
      "loss": 2.7,
      "theoretical_loss": 3.980639147255015,
      "tokens_seen": 429195264
    },
    {
      "epoch": 0.13,
      "objective/train/advantage_avg": 0.4945264756679535,
      "objective/train/docs_used": 255066,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.8185579776763916,
      "objective/train/original_loss": 2.8185579776763916,
      "objective/train/theoretical_loss": 3.98057195458059,
      "objective/train/tokens_used": 449720800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2446141093969345,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506960153579712,
      "objective/train/weighted_lm_loss": 2.961355686187744,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0417574644088745,
      "theoretical_loss": 3.98057195458059,
      "tokens_seen": 429260800
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.000439358074222668,
      "loss": 2.8913,
      "theoretical_loss": 3.98057195458059,
      "tokens_seen": 429260800
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004393480441323972,
      "loss": 2.8974,
      "theoretical_loss": 3.980504775035665,
      "tokens_seen": 429326336
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043933801404212635,
      "loss": 2.9076,
      "theoretical_loss": 3.98043760861567,
      "tokens_seen": 429391872
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004393279839518556,
      "loss": 2.9287,
      "theoretical_loss": 3.980370455316039,
      "tokens_seen": 429457408
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043931795386158477,
      "loss": 2.9874,
      "theoretical_loss": 3.980303315132206,
      "tokens_seen": 429522944
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043930792377131395,
      "loss": 3.0694,
      "theoretical_loss": 3.9802361880596093,
      "tokens_seen": 429588480
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043929789368104313,
      "loss": 3.2185,
      "theoretical_loss": 3.9801690740936895,
      "tokens_seen": 429654016
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004392878635907723,
      "loss": 3.1335,
      "theoretical_loss": 3.980101973229889,
      "tokens_seen": 429719552
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004392778335005015,
      "loss": 3.2648,
      "theoretical_loss": 3.9800348854636516,
      "tokens_seen": 429785088
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004392678034102307,
      "loss": 3.0478,
      "theoretical_loss": 3.9799678107904244,
      "tokens_seen": 429850624
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043925777331995985,
      "loss": 3.1061,
      "theoretical_loss": 3.979900749205657,
      "tokens_seen": 429916160
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004392477432296891,
      "loss": 3.3991,
      "theoretical_loss": 3.9798337007048,
      "tokens_seen": 429981696
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004392377131394182,
      "loss": 2.864,
      "theoretical_loss": 3.9797666652833072,
      "tokens_seen": 430047232
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043922768304914745,
      "loss": 3.072,
      "theoretical_loss": 3.9796996429366356,
      "tokens_seen": 430112768
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043921765295887663,
      "loss": 2.9304,
      "theoretical_loss": 3.9796326336602426,
      "tokens_seen": 430178304
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004392076228686058,
      "loss": 2.8075,
      "theoretical_loss": 3.9795656374495882,
      "tokens_seen": 430243840
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.000439197592778335,
      "loss": 2.9344,
      "theoretical_loss": 3.9794986543001363,
      "tokens_seen": 430309376
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043918756268806423,
      "loss": 3.077,
      "theoretical_loss": 3.9794316842073516,
      "tokens_seen": 430374912
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004391775325977934,
      "loss": 3.2589,
      "theoretical_loss": 3.9793647271667005,
      "tokens_seen": 430440448
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004391675025075226,
      "loss": 3.2268,
      "theoretical_loss": 3.979297783173654,
      "tokens_seen": 430505984
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043915747241725177,
      "loss": 3.2799,
      "theoretical_loss": 3.979230852223682,
      "tokens_seen": 430571520
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043914744232698095,
      "loss": 2.9662,
      "theoretical_loss": 3.9791639343122602,
      "tokens_seen": 430637056
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004391374122367102,
      "loss": 3.2485,
      "theoretical_loss": 3.979097029434865,
      "tokens_seen": 430702592
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004391273821464393,
      "loss": 3.0161,
      "theoretical_loss": 3.979030137586974,
      "tokens_seen": 430768128
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043911735205616855,
      "loss": 3.3749,
      "theoretical_loss": 3.978963258764068,
      "tokens_seen": 430833664
    },
    {
      "epoch": 0.13,
      "objective/train/advantage_avg": 0.4933794140815735,
      "objective/train/docs_used": 256338,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.7953133583068848,
      "objective/train/original_loss": 2.795313596725464,
      "objective/train/theoretical_loss": 3.9788963929616314,
      "objective/train/tokens_used": 451359200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24351128935813904,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505757331848145,
      "objective/train/weighted_lm_loss": 2.936458110809326,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0338950157165527,
      "theoretical_loss": 3.9788963929616314,
      "tokens_seen": 430899200
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004391073219658977,
      "loss": 3.1078,
      "theoretical_loss": 3.9788963929616314,
      "tokens_seen": 430899200
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004390972918756269,
      "loss": 3.3791,
      "theoretical_loss": 3.9788295401751483,
      "tokens_seen": 430964736
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004390872617853561,
      "loss": 2.9848,
      "theoretical_loss": 3.9787627004001074,
      "tokens_seen": 431030272
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004390772316950853,
      "loss": 2.8944,
      "theoretical_loss": 3.978695873631997,
      "tokens_seen": 431095808
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043906720160481446,
      "loss": 3.1933,
      "theoretical_loss": 3.9786290598663108,
      "tokens_seen": 431161344
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043905717151454364,
      "loss": 3.0159,
      "theoretical_loss": 3.978562259098542,
      "tokens_seen": 431226880
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004390471414242728,
      "loss": 3.1903,
      "theoretical_loss": 3.9784954713241882,
      "tokens_seen": 431292416
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043903711133400205,
      "loss": 3.0117,
      "theoretical_loss": 3.9784286965387468,
      "tokens_seen": 431357952
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004390270812437312,
      "loss": 3.0476,
      "theoretical_loss": 3.9783619347377206,
      "tokens_seen": 431423488
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004390170511534604,
      "loss": 3.2442,
      "theoretical_loss": 3.978295185916612,
      "tokens_seen": 431489024
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004390070210631896,
      "loss": 2.9121,
      "theoretical_loss": 3.9782284500709264,
      "tokens_seen": 431554560
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004389969909729188,
      "loss": 3.1683,
      "theoretical_loss": 3.978161727196172,
      "tokens_seen": 431620096
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043898696088264796,
      "loss": 2.7393,
      "theoretical_loss": 3.978095017287859,
      "tokens_seen": 431685632
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043897693079237714,
      "loss": 2.7922,
      "theoretical_loss": 3.9780283203414992,
      "tokens_seen": 431751168
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004389669007021063,
      "loss": 3.0885,
      "theoretical_loss": 3.9779616363526076,
      "tokens_seen": 431816704
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043895687061183556,
      "loss": 3.1904,
      "theoretical_loss": 3.9778949653167004,
      "tokens_seen": 431882240
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004389468405215647,
      "loss": 2.837,
      "theoretical_loss": 3.977828307229297,
      "tokens_seen": 431947776
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004389368104312939,
      "loss": 3.1507,
      "theoretical_loss": 3.9777616620859186,
      "tokens_seen": 432013312
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043892678034102305,
      "loss": 2.9704,
      "theoretical_loss": 3.9776950298820886,
      "tokens_seen": 432078848
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004389167502507523,
      "loss": 3.2546,
      "theoretical_loss": 3.977628410613332,
      "tokens_seen": 432144384
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043890672016048146,
      "loss": 2.8288,
      "theoretical_loss": 3.9775618042751777,
      "tokens_seen": 432209920
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043889669007021064,
      "loss": 2.9964,
      "theoretical_loss": 3.9774952108631556,
      "tokens_seen": 432275456
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004388866599799398,
      "loss": 2.6385,
      "theoretical_loss": 3.9774286303727973,
      "tokens_seen": 432340992
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.000438876629889669,
      "loss": 3.0167,
      "theoretical_loss": 3.977362062799638,
      "tokens_seen": 432406528
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004388665997993982,
      "loss": 3.2411,
      "theoretical_loss": 3.977295508139215,
      "tokens_seen": 432472064
    },
    {
      "epoch": 0.13,
      "objective/train/advantage_avg": 0.4949647784233093,
      "objective/train/docs_used": 257001,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.2045726776123047,
      "objective/train/original_loss": 2.2045726776123047,
      "objective/train/theoretical_loss": 3.9772289663870657,
      "objective/train/tokens_used": 452997600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24510882794857025,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507426261901855,
      "objective/train/weighted_lm_loss": 2.316154956817627,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0338950157165527,
      "theoretical_loss": 3.9772289663870657,
      "tokens_seen": 432537600
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004388565697091274,
      "loss": 2.7493,
      "theoretical_loss": 3.9772289663870657,
      "tokens_seen": 432537600
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043884653961885655,
      "loss": 3.1328,
      "theoretical_loss": 3.977162437538732,
      "tokens_seen": 432603136
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004388365095285858,
      "loss": 3.1391,
      "theoretical_loss": 3.9770959215897586,
      "tokens_seen": 432668672
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043882647943831497,
      "loss": 3.0681,
      "theoretical_loss": 3.97702941853569,
      "tokens_seen": 432734208
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043881644934804415,
      "loss": 3.1698,
      "theoretical_loss": 3.9769629283720738,
      "tokens_seen": 432799744
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043880641925777333,
      "loss": 3.1544,
      "theoretical_loss": 3.9768964510944604,
      "tokens_seen": 432865280
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004387963891675025,
      "loss": 3.2978,
      "theoretical_loss": 3.9768299866984025,
      "tokens_seen": 432930816
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004387863590772317,
      "loss": 3.268,
      "theoretical_loss": 3.9767635351794532,
      "tokens_seen": 432996352
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004387763289869609,
      "loss": 3.4267,
      "theoretical_loss": 3.976697096533171,
      "tokens_seen": 433061888
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043876629889669005,
      "loss": 3.1618,
      "theoretical_loss": 3.976630670755114,
      "tokens_seen": 433127424
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004387562688064193,
      "loss": 3.2312,
      "theoretical_loss": 3.9765642578408427,
      "tokens_seen": 433192960
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004387462387161484,
      "loss": 3.1994,
      "theoretical_loss": 3.9764978577859216,
      "tokens_seen": 433258496
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043873620862587765,
      "loss": 2.8955,
      "theoretical_loss": 3.976431470585915,
      "tokens_seen": 433324032
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043872617853560683,
      "loss": 2.9661,
      "theoretical_loss": 3.9763650962363917,
      "tokens_seen": 433389568
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.000438716148445336,
      "loss": 2.9687,
      "theoretical_loss": 3.9762987347329206,
      "tokens_seen": 433455104
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004387061183550652,
      "loss": 3.076,
      "theoretical_loss": 3.976232386071074,
      "tokens_seen": 433520640
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043869608826479443,
      "loss": 3.0578,
      "theoretical_loss": 3.976166050246427,
      "tokens_seen": 433586176
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043868605817452356,
      "loss": 3.1327,
      "theoretical_loss": 3.9760997272545557,
      "tokens_seen": 433651712
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004386760280842528,
      "loss": 2.8302,
      "theoretical_loss": 3.9760334170910383,
      "tokens_seen": 433717248
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004386659979939819,
      "loss": 3.1258,
      "theoretical_loss": 3.9759671197514557,
      "tokens_seen": 433782784
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043865596790371115,
      "loss": 3.307,
      "theoretical_loss": 3.975900835231391,
      "tokens_seen": 433848320
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043864593781344033,
      "loss": 3.2526,
      "theoretical_loss": 3.9758345635264307,
      "tokens_seen": 433913856
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004386359077231695,
      "loss": 3.0724,
      "theoretical_loss": 3.9757683046321604,
      "tokens_seen": 433979392
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004386258776328987,
      "loss": 3.0124,
      "theoretical_loss": 3.9757020585441705,
      "tokens_seen": 434044928
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004386158475426279,
      "loss": 3.0301,
      "theoretical_loss": 3.975635825258053,
      "tokens_seen": 434110464
    },
    {
      "epoch": 0.13,
      "objective/train/advantage_avg": 0.4952355921268463,
      "objective/train/docs_used": 257982,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.5731234550476074,
      "objective/train/original_loss": 2.5731232166290283,
      "objective/train/theoretical_loss": 3.975569604769402,
      "objective/train/tokens_used": 454636000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24530571699142456,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507705211639404,
      "objective/train/weighted_lm_loss": 2.7036213874816895,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0417574644088745,
      "theoretical_loss": 3.975569604769402,
      "tokens_seen": 434176000
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043860581745235706,
      "loss": 3.058,
      "theoretical_loss": 3.975569604769402,
      "tokens_seen": 434176000
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004385957873620863,
      "loss": 3.1037,
      "theoretical_loss": 3.9755033970738127,
      "tokens_seen": 434241536
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004385857572718154,
      "loss": 3.3072,
      "theoretical_loss": 3.9754372021668845,
      "tokens_seen": 434307072
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043857572718154466,
      "loss": 2.9794,
      "theoretical_loss": 3.975371020044218,
      "tokens_seen": 434372608
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004385656970912738,
      "loss": 2.9723,
      "theoretical_loss": 3.9753048507014146,
      "tokens_seen": 434438144
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.000438555667001003,
      "loss": 3.0335,
      "theoretical_loss": 3.9752386941340805,
      "tokens_seen": 434503680
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004385456369107322,
      "loss": 2.9481,
      "theoretical_loss": 3.975172550337822,
      "tokens_seen": 434569216
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004385356068204614,
      "loss": 3.2772,
      "theoretical_loss": 3.975106419308249,
      "tokens_seen": 434634752
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043852557673019056,
      "loss": 3.216,
      "theoretical_loss": 3.9750403010409716,
      "tokens_seen": 434700288
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004385155466399198,
      "loss": 3.2404,
      "theoretical_loss": 3.9749741955316056,
      "tokens_seen": 434765824
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004385055165496489,
      "loss": 3.2176,
      "theoretical_loss": 3.974908102775764,
      "tokens_seen": 434831360
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043849548645937816,
      "loss": 3.1916,
      "theoretical_loss": 3.9748420227690673,
      "tokens_seen": 434896896
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004384854563691073,
      "loss": 3.0635,
      "theoretical_loss": 3.9747759555071336,
      "tokens_seen": 434962432
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004384754262788365,
      "loss": 3.2103,
      "theoretical_loss": 3.974709900985586,
      "tokens_seen": 435027968
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004384653961885657,
      "loss": 3.2702,
      "theoretical_loss": 3.9746438592000484,
      "tokens_seen": 435093504
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004384553660982949,
      "loss": 3.0074,
      "theoretical_loss": 3.9745778301461483,
      "tokens_seen": 435159040
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043844533600802407,
      "loss": 2.8953,
      "theoretical_loss": 3.9745118138195137,
      "tokens_seen": 435224576
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043843530591775325,
      "loss": 3.0828,
      "theoretical_loss": 3.9744458102157756,
      "tokens_seen": 435290112
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004384252758274825,
      "loss": 3.1016,
      "theoretical_loss": 3.974379819330567,
      "tokens_seen": 435355648
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043841524573721166,
      "loss": 3.1261,
      "theoretical_loss": 3.974313841159523,
      "tokens_seen": 435421184
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043840521564694084,
      "loss": 2.6697,
      "theoretical_loss": 3.974247875698281,
      "tokens_seen": 435486720
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043839518555667,
      "loss": 3.0705,
      "theoretical_loss": 3.974181922942481,
      "tokens_seen": 435552256
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004383851554663992,
      "loss": 3.0162,
      "theoretical_loss": 3.9741159828877635,
      "tokens_seen": 435617792
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004383751253761284,
      "loss": 3.2224,
      "theoretical_loss": 3.9740500555297738,
      "tokens_seen": 435683328
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004383650952858576,
      "loss": 3.1517,
      "theoretical_loss": 3.9739841408641565,
      "tokens_seen": 435748864
    },
    {
      "epoch": 0.13,
      "objective/train/advantage_avg": 0.49500441551208496,
      "objective/train/docs_used": 258654,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.462207317352295,
      "objective/train/original_loss": 2.462207555770874,
      "objective/train/theoretical_loss": 3.9739182388865606,
      "objective/train/tokens_used": 456274400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24508091807365417,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507464408874512,
      "objective/train/weighted_lm_loss": 2.586916446685791,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0417574644088745,
      "theoretical_loss": 3.9739182388865606,
      "tokens_seen": 435814400
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043835506519558675,
      "loss": 2.8955,
      "theoretical_loss": 3.9739182388865606,
      "tokens_seen": 435814400
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.000438345035105316,
      "loss": 3.1225,
      "theoretical_loss": 3.9738523495926357,
      "tokens_seen": 435879936
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043833500501504517,
      "loss": 3.1947,
      "theoretical_loss": 3.9737864729780346,
      "tokens_seen": 435945472
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043832497492477435,
      "loss": 2.9005,
      "theoretical_loss": 3.973720609038412,
      "tokens_seen": 436011008
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043831494483450353,
      "loss": 3.0745,
      "theoretical_loss": 3.9736547577694235,
      "tokens_seen": 436076544
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004383049147442327,
      "loss": 3.2507,
      "theoretical_loss": 3.9735889191667293,
      "tokens_seen": 436142080
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004382948846539619,
      "loss": 3.121,
      "theoretical_loss": 3.9735230932259893,
      "tokens_seen": 436207616
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004382848545636911,
      "loss": 2.8564,
      "theoretical_loss": 3.973457279942868,
      "tokens_seen": 436273152
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043827482447342025,
      "loss": 3.2292,
      "theoretical_loss": 3.9733914793130287,
      "tokens_seen": 436338688
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004382647943831495,
      "loss": 3.2003,
      "theoretical_loss": 3.9733256913321395,
      "tokens_seen": 436404224
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004382547642928786,
      "loss": 3.0587,
      "theoretical_loss": 3.973259915995871,
      "tokens_seen": 436469760
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043824473420260785,
      "loss": 2.7753,
      "theoretical_loss": 3.973194153299893,
      "tokens_seen": 436535296
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043823470411233703,
      "loss": 2.9775,
      "theoretical_loss": 3.9731284032398806,
      "tokens_seen": 436600832
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004382246740220662,
      "loss": 2.9701,
      "theoretical_loss": 3.9730626658115096,
      "tokens_seen": 436666368
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004382146439317954,
      "loss": 2.8682,
      "theoretical_loss": 3.9729969410104573,
      "tokens_seen": 436731904
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043820461384152463,
      "loss": 3.2346,
      "theoretical_loss": 3.9729312288324037,
      "tokens_seen": 436797440
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043819458375125376,
      "loss": 3.0309,
      "theoretical_loss": 3.972865529273032,
      "tokens_seen": 436862976
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.000438184553660983,
      "loss": 3.2558,
      "theoretical_loss": 3.972799842328027,
      "tokens_seen": 436928512
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004381745235707121,
      "loss": 3.0582,
      "theoretical_loss": 3.9727341679930737,
      "tokens_seen": 436994048
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043816449348044135,
      "loss": 3.1033,
      "theoretical_loss": 3.9726685062638616,
      "tokens_seen": 437059584
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043815446339017053,
      "loss": 3.3182,
      "theoretical_loss": 3.9726028571360814,
      "tokens_seen": 437125120
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004381444332998997,
      "loss": 3.0783,
      "theoretical_loss": 3.972537220605426,
      "tokens_seen": 437190656
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004381344032096289,
      "loss": 3.1737,
      "theoretical_loss": 3.9724715966675896,
      "tokens_seen": 437256192
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004381243731193581,
      "loss": 3.2809,
      "theoretical_loss": 3.9724059853182707,
      "tokens_seen": 437321728
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043811434302908726,
      "loss": 3.2315,
      "theoretical_loss": 3.9723403865531677,
      "tokens_seen": 437387264
    },
    {
      "epoch": 0.13,
      "objective/train/advantage_avg": 0.4903266429901123,
      "objective/train/docs_used": 259916,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.5855929851531982,
      "objective/train/original_loss": 2.5855929851531982,
      "objective/train/theoretical_loss": 3.9722748003679818,
      "objective/train/tokens_used": 457912800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2406104952096939,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0502556562423706,
      "objective/train/weighted_lm_loss": 2.715423583984375,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0304546356201172,
      "theoretical_loss": 3.9722748003679818,
      "tokens_seen": 437452800
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004381043129388165,
      "loss": 3.2693,
      "theoretical_loss": 3.9722748003679818,
      "tokens_seen": 437452800
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004380942828485456,
      "loss": 2.886,
      "theoretical_loss": 3.972209226758417,
      "tokens_seen": 437518336
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043808425275827486,
      "loss": 2.8721,
      "theoretical_loss": 3.972143665720178,
      "tokens_seen": 437583872
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.000438074222668004,
      "loss": 3.1258,
      "theoretical_loss": 3.9720781172489734,
      "tokens_seen": 437649408
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004380641925777332,
      "loss": 3.0995,
      "theoretical_loss": 3.972012581340513,
      "tokens_seen": 437714944
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004380541624874624,
      "loss": 3.4119,
      "theoretical_loss": 3.9719470579905076,
      "tokens_seen": 437780480
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004380441323971916,
      "loss": 3.2205,
      "theoretical_loss": 3.9718815471946725,
      "tokens_seen": 437846016
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043803410230692076,
      "loss": 3.0626,
      "theoretical_loss": 3.971816048948723,
      "tokens_seen": 437911552
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043802407221665,
      "loss": 3.3305,
      "theoretical_loss": 3.9717505632483774,
      "tokens_seen": 437977088
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004380140421263791,
      "loss": 3.2086,
      "theoretical_loss": 3.9716850900893563,
      "tokens_seen": 438042624
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043800401203610836,
      "loss": 3.3768,
      "theoretical_loss": 3.971619629467382,
      "tokens_seen": 438108160
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004379939819458375,
      "loss": 3.2621,
      "theoretical_loss": 3.971554181378178,
      "tokens_seen": 438173696
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004379839518555667,
      "loss": 2.9302,
      "theoretical_loss": 3.9714887458174726,
      "tokens_seen": 438239232
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004379739217652959,
      "loss": 3.2557,
      "theoretical_loss": 3.9714233227809936,
      "tokens_seen": 438304768
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004379638916750251,
      "loss": 2.9548,
      "theoretical_loss": 3.9713579122644718,
      "tokens_seen": 438370304
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043795386158475427,
      "loss": 2.9004,
      "theoretical_loss": 3.97129251426364,
      "tokens_seen": 438435840
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043794383149448345,
      "loss": 3.1027,
      "theoretical_loss": 3.971227128774234,
      "tokens_seen": 438501376
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043793380140421263,
      "loss": 3.3575,
      "theoretical_loss": 3.971161755791989,
      "tokens_seen": 438566912
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043792377131394186,
      "loss": 3.0241,
      "theoretical_loss": 3.971096395312646,
      "tokens_seen": 438632448
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.000437913741223671,
      "loss": 2.9491,
      "theoretical_loss": 3.9710310473319455,
      "tokens_seen": 438697984
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004379037111334002,
      "loss": 3.0428,
      "theoretical_loss": 3.970965711845631,
      "tokens_seen": 438763520
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043789368104312935,
      "loss": 3.1834,
      "theoretical_loss": 3.970900388849447,
      "tokens_seen": 438829056
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004378836509528586,
      "loss": 3.0194,
      "theoretical_loss": 3.970835078339143,
      "tokens_seen": 438894592
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043787362086258777,
      "loss": 3.3237,
      "theoretical_loss": 3.9707697803104667,
      "tokens_seen": 438960128
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043786359077231695,
      "loss": 3.1495,
      "theoretical_loss": 3.970704494759171,
      "tokens_seen": 439025664
    },
    {
      "epoch": 0.13,
      "objective/train/advantage_avg": 0.4948321282863617,
      "objective/train/docs_used": 260524,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.3129186630249023,
      "objective/train/original_loss": 2.3129186630249023,
      "objective/train/theoretical_loss": 3.9706392216810085,
      "objective/train/tokens_used": 459551200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24492323398590088,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507283210754395,
      "objective/train/weighted_lm_loss": 2.430363893508911,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0417574644088745,
      "theoretical_loss": 3.9706392216810085,
      "tokens_seen": 439091200
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043785356068204613,
      "loss": 3.0167,
      "theoretical_loss": 3.9706392216810085,
      "tokens_seen": 439091200
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043784353059177537,
      "loss": 3.042,
      "theoretical_loss": 3.970573961071736,
      "tokens_seen": 439156736
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004378335005015045,
      "loss": 3.0537,
      "theoretical_loss": 3.97050871292711,
      "tokens_seen": 439222272
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043782347041123373,
      "loss": 3.2326,
      "theoretical_loss": 3.9704434772428923,
      "tokens_seen": 439287808
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043781344032096286,
      "loss": 3.0659,
      "theoretical_loss": 3.970378254014844,
      "tokens_seen": 439353344
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004378034102306921,
      "loss": 2.8753,
      "theoretical_loss": 3.9703130432387295,
      "tokens_seen": 439418880
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043779338014042127,
      "loss": 3.0698,
      "theoretical_loss": 3.9702478449103142,
      "tokens_seen": 439484416
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043778335005015045,
      "loss": 3.2233,
      "theoretical_loss": 3.9701826590253675,
      "tokens_seen": 439549952
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043777331995987963,
      "loss": 2.914,
      "theoretical_loss": 3.9701174855796593,
      "tokens_seen": 439615488
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004377632898696088,
      "loss": 3.095,
      "theoretical_loss": 3.970052324568961,
      "tokens_seen": 439681024
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.000437753259779338,
      "loss": 3.2278,
      "theoretical_loss": 3.969987175989049,
      "tokens_seen": 439746560
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043774322968906723,
      "loss": 3.0707,
      "theoretical_loss": 3.9699220398356982,
      "tokens_seen": 439812096
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043773319959879636,
      "loss": 3.0017,
      "theoretical_loss": 3.9698569161046877,
      "tokens_seen": 439877632
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004377231695085256,
      "loss": 3.1599,
      "theoretical_loss": 3.9697918047917984,
      "tokens_seen": 439943168
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004377131394182547,
      "loss": 3.1258,
      "theoretical_loss": 3.969726705892813,
      "tokens_seen": 440008704
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043770310932798396,
      "loss": 3.1931,
      "theoretical_loss": 3.9696616194035164,
      "tokens_seen": 440074240
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043769307923771314,
      "loss": 3.2469,
      "theoretical_loss": 3.9695965453196944,
      "tokens_seen": 440139776
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004376830491474423,
      "loss": 2.7644,
      "theoretical_loss": 3.9695314836371365,
      "tokens_seen": 440205312
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043767301905717155,
      "loss": 3.0693,
      "theoretical_loss": 3.969466434351635,
      "tokens_seen": 440270848
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043766298896690073,
      "loss": 3.1413,
      "theoretical_loss": 3.969401397458981,
      "tokens_seen": 440336384
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004376529588766299,
      "loss": 3.1716,
      "theoretical_loss": 3.96933637295497,
      "tokens_seen": 440401920
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004376429287863591,
      "loss": 2.9632,
      "theoretical_loss": 3.9692713608353998,
      "tokens_seen": 440467456
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004376328986960883,
      "loss": 3.053,
      "theoretical_loss": 3.969206361096069,
      "tokens_seen": 440532992
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043762286860581746,
      "loss": 3.1438,
      "theoretical_loss": 3.9691413737327785,
      "tokens_seen": 440598528
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004376128385155467,
      "loss": 2.9921,
      "theoretical_loss": 3.9690763987413327,
      "tokens_seen": 440664064
    },
    {
      "epoch": 0.13,
      "objective/train/advantage_avg": 0.49579471349716187,
      "objective/train/docs_used": 261739,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.772521495819092,
      "objective/train/original_loss": 2.7725212574005127,
      "objective/train/theoretical_loss": 3.969011436117536,
      "objective/train/tokens_used": 461189600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24586226046085358,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0508291721343994,
      "objective/train/weighted_lm_loss": 2.913348436355591,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0373471975326538,
      "theoretical_loss": 3.969011436117536,
      "tokens_seen": 440729600
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004376028084252758,
      "loss": 3.1322,
      "theoretical_loss": 3.969011436117536,
      "tokens_seen": 440729600
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043759277833500506,
      "loss": 3.1262,
      "theoretical_loss": 3.968946485857196,
      "tokens_seen": 440795136
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004375827482447342,
      "loss": 2.7114,
      "theoretical_loss": 3.968881547956122,
      "tokens_seen": 440860672
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004375727181544634,
      "loss": 3.1572,
      "theoretical_loss": 3.9688166224101256,
      "tokens_seen": 440926208
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004375626880641926,
      "loss": 3.1821,
      "theoretical_loss": 3.96875170921502,
      "tokens_seen": 440991744
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004375526579739218,
      "loss": 3.2413,
      "theoretical_loss": 3.968686808366621,
      "tokens_seen": 441057280
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043754262788365096,
      "loss": 3.0884,
      "theoretical_loss": 3.9686219198607464,
      "tokens_seen": 441122816
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004375325977933802,
      "loss": 2.9333,
      "theoretical_loss": 3.968557043693215,
      "tokens_seen": 441188352
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004375225677031093,
      "loss": 3.1214,
      "theoretical_loss": 3.968492179859849,
      "tokens_seen": 441253888
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043751253761283856,
      "loss": 3.1402,
      "theoretical_loss": 3.968427328356472,
      "tokens_seen": 441319424
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004375025075225677,
      "loss": 3.1407,
      "theoretical_loss": 3.9683624891789098,
      "tokens_seen": 441384960
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004374924774322969,
      "loss": 3.3146,
      "theoretical_loss": 3.96829766232299,
      "tokens_seen": 441450496
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004374824473420261,
      "loss": 3.101,
      "theoretical_loss": 3.9682328477845425,
      "tokens_seen": 441516032
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004374724172517553,
      "loss": 3.2336,
      "theoretical_loss": 3.9681680455593984,
      "tokens_seen": 441581568
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043746238716148447,
      "loss": 3.2437,
      "theoretical_loss": 3.968103255643393,
      "tokens_seen": 441647104
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043745235707121365,
      "loss": 3.2592,
      "theoretical_loss": 3.9680384780323603,
      "tokens_seen": 441712640
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043744232698094283,
      "loss": 3.2296,
      "theoretical_loss": 3.9679737127221397,
      "tokens_seen": 441778176
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043743229689067206,
      "loss": 3.1343,
      "theoretical_loss": 3.9679089597085704,
      "tokens_seen": 441843712
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004374222668004012,
      "loss": 3.1037,
      "theoretical_loss": 3.9678442189874947,
      "tokens_seen": 441909248
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004374122367101304,
      "loss": 3.0624,
      "theoretical_loss": 3.967779490554756,
      "tokens_seen": 441974784
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043740220661985955,
      "loss": 3.0396,
      "theoretical_loss": 3.967714774406201,
      "tokens_seen": 442040320
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004373921765295888,
      "loss": 3.3268,
      "theoretical_loss": 3.967650070537677,
      "tokens_seen": 442105856
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043738214643931797,
      "loss": 3.252,
      "theoretical_loss": 3.9675853789450346,
      "tokens_seen": 442171392
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043737211634904715,
      "loss": 3.0547,
      "theoretical_loss": 3.9675206996241252,
      "tokens_seen": 442236928
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043736208625877633,
      "loss": 2.9607,
      "theoretical_loss": 3.9674560325708037,
      "tokens_seen": 442302464
    },
    {
      "epoch": 0.13,
      "objective/train/advantage_avg": 0.4929172396659851,
      "objective/train/docs_used": 262441,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.428363084793091,
      "objective/train/original_loss": 2.428363084793091,
      "objective/train/theoretical_loss": 3.9673913777809253,
      "objective/train/tokens_used": 462828000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2430526167154312,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505272150039673,
      "objective/train/weighted_lm_loss": 2.55107045173645,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0408108234405518,
      "theoretical_loss": 3.9673913777809253,
      "tokens_seen": 442368000
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043735205616850557,
      "loss": 2.9575,
      "theoretical_loss": 3.9673913777809253,
      "tokens_seen": 442368000
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004373420260782347,
      "loss": 3.1343,
      "theoretical_loss": 3.9673267352503485,
      "tokens_seen": 442433536
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043733199598796393,
      "loss": 2.5887,
      "theoretical_loss": 3.9672621049749335,
      "tokens_seen": 442499072
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043732196589769306,
      "loss": 3.1448,
      "theoretical_loss": 3.9671974869505418,
      "tokens_seen": 442564608
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004373119358074223,
      "loss": 2.9872,
      "theoretical_loss": 3.967132881173038,
      "tokens_seen": 442630144
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043730190571715147,
      "loss": 3.1224,
      "theoretical_loss": 3.9670682876382886,
      "tokens_seen": 442695680
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043729187562688065,
      "loss": 2.9149,
      "theoretical_loss": 3.967003706342161,
      "tokens_seen": 442761216
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043728184553660983,
      "loss": 3.1669,
      "theoretical_loss": 3.9669391372805256,
      "tokens_seen": 442826752
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.000437271815446339,
      "loss": 2.8606,
      "theoretical_loss": 3.9668745804492542,
      "tokens_seen": 442892288
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004372617853560682,
      "loss": 2.9056,
      "theoretical_loss": 3.9668100358442215,
      "tokens_seen": 442957824
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043725175526579743,
      "loss": 2.9594,
      "theoretical_loss": 3.9667455034613033,
      "tokens_seen": 443023360
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043724172517552656,
      "loss": 2.8408,
      "theoretical_loss": 3.966680983296378,
      "tokens_seen": 443088896
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004372316950852558,
      "loss": 3.0439,
      "theoretical_loss": 3.966616475345326,
      "tokens_seen": 443154432
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004372216649949849,
      "loss": 3.1564,
      "theoretical_loss": 3.966551979604028,
      "tokens_seen": 443219968
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043721163490471416,
      "loss": 3.3222,
      "theoretical_loss": 3.96648749606837,
      "tokens_seen": 443285504
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043720160481444334,
      "loss": 3.0879,
      "theoretical_loss": 3.9664230247342367,
      "tokens_seen": 443351040
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004371915747241725,
      "loss": 3.0978,
      "theoretical_loss": 3.966358565597517,
      "tokens_seen": 443416576
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004371815446339017,
      "loss": 3.0486,
      "theoretical_loss": 3.9662941186541008,
      "tokens_seen": 443482112
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043717151454363093,
      "loss": 2.96,
      "theoretical_loss": 3.96622968389988,
      "tokens_seen": 443547648
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043716148445336006,
      "loss": 3.223,
      "theoretical_loss": 3.9661652613307496,
      "tokens_seen": 443613184
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004371514543630893,
      "loss": 2.8743,
      "theoretical_loss": 3.9661008509426043,
      "tokens_seen": 443678720
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004371414242728184,
      "loss": 3.0918,
      "theoretical_loss": 3.9660364527313434,
      "tokens_seen": 443744256
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043713139418254766,
      "loss": 3.2612,
      "theoretical_loss": 3.965972066692866,
      "tokens_seen": 443809792
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043712136409227684,
      "loss": 3.1438,
      "theoretical_loss": 3.965907692823075,
      "tokens_seen": 443875328
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.000437111334002006,
      "loss": 2.994,
      "theoretical_loss": 3.965843331117874,
      "tokens_seen": 443940864
    },
    {
      "epoch": 0.13,
      "objective/train/advantage_avg": 0.49212518334388733,
      "objective/train/docs_used": 263839,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.1218082904815674,
      "objective/train/original_loss": 3.1218080520629883,
      "objective/train/theoretical_loss": 3.9657789815731688,
      "objective/train/tokens_used": 464466400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24240030348300934,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0504446029663086,
      "objective/train/weighted_lm_loss": 3.280090093612671,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0363596677780151,
      "theoretical_loss": 3.9657789815731688,
      "tokens_seen": 444006400
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004371013039117352,
      "loss": 3.053,
      "theoretical_loss": 3.9657789815731688,
      "tokens_seen": 444006400
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004370912738214644,
      "loss": 3.0526,
      "theoretical_loss": 3.9657146441848674,
      "tokens_seen": 444071936
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043708124373119356,
      "loss": 3.1389,
      "theoretical_loss": 3.96565031894888,
      "tokens_seen": 444137472
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004370712136409228,
      "loss": 3.1749,
      "theoretical_loss": 3.965586005861119,
      "tokens_seen": 444203008
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004370611835506519,
      "loss": 2.8786,
      "theoretical_loss": 3.965521704917498,
      "tokens_seen": 444268544
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043705115346038116,
      "loss": 3.0864,
      "theoretical_loss": 3.9654574161139324,
      "tokens_seen": 444334080
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004370411233701103,
      "loss": 3.1319,
      "theoretical_loss": 3.965393139446341,
      "tokens_seen": 444399616
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004370310932798395,
      "loss": 3.1544,
      "theoretical_loss": 3.965328874910643,
      "tokens_seen": 444465152
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004370210631895687,
      "loss": 2.8823,
      "theoretical_loss": 3.96526462250276,
      "tokens_seen": 444530688
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004370110330992979,
      "loss": 3.1216,
      "theoretical_loss": 3.9652003822186166,
      "tokens_seen": 444596224
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043700100300902707,
      "loss": 3.306,
      "theoretical_loss": 3.9651361540541386,
      "tokens_seen": 444661760
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004369909729187563,
      "loss": 3.1011,
      "theoretical_loss": 3.965071938005252,
      "tokens_seen": 444727296
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043698094282848543,
      "loss": 3.0097,
      "theoretical_loss": 3.965007734067889,
      "tokens_seen": 444792832
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043697091273821467,
      "loss": 3.1305,
      "theoretical_loss": 3.96494354223798,
      "tokens_seen": 444858368
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004369608826479438,
      "loss": 3.1976,
      "theoretical_loss": 3.964879362511459,
      "tokens_seen": 444923904
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043695085255767303,
      "loss": 3.0934,
      "theoretical_loss": 3.9648151948842605,
      "tokens_seen": 444989440
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004369408224674022,
      "loss": 2.9936,
      "theoretical_loss": 3.9647510393523238,
      "tokens_seen": 445054976
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004369307923771314,
      "loss": 3.1635,
      "theoretical_loss": 3.9646868959115875,
      "tokens_seen": 445120512
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.0004369207622868606,
      "loss": 3.0925,
      "theoretical_loss": 3.9646227645579932,
      "tokens_seen": 445186048
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043691073219658975,
      "loss": 3.2439,
      "theoretical_loss": 3.9645586452874846,
      "tokens_seen": 445251584
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.000436900702106319,
      "loss": 2.9022,
      "theoretical_loss": 3.964494538096006,
      "tokens_seen": 445317120
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043689067201604817,
      "loss": 3.0464,
      "theoretical_loss": 3.9644304429795065,
      "tokens_seen": 445382656
    },
    {
      "epoch": 0.13,
      "learning_rate": 0.00043688064192577735,
      "loss": 3.1701,
      "theoretical_loss": 3.9643663599339343,
      "tokens_seen": 445448192
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043687061183550653,
      "loss": 3.24,
      "theoretical_loss": 3.9643022889552415,
      "tokens_seen": 445513728
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043686058174523577,
      "loss": 2.9572,
      "theoretical_loss": 3.9642382300393804,
      "tokens_seen": 445579264
    },
    {
      "epoch": 0.14,
      "objective/train/advantage_avg": 0.493392676115036,
      "objective/train/docs_used": 264540,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.11641001701355,
      "objective/train/original_loss": 3.1164095401763916,
      "objective/train/theoretical_loss": 3.9641741831823065,
      "objective/train/tokens_used": 466104800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24354027211666107,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050577163696289,
      "objective/train/weighted_lm_loss": 3.274244546890259,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0253151655197144,
      "theoretical_loss": 3.9641741831823065,
      "tokens_seen": 445644800
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004368505516549649,
      "loss": 3.1733,
      "theoretical_loss": 3.9641741831823065,
      "tokens_seen": 445644800
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043684052156469413,
      "loss": 3.0651,
      "theoretical_loss": 3.964110148379977,
      "tokens_seen": 445710336
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043683049147442326,
      "loss": 2.9245,
      "theoretical_loss": 3.964046125628351,
      "tokens_seen": 445775872
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004368204613841525,
      "loss": 2.8201,
      "theoretical_loss": 3.96398211492339,
      "tokens_seen": 445841408
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043681043129388167,
      "loss": 3.1293,
      "theoretical_loss": 3.9639181162610564,
      "tokens_seen": 445906944
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043680040120361085,
      "loss": 3.0946,
      "theoretical_loss": 3.9638541296373155,
      "tokens_seen": 445972480
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043679037111334003,
      "loss": 2.868,
      "theoretical_loss": 3.963790155048133,
      "tokens_seen": 446038016
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004367803410230692,
      "loss": 3.3765,
      "theoretical_loss": 3.9637261924894798,
      "tokens_seen": 446103552
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004367703109327984,
      "loss": 3.2156,
      "theoretical_loss": 3.9636622419573246,
      "tokens_seen": 446169088
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043676028084252763,
      "loss": 3.1658,
      "theoretical_loss": 3.963598303447642,
      "tokens_seen": 446234624
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043675025075225676,
      "loss": 2.8691,
      "theoretical_loss": 3.9635343769564053,
      "tokens_seen": 446300160
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.000436740220661986,
      "loss": 2.9416,
      "theoretical_loss": 3.9634704624795916,
      "tokens_seen": 446365696
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004367301905717151,
      "loss": 3.2271,
      "theoretical_loss": 3.963406560013179,
      "tokens_seen": 446431232
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043672016048144436,
      "loss": 3.2104,
      "theoretical_loss": 3.9633426695531484,
      "tokens_seen": 446496768
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043671013039117354,
      "loss": 2.7949,
      "theoretical_loss": 3.963278791095482,
      "tokens_seen": 446562304
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004367001003009027,
      "loss": 3.0485,
      "theoretical_loss": 3.9632149246361643,
      "tokens_seen": 446627840
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004366900702106319,
      "loss": 3.1938,
      "theoretical_loss": 3.9631510701711816,
      "tokens_seen": 446693376
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043668004012036113,
      "loss": 2.9694,
      "theoretical_loss": 3.963087227696521,
      "tokens_seen": 446758912
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043667001003009026,
      "loss": 3.2737,
      "theoretical_loss": 3.9630233972081745,
      "tokens_seen": 446824448
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004366599799398195,
      "loss": 3.0751,
      "theoretical_loss": 3.962959578702133,
      "tokens_seen": 446889984
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004366499498495486,
      "loss": 3.0227,
      "theoretical_loss": 3.9628957721743907,
      "tokens_seen": 446955520
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043663991975927786,
      "loss": 3.2691,
      "theoretical_loss": 3.962831977620943,
      "tokens_seen": 447021056
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043662988966900704,
      "loss": 2.9762,
      "theoretical_loss": 3.962768195037788,
      "tokens_seen": 447086592
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004366198595787362,
      "loss": 3.1099,
      "theoretical_loss": 3.9627044244209264,
      "tokens_seen": 447152128
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004366098294884654,
      "loss": 2.9972,
      "theoretical_loss": 3.962640665766359,
      "tokens_seen": 447217664
    },
    {
      "epoch": 0.14,
      "objective/train/advantage_avg": 0.49465104937553406,
      "objective/train/docs_used": 265849,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.858397960662842,
      "objective/train/original_loss": 2.858397960662842,
      "objective/train/theoretical_loss": 3.962576919070089,
      "objective/train/tokens_used": 467743200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2447502613067627,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507091283798218,
      "objective/train/weighted_lm_loss": 3.003237724304199,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 3.962576919070089,
      "tokens_seen": 447283200
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004365997993981946,
      "loss": 3.0166,
      "theoretical_loss": 3.962576919070089,
      "tokens_seen": 447283200
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043658976930792376,
      "loss": 3.1482,
      "theoretical_loss": 3.962513184328122,
      "tokens_seen": 447348736
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.000436579739217653,
      "loss": 2.7809,
      "theoretical_loss": 3.9624494615364667,
      "tokens_seen": 447414272
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043656970912738213,
      "loss": 3.1861,
      "theoretical_loss": 3.9623857506911313,
      "tokens_seen": 447479808
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043655967903711136,
      "loss": 3.0774,
      "theoretical_loss": 3.9623220517881275,
      "tokens_seen": 447545344
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004365496489468405,
      "loss": 3.0244,
      "theoretical_loss": 3.962258364823468,
      "tokens_seen": 447610880
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004365396188565697,
      "loss": 3.2293,
      "theoretical_loss": 3.9621946897931686,
      "tokens_seen": 447676416
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004365295887662989,
      "loss": 2.971,
      "theoretical_loss": 3.9621310266932457,
      "tokens_seen": 447741952
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004365195586760281,
      "loss": 3.3178,
      "theoretical_loss": 3.9620673755197187,
      "tokens_seen": 447807488
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043650952858575727,
      "loss": 3.1172,
      "theoretical_loss": 3.962003736268608,
      "tokens_seen": 447873024
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004364994984954865,
      "loss": 3.0086,
      "theoretical_loss": 3.961940108935937,
      "tokens_seen": 447938560
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043648946840521563,
      "loss": 2.8635,
      "theoretical_loss": 3.961876493517729,
      "tokens_seen": 448004096
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043647943831494487,
      "loss": 3.1213,
      "theoretical_loss": 3.9618128900100125,
      "tokens_seen": 448069632
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.000436469408224674,
      "loss": 2.9658,
      "theoretical_loss": 3.9617492984088143,
      "tokens_seen": 448135168
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043645937813440323,
      "loss": 3.2501,
      "theoretical_loss": 3.9616857187101653,
      "tokens_seen": 448200704
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004364493480441324,
      "loss": 3.158,
      "theoretical_loss": 3.961622150910098,
      "tokens_seen": 448266240
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004364393179538616,
      "loss": 3.0148,
      "theoretical_loss": 3.9615585950046466,
      "tokens_seen": 448331776
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043642928786359077,
      "loss": 2.775,
      "theoretical_loss": 3.961495050989847,
      "tokens_seen": 448397312
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043641925777331995,
      "loss": 3.0015,
      "theoretical_loss": 3.9614315188617377,
      "tokens_seen": 448462848
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043640922768304913,
      "loss": 3.0162,
      "theoretical_loss": 3.9613679986163572,
      "tokens_seen": 448528384
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043639919759277837,
      "loss": 3.0439,
      "theoretical_loss": 3.9613044902497485,
      "tokens_seen": 448593920
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004363891675025075,
      "loss": 3.1709,
      "theoretical_loss": 3.961240993757955,
      "tokens_seen": 448659456
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043637913741223673,
      "loss": 3.0896,
      "theoretical_loss": 3.9611775091370225,
      "tokens_seen": 448724992
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004363691073219659,
      "loss": 3.0575,
      "theoretical_loss": 3.9611140363829977,
      "tokens_seen": 448790528
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004363590772316951,
      "loss": 3.0301,
      "theoretical_loss": 3.9610505754919307,
      "tokens_seen": 448856064
    },
    {
      "epoch": 0.14,
      "objective/train/advantage_avg": 0.49354153871536255,
      "objective/train/docs_used": 266141,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.0965635776519775,
      "objective/train/original_loss": 3.0965638160705566,
      "objective/train/theoretical_loss": 3.960987126459872,
      "objective/train/tokens_used": 469381600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24366524815559387,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505927801132202,
      "objective/train/weighted_lm_loss": 3.253385543823242,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 3.960987126459872,
      "tokens_seen": 448921600
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004363490471414243,
      "loss": 3.0579,
      "theoretical_loss": 3.960987126459872,
      "tokens_seen": 448921600
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043633901705115346,
      "loss": 2.9835,
      "theoretical_loss": 3.960923689282876,
      "tokens_seen": 448987136
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043632898696088264,
      "loss": 3.0552,
      "theoretical_loss": 3.9608602639569965,
      "tokens_seen": 449052672
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043631895687061187,
      "loss": 3.0525,
      "theoretical_loss": 3.9607968504782907,
      "tokens_seen": 449118208
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.000436308926780341,
      "loss": 3.1653,
      "theoretical_loss": 3.9607334488428174,
      "tokens_seen": 449183744
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043629889669007023,
      "loss": 3.2362,
      "theoretical_loss": 3.960670059046638,
      "tokens_seen": 449249280
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043628886659979936,
      "loss": 3.2345,
      "theoretical_loss": 3.9606066810858143,
      "tokens_seen": 449314816
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004362788365095286,
      "loss": 3.3655,
      "theoretical_loss": 3.960543314956411,
      "tokens_seen": 449380352
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004362688064192578,
      "loss": 3.2351,
      "theoretical_loss": 3.960479960654494,
      "tokens_seen": 449445888
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043625877632898696,
      "loss": 2.9957,
      "theoretical_loss": 3.960416618176132,
      "tokens_seen": 449511424
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043624874623871614,
      "loss": 3.0255,
      "theoretical_loss": 3.9603532875173957,
      "tokens_seen": 449576960
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004362387161484453,
      "loss": 3.1484,
      "theoretical_loss": 3.9602899686743553,
      "tokens_seen": 449642496
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004362286860581745,
      "loss": 3.2035,
      "theoretical_loss": 3.9602266616430866,
      "tokens_seen": 449708032
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043621865596790374,
      "loss": 3.1261,
      "theoretical_loss": 3.9601633664196645,
      "tokens_seen": 449773568
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043620862587763286,
      "loss": 3.1984,
      "theoretical_loss": 3.9601000830001665,
      "tokens_seen": 449839104
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004361985957873621,
      "loss": 2.9476,
      "theoretical_loss": 3.9600368113806717,
      "tokens_seen": 449904640
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004361885656970913,
      "loss": 3.3689,
      "theoretical_loss": 3.9599735515572627,
      "tokens_seen": 449970176
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043617853560682046,
      "loss": 2.9735,
      "theoretical_loss": 3.959910303526021,
      "tokens_seen": 450035712
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004361685055165497,
      "loss": 2.877,
      "theoretical_loss": 3.9598470672830333,
      "tokens_seen": 450101248
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004361584754262788,
      "loss": 3.0444,
      "theoretical_loss": 3.9597838428243857,
      "tokens_seen": 450166784
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043614844533600806,
      "loss": 3.1377,
      "theoretical_loss": 3.959720630146167,
      "tokens_seen": 450232320
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043613841524573724,
      "loss": 3.2859,
      "theoretical_loss": 3.9596574292444684,
      "tokens_seen": 450297856
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004361283851554664,
      "loss": 2.9595,
      "theoretical_loss": 3.9595942401153823,
      "tokens_seen": 450363392
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004361183550651956,
      "loss": 3.0903,
      "theoretical_loss": 3.959531062755003,
      "tokens_seen": 450428928
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004361083249749248,
      "loss": 2.9615,
      "theoretical_loss": 3.959467897159427,
      "tokens_seen": 450494464
    },
    {
      "epoch": 0.14,
      "objective/train/advantage_avg": 0.4943898916244507,
      "objective/train/docs_used": 266764,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.564117670059204,
      "objective/train/original_loss": 3.564117670059204,
      "objective/train/theoretical_loss": 3.959404743324752,
      "objective/train/tokens_used": 471020000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2445097118616104,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506819486618042,
      "objective/train/weighted_lm_loss": 3.744762659072876,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0338950157165527,
      "theoretical_loss": 3.959404743324752,
      "tokens_seen": 450560000
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043609829488465397,
      "loss": 3.1961,
      "theoretical_loss": 3.959404743324752,
      "tokens_seen": 450560000
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004360882647943832,
      "loss": 3.1231,
      "theoretical_loss": 3.9593416012470786,
      "tokens_seen": 450625536
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043607823470411233,
      "loss": 2.9049,
      "theoretical_loss": 3.9592784709225084,
      "tokens_seen": 450691072
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043606820461384156,
      "loss": 2.9401,
      "theoretical_loss": 3.959215352347145,
      "tokens_seen": 450756608
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004360581745235707,
      "loss": 3.0095,
      "theoretical_loss": 3.959152245517094,
      "tokens_seen": 450822144
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004360481444332999,
      "loss": 2.9733,
      "theoretical_loss": 3.9590891504284635,
      "tokens_seen": 450887680
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004360381143430291,
      "loss": 3.2228,
      "theoretical_loss": 3.959026067077362,
      "tokens_seen": 450953216
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004360280842527583,
      "loss": 2.9563,
      "theoretical_loss": 3.958962995459901,
      "tokens_seen": 451018752
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043601805416248747,
      "loss": 3.029,
      "theoretical_loss": 3.9588999355721937,
      "tokens_seen": 451084288
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004360080240722167,
      "loss": 2.9303,
      "theoretical_loss": 3.958836887410354,
      "tokens_seen": 451149824
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043599799398194583,
      "loss": 3.0471,
      "theoretical_loss": 3.9587738509705,
      "tokens_seen": 451215360
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043598796389167507,
      "loss": 3.0177,
      "theoretical_loss": 3.95871082624875,
      "tokens_seen": 451280896
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004359779338014042,
      "loss": 3.0932,
      "theoretical_loss": 3.9586478132412233,
      "tokens_seen": 451346432
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043596790371113343,
      "loss": 3.0643,
      "theoretical_loss": 3.9585848119440437,
      "tokens_seen": 451411968
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004359578736208626,
      "loss": 2.8667,
      "theoretical_loss": 3.958521822353334,
      "tokens_seen": 451477504
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004359478435305918,
      "loss": 3.3013,
      "theoretical_loss": 3.9584588444652207,
      "tokens_seen": 451543040
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043593781344032097,
      "loss": 3.2367,
      "theoretical_loss": 3.9583958782758315,
      "tokens_seen": 451608576
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043592778335005015,
      "loss": 3.1987,
      "theoretical_loss": 3.9583329237812968,
      "tokens_seen": 451674112
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043591775325977933,
      "loss": 2.8295,
      "theoretical_loss": 3.9582699809777466,
      "tokens_seen": 451739648
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043590772316950857,
      "loss": 3.3859,
      "theoretical_loss": 3.9582070498613158,
      "tokens_seen": 451805184
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004358976930792377,
      "loss": 2.8396,
      "theoretical_loss": 3.958144130428139,
      "tokens_seen": 451870720
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043588766298896693,
      "loss": 3.2143,
      "theoretical_loss": 3.9580812226743523,
      "tokens_seen": 451936256
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004358776328986961,
      "loss": 3.0271,
      "theoretical_loss": 3.9580183265960955,
      "tokens_seen": 452001792
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004358676028084253,
      "loss": 3.1532,
      "theoretical_loss": 3.9579554421895096,
      "tokens_seen": 452067328
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004358575727181545,
      "loss": 2.9781,
      "theoretical_loss": 3.9578925694507356,
      "tokens_seen": 452132864
    },
    {
      "epoch": 0.14,
      "objective/train/advantage_avg": 0.4937143921852112,
      "objective/train/docs_used": 268192,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.759460687637329,
      "objective/train/original_loss": 2.75946044921875,
      "objective/train/theoretical_loss": 3.9578297083759195,
      "objective/train/tokens_used": 472658400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24379627406597137,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506106615066528,
      "objective/train/weighted_lm_loss": 2.8990302085876465,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0413066148757935,
      "theoretical_loss": 3.9578297083759195,
      "tokens_seen": 452198400
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043584754262788366,
      "loss": 2.9481,
      "theoretical_loss": 3.9578297083759195,
      "tokens_seen": 452198400
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043583751253761284,
      "loss": 2.9895,
      "theoretical_loss": 3.957766858961207,
      "tokens_seen": 452263936
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043582748244734207,
      "loss": 3.1853,
      "theoretical_loss": 3.957704021202746,
      "tokens_seen": 452329472
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004358174523570712,
      "loss": 3.0303,
      "theoretical_loss": 3.957641195096685,
      "tokens_seen": 452395008
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043580742226680043,
      "loss": 3.106,
      "theoretical_loss": 3.9575783806391787,
      "tokens_seen": 452460544
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043579739217652956,
      "loss": 3.1358,
      "theoretical_loss": 3.9575155778263778,
      "tokens_seen": 452526080
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004357873620862588,
      "loss": 3.2593,
      "theoretical_loss": 3.9574527866544384,
      "tokens_seen": 452591616
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.000435777331995988,
      "loss": 3.3,
      "theoretical_loss": 3.9573900071195185,
      "tokens_seen": 452657152
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043576730190571716,
      "loss": 3.2577,
      "theoretical_loss": 3.9573272392177765,
      "tokens_seen": 452722688
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043575727181544634,
      "loss": 3.1172,
      "theoretical_loss": 3.957264482945373,
      "tokens_seen": 452788224
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004357472417251755,
      "loss": 3.0682,
      "theoretical_loss": 3.9572017382984708,
      "tokens_seen": 452853760
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004357372116349047,
      "loss": 2.996,
      "theoretical_loss": 3.9571390052732345,
      "tokens_seen": 452919296
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043572718154463394,
      "loss": 2.953,
      "theoretical_loss": 3.95707628386583,
      "tokens_seen": 452984832
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043571715145436306,
      "loss": 3.1943,
      "theoretical_loss": 3.957013574072426,
      "tokens_seen": 453050368
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004357071213640923,
      "loss": 2.9849,
      "theoretical_loss": 3.9569508758891914,
      "tokens_seen": 453115904
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004356970912738215,
      "loss": 3.0902,
      "theoretical_loss": 3.9568881893122994,
      "tokens_seen": 453181440
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043568706118355066,
      "loss": 2.9782,
      "theoretical_loss": 3.956825514337922,
      "tokens_seen": 453246976
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043567703109327984,
      "loss": 2.9724,
      "theoretical_loss": 3.9567628509622352,
      "tokens_seen": 453312512
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.000435667001003009,
      "loss": 3.0363,
      "theoretical_loss": 3.956700199181416,
      "tokens_seen": 453378048
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004356569709127382,
      "loss": 3.2951,
      "theoretical_loss": 3.9566375589916443,
      "tokens_seen": 453443584
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043564694082246744,
      "loss": 3.276,
      "theoretical_loss": 3.9565749303891,
      "tokens_seen": 453509120
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043563691073219657,
      "loss": 3.2606,
      "theoretical_loss": 3.956512313369965,
      "tokens_seen": 453574656
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004356268806419258,
      "loss": 3.0334,
      "theoretical_loss": 3.956449707930425,
      "tokens_seen": 453640192
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043561685055165493,
      "loss": 2.9894,
      "theoretical_loss": 3.9563871140666658,
      "tokens_seen": 453705728
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043560682046138417,
      "loss": 2.9865,
      "theoretical_loss": 3.9563245317748756,
      "tokens_seen": 453771264
    },
    {
      "epoch": 0.14,
      "objective/train/advantage_avg": 0.494680792093277,
      "objective/train/docs_used": 268717,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.8005006313323975,
      "objective/train/original_loss": 2.8005006313323975,
      "objective/train/theoretical_loss": 3.9562619610512435,
      "objective/train/tokens_used": 474296800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24475814402103424,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.05071222782135,
      "objective/train/weighted_lm_loss": 2.942582845687866,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0396549701690674,
      "theoretical_loss": 3.9562619610512435,
      "tokens_seen": 453836800
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043559679037111335,
      "loss": 2.9357,
      "theoretical_loss": 3.9562619610512435,
      "tokens_seen": 453836800
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043558676028084253,
      "loss": 3.1268,
      "theoretical_loss": 3.956199401891962,
      "tokens_seen": 453902336
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004355767301905717,
      "loss": 2.9422,
      "theoretical_loss": 3.9561368542932245,
      "tokens_seen": 453967872
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004355667001003009,
      "loss": 3.2965,
      "theoretical_loss": 3.9560743182512255,
      "tokens_seen": 454033408
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043555667001003007,
      "loss": 3.0888,
      "theoretical_loss": 3.956011793762162,
      "tokens_seen": 454098944
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004355466399197593,
      "loss": 3.137,
      "theoretical_loss": 3.9559492808222334,
      "tokens_seen": 454164480
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043553660982948843,
      "loss": 3.0819,
      "theoretical_loss": 3.9558867794276407,
      "tokens_seen": 454230016
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043552657973921767,
      "loss": 2.9804,
      "theoretical_loss": 3.955824289574586,
      "tokens_seen": 454295552
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043551654964894685,
      "loss": 3.0108,
      "theoretical_loss": 3.955761811259273,
      "tokens_seen": 454361088
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043550651955867603,
      "loss": 3.1386,
      "theoretical_loss": 3.9556993444779076,
      "tokens_seen": 454426624
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004354964894684052,
      "loss": 3.2371,
      "theoretical_loss": 3.9556368892266986,
      "tokens_seen": 454492160
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004354864593781344,
      "loss": 3.1761,
      "theoretical_loss": 3.9555744455018553,
      "tokens_seen": 454557696
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004354764292878636,
      "loss": 3.1208,
      "theoretical_loss": 3.9555120132995887,
      "tokens_seen": 454623232
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004354663991975928,
      "loss": 2.9398,
      "theoretical_loss": 3.955449592616112,
      "tokens_seen": 454688768
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043545636910732194,
      "loss": 2.9141,
      "theoretical_loss": 3.9553871834476406,
      "tokens_seen": 454754304
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043544633901705117,
      "loss": 3.0327,
      "theoretical_loss": 3.955324785790391,
      "tokens_seen": 454819840
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004354363089267803,
      "loss": 3.1207,
      "theoretical_loss": 3.9552623996405822,
      "tokens_seen": 454885376
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043542627883650953,
      "loss": 2.9203,
      "theoretical_loss": 3.9552000249944337,
      "tokens_seen": 454950912
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043541624874623877,
      "loss": 2.9208,
      "theoretical_loss": 3.9551376618481684,
      "tokens_seen": 455016448
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004354062186559679,
      "loss": 2.8504,
      "theoretical_loss": 3.9550753101980103,
      "tokens_seen": 455081984
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043539618856569713,
      "loss": 3.1481,
      "theoretical_loss": 3.955012970040184,
      "tokens_seen": 455147520
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004353861584754263,
      "loss": 3.1065,
      "theoretical_loss": 3.9549506413709183,
      "tokens_seen": 455213056
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004353761283851555,
      "loss": 3.3557,
      "theoretical_loss": 3.954888324186442,
      "tokens_seen": 455278592
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004353660982948847,
      "loss": 2.9473,
      "theoretical_loss": 3.954826018482986,
      "tokens_seen": 455344128
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043535606820461386,
      "loss": 3.0107,
      "theoretical_loss": 3.9547637242567832,
      "tokens_seen": 455409664
    },
    {
      "epoch": 0.14,
      "objective/train/advantage_avg": 0.49434995651245117,
      "objective/train/docs_used": 269967,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.8045313358306885,
      "objective/train/original_loss": 2.8045310974121094,
      "objective/train/theoretical_loss": 3.954701441504068,
      "objective/train/tokens_used": 475935200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2444310039281845,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506775379180908,
      "objective/train/weighted_lm_loss": 2.9466047286987305,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0396549701690674,
      "theoretical_loss": 3.954701441504068,
      "tokens_seen": 455475200
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043534603811434304,
      "loss": 3.0093,
      "theoretical_loss": 3.954701441504068,
      "tokens_seen": 455475200
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043533600802407227,
      "loss": 2.8971,
      "theoretical_loss": 3.9546391702210766,
      "tokens_seen": 455540736
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004353259779338014,
      "loss": 3.0053,
      "theoretical_loss": 3.9545769104040485,
      "tokens_seen": 455606272
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043531594784353063,
      "loss": 2.9347,
      "theoretical_loss": 3.954514662049222,
      "tokens_seen": 455671808
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043530591775325976,
      "loss": 3.05,
      "theoretical_loss": 3.95445242515284,
      "tokens_seen": 455737344
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.000435295887662989,
      "loss": 2.9472,
      "theoretical_loss": 3.9543901997111446,
      "tokens_seen": 455802880
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004352858575727182,
      "loss": 3.1818,
      "theoretical_loss": 3.954327985720383,
      "tokens_seen": 455868416
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043527582748244736,
      "loss": 2.9881,
      "theoretical_loss": 3.9542657831768,
      "tokens_seen": 455933952
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043526579739217654,
      "loss": 3.2061,
      "theoretical_loss": 3.9542035920766465,
      "tokens_seen": 455999488
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004352557673019057,
      "loss": 3.1286,
      "theoretical_loss": 3.9541414124161713,
      "tokens_seen": 456065024
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004352457372116349,
      "loss": 3.0515,
      "theoretical_loss": 3.954079244191628,
      "tokens_seen": 456130560
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043523570712136414,
      "loss": 2.9976,
      "theoretical_loss": 3.9540170873992704,
      "tokens_seen": 456196096
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043522567703109326,
      "loss": 2.9756,
      "theoretical_loss": 3.9539549420353532,
      "tokens_seen": 456261632
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004352156469408225,
      "loss": 2.9509,
      "theoretical_loss": 3.9538928080961355,
      "tokens_seen": 456327168
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004352056168505517,
      "loss": 3.0328,
      "theoretical_loss": 3.9538306855778766,
      "tokens_seen": 456392704
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043519558676028086,
      "loss": 3.0657,
      "theoretical_loss": 3.953768574476837,
      "tokens_seen": 456458240
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043518555667001004,
      "loss": 3.124,
      "theoretical_loss": 3.9537064747892794,
      "tokens_seen": 456523776
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004351755265797392,
      "loss": 2.814,
      "theoretical_loss": 3.953644386511469,
      "tokens_seen": 456589312
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004351654964894684,
      "loss": 3.054,
      "theoretical_loss": 3.9535823096396725,
      "tokens_seen": 456654848
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043515546639919764,
      "loss": 3.1187,
      "theoretical_loss": 3.953520244170157,
      "tokens_seen": 456720384
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043514543630892677,
      "loss": 2.9979,
      "theoretical_loss": 3.9534581900991936,
      "tokens_seen": 456785920
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.000435135406218656,
      "loss": 3.1273,
      "theoretical_loss": 3.9533961474230535,
      "tokens_seen": 456851456
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043512537612838513,
      "loss": 3.1408,
      "theoretical_loss": 3.9533341161380102,
      "tokens_seen": 456916992
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043511534603811437,
      "loss": 3.0821,
      "theoretical_loss": 3.9532720962403385,
      "tokens_seen": 456982528
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043510531594784355,
      "loss": 3.1813,
      "theoretical_loss": 3.953210087726316,
      "tokens_seen": 457048064
    },
    {
      "epoch": 0.14,
      "objective/train/advantage_avg": 0.49401727318763733,
      "objective/train/docs_used": 270622,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.475588083267212,
      "objective/train/original_loss": 3.475588321685791,
      "objective/train/theoretical_loss": 3.9531480905922205,
      "objective/train/tokens_used": 477573600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24411353468894958,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506426095962524,
      "objective/train/weighted_lm_loss": 3.651499032974243,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0432155132293701,
      "theoretical_loss": 3.9531480905922205,
      "tokens_seen": 457113600
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043509528585757273,
      "loss": 3.1909,
      "theoretical_loss": 3.9531480905922205,
      "tokens_seen": 457113600
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004350852557673019,
      "loss": 2.7788,
      "theoretical_loss": 3.953086104834334,
      "tokens_seen": 457179136
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004350752256770311,
      "loss": 3.0028,
      "theoretical_loss": 3.953024130448937,
      "tokens_seen": 457244672
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043506519558676027,
      "loss": 2.8479,
      "theoretical_loss": 3.952962167432314,
      "tokens_seen": 457310208
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004350551654964895,
      "loss": 3.1185,
      "theoretical_loss": 3.9529002157807516,
      "tokens_seen": 457375744
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043504513540621863,
      "loss": 2.9092,
      "theoretical_loss": 3.952838275490536,
      "tokens_seen": 457441280
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043503510531594787,
      "loss": 3.053,
      "theoretical_loss": 3.9527763465579575,
      "tokens_seen": 457506816
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043502507522567705,
      "loss": 3.1446,
      "theoretical_loss": 3.9527144289793057,
      "tokens_seen": 457572352
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043501504513540623,
      "loss": 3.1608,
      "theoretical_loss": 3.952652522750874,
      "tokens_seen": 457637888
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004350050150451354,
      "loss": 3.0346,
      "theoretical_loss": 3.9525906278689575,
      "tokens_seen": 457703424
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004349949849548646,
      "loss": 3.0461,
      "theoretical_loss": 3.952528744329851,
      "tokens_seen": 457768960
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004349849548645938,
      "loss": 3.2551,
      "theoretical_loss": 3.952466872129854,
      "tokens_seen": 457834496
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.000434974924774323,
      "loss": 2.9728,
      "theoretical_loss": 3.952405011265264,
      "tokens_seen": 457900032
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043496489468405214,
      "loss": 3.2013,
      "theoretical_loss": 3.9523431617323848,
      "tokens_seen": 457965568
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043495486459378137,
      "loss": 3.1017,
      "theoretical_loss": 3.952281323527517,
      "tokens_seen": 458031104
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004349448345035105,
      "loss": 2.9775,
      "theoretical_loss": 3.9522194966469675,
      "tokens_seen": 458096640
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043493480441323973,
      "loss": 3.172,
      "theoretical_loss": 3.9521576810870416,
      "tokens_seen": 458162176
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004349247743229689,
      "loss": 3.0816,
      "theoretical_loss": 3.9520958768440484,
      "tokens_seen": 458227712
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004349147442326981,
      "loss": 2.979,
      "theoretical_loss": 3.952034083914297,
      "tokens_seen": 458293248
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004349047141424273,
      "loss": 2.9722,
      "theoretical_loss": 3.9519723022941005,
      "tokens_seen": 458358784
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004348946840521565,
      "loss": 3.1538,
      "theoretical_loss": 3.951910531979771,
      "tokens_seen": 458424320
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043488465396188564,
      "loss": 2.9482,
      "theoretical_loss": 3.951848772967625,
      "tokens_seen": 458489856
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004348746238716149,
      "loss": 3.1046,
      "theoretical_loss": 3.9517870252539784,
      "tokens_seen": 458555392
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.000434864593781344,
      "loss": 3.2591,
      "theoretical_loss": 3.95172528883515,
      "tokens_seen": 458620928
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043485456369107324,
      "loss": 3.277,
      "theoretical_loss": 3.9516635637074615,
      "tokens_seen": 458686464
    },
    {
      "epoch": 0.14,
      "objective/train/advantage_avg": 0.4938891530036926,
      "objective/train/docs_used": 271952,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.735518217086792,
      "objective/train/original_loss": 2.735517978668213,
      "objective/train/theoretical_loss": 3.951601849867233,
      "objective/train/tokens_used": 479212000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24400101602077484,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506292581558228,
      "objective/train/weighted_lm_loss": 2.8739986419677734,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 3.951601849867233,
      "tokens_seen": 458752000
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004348445336008024,
      "loss": 2.988,
      "theoretical_loss": 3.951601849867233,
      "tokens_seen": 458752000
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004348345035105316,
      "loss": 3.0813,
      "theoretical_loss": 3.95154014731079,
      "tokens_seen": 458817536
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004348244734202608,
      "loss": 3.0137,
      "theoretical_loss": 3.951478456034457,
      "tokens_seen": 458883072
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043481444332998996,
      "loss": 3.0986,
      "theoretical_loss": 3.951416776034562,
      "tokens_seen": 458948608
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043480441323971914,
      "loss": 3.2382,
      "theoretical_loss": 3.951355107307434,
      "tokens_seen": 459014144
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004347943831494484,
      "loss": 3.1498,
      "theoretical_loss": 3.951293449849403,
      "tokens_seen": 459079680
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004347843530591775,
      "loss": 3.327,
      "theoretical_loss": 3.9512318036568024,
      "tokens_seen": 459145216
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043477432296890674,
      "loss": 2.8525,
      "theoretical_loss": 3.9511701687259655,
      "tokens_seen": 459210752
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043476429287863587,
      "loss": 3.1166,
      "theoretical_loss": 3.951108545053229,
      "tokens_seen": 459276288
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004347542627883651,
      "loss": 2.8654,
      "theoretical_loss": 3.9510469326349296,
      "tokens_seen": 459341824
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004347442326980943,
      "loss": 3.0833,
      "theoretical_loss": 3.950985331467407,
      "tokens_seen": 459407360
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043473420260782346,
      "loss": 3.0487,
      "theoretical_loss": 3.9509237415470024,
      "tokens_seen": 459472896
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043472417251755265,
      "loss": 3.0851,
      "theoretical_loss": 3.9508621628700586,
      "tokens_seen": 459538432
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004347141424272819,
      "loss": 3.2537,
      "theoretical_loss": 3.95080059543292,
      "tokens_seen": 459603968
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.000434704112337011,
      "loss": 2.9241,
      "theoretical_loss": 3.950739039231933,
      "tokens_seen": 459669504
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043469408224674024,
      "loss": 2.9852,
      "theoretical_loss": 3.950677494263444,
      "tokens_seen": 459735040
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004346840521564694,
      "loss": 3.1686,
      "theoretical_loss": 3.9506159605238045,
      "tokens_seen": 459800576
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004346740220661986,
      "loss": 3.1388,
      "theoretical_loss": 3.9505544380093642,
      "tokens_seen": 459866112
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043466399197592784,
      "loss": 3.1912,
      "theoretical_loss": 3.9504929267164774,
      "tokens_seen": 459931648
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043465396188565697,
      "loss": 2.9293,
      "theoretical_loss": 3.950431426641498,
      "tokens_seen": 459997184
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004346439317953862,
      "loss": 3.1988,
      "theoretical_loss": 3.950369937780783,
      "tokens_seen": 460062720
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043463390170511533,
      "loss": 2.9973,
      "theoretical_loss": 3.9503084601306897,
      "tokens_seen": 460128256
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043462387161484457,
      "loss": 3.2142,
      "theoretical_loss": 3.9502469936875784,
      "tokens_seen": 460193792
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043461384152457375,
      "loss": 3.0882,
      "theoretical_loss": 3.950185538447811,
      "tokens_seen": 460259328
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043460381143430293,
      "loss": 2.9116,
      "theoretical_loss": 3.9501240944077494,
      "tokens_seen": 460324864
    },
    {
      "epoch": 0.14,
      "objective/train/advantage_avg": 0.49573150277137756,
      "objective/train/docs_used": 272724,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.4928669929504395,
      "objective/train/original_loss": 3.492866277694702,
      "objective/train/theoretical_loss": 3.9500626615637597,
      "objective/train/tokens_used": 480850400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24579942226409912,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0508226156234741,
      "objective/train/weighted_lm_loss": 3.670428514480591,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0425469875335693,
      "theoretical_loss": 3.9500626615637597,
      "tokens_seen": 460390400
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004345937813440321,
      "loss": 3.2421,
      "theoretical_loss": 3.9500626615637597,
      "tokens_seen": 460390400
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004345837512537613,
      "loss": 3.1251,
      "theoretical_loss": 3.9500012399122078,
      "tokens_seen": 460455936
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043457372116349047,
      "loss": 3.0798,
      "theoretical_loss": 3.9499398294494625,
      "tokens_seen": 460521472
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004345636910732197,
      "loss": 2.9754,
      "theoretical_loss": 3.949878430171893,
      "tokens_seen": 460587008
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043455366098294883,
      "loss": 3.1097,
      "theoretical_loss": 3.949817042075872,
      "tokens_seen": 460652544
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043454363089267807,
      "loss": 3.2438,
      "theoretical_loss": 3.9497556651577725,
      "tokens_seen": 460718080
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043453360080240725,
      "loss": 3.2056,
      "theoretical_loss": 3.949694299413969,
      "tokens_seen": 460783616
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043452357071213643,
      "loss": 2.9083,
      "theoretical_loss": 3.9496329448408387,
      "tokens_seen": 460849152
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004345135406218656,
      "loss": 3.0567,
      "theoretical_loss": 3.9495716014347604,
      "tokens_seen": 460914688
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004345035105315948,
      "loss": 3.0183,
      "theoretical_loss": 3.9495102691921136,
      "tokens_seen": 460980224
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.000434493480441324,
      "loss": 3.0621,
      "theoretical_loss": 3.9494489481092803,
      "tokens_seen": 461045760
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004344834503510532,
      "loss": 2.9536,
      "theoretical_loss": 3.9493876381826443,
      "tokens_seen": 461111296
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043447342026078234,
      "loss": 3.2228,
      "theoretical_loss": 3.9493263394085902,
      "tokens_seen": 461176832
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043446339017051157,
      "loss": 3.181,
      "theoretical_loss": 3.9492650517835055,
      "tokens_seen": 461242368
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004344533600802407,
      "loss": 2.8296,
      "theoretical_loss": 3.9492037753037783,
      "tokens_seen": 461307904
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043444332998996993,
      "loss": 3.0552,
      "theoretical_loss": 3.949142509965799,
      "tokens_seen": 461373440
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004344332998996991,
      "loss": 3.0496,
      "theoretical_loss": 3.9490812557659596,
      "tokens_seen": 461438976
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004344232698094283,
      "loss": 3.1483,
      "theoretical_loss": 3.949020012700654,
      "tokens_seen": 461504512
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004344132397191575,
      "loss": 3.1332,
      "theoretical_loss": 3.9489587807662763,
      "tokens_seen": 461570048
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004344032096288867,
      "loss": 3.0024,
      "theoretical_loss": 3.9488975599592244,
      "tokens_seen": 461635584
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043439317953861584,
      "loss": 2.8445,
      "theoretical_loss": 3.948836350275897,
      "tokens_seen": 461701120
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004343831494483451,
      "loss": 3.2929,
      "theoretical_loss": 3.9487751517126948,
      "tokens_seen": 461766656
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004343731193580742,
      "loss": 3.3667,
      "theoretical_loss": 3.9487139642660187,
      "tokens_seen": 461832192
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043436308926780344,
      "loss": 3.0887,
      "theoretical_loss": 3.9486527879322733,
      "tokens_seen": 461897728
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004343530591775326,
      "loss": 2.9982,
      "theoretical_loss": 3.9485916227078626,
      "tokens_seen": 461963264
    },
    {
      "epoch": 0.14,
      "objective/train/advantage_avg": 0.4947793781757355,
      "objective/train/docs_used": 273842,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.1547701358795166,
      "objective/train/original_loss": 3.1547698974609375,
      "objective/train/theoretical_loss": 3.948530468589195,
      "objective/train/tokens_used": 482488800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2448568195104599,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507227182388306,
      "objective/train/weighted_lm_loss": 3.3145408630371094,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0460278987884521,
      "theoretical_loss": 3.948530468589195,
      "tokens_seen": 462028800
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004343430290872618,
      "loss": 3.1567,
      "theoretical_loss": 3.948530468589195,
      "tokens_seen": 462028800
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.000434332998996991,
      "loss": 3.0741,
      "theoretical_loss": 3.9484693255726793,
      "tokens_seen": 462094336
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043432296890672016,
      "loss": 3.0354,
      "theoretical_loss": 3.9484081936547244,
      "tokens_seen": 462159872
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043431293881644934,
      "loss": 3.0604,
      "theoretical_loss": 3.948347072831744,
      "tokens_seen": 462225408
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004343029087261786,
      "loss": 3.1645,
      "theoretical_loss": 3.94828596310015,
      "tokens_seen": 462290944
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004342928786359077,
      "loss": 3.2171,
      "theoretical_loss": 3.9482248644563596,
      "tokens_seen": 462356480
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043428284854563694,
      "loss": 2.7867,
      "theoretical_loss": 3.9481637768967883,
      "tokens_seen": 462422016
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043427281845536607,
      "loss": 2.8758,
      "theoretical_loss": 3.9481027004178557,
      "tokens_seen": 462487552
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004342627883650953,
      "loss": 3.0022,
      "theoretical_loss": 3.948041635015982,
      "tokens_seen": 462553088
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004342527582748245,
      "loss": 2.7936,
      "theoretical_loss": 3.947980580687589,
      "tokens_seen": 462618624
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043424272818455367,
      "loss": 3.0998,
      "theoretical_loss": 3.947919537429101,
      "tokens_seen": 462684160
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043423269809428285,
      "loss": 3.1175,
      "theoretical_loss": 3.947858505236942,
      "tokens_seen": 462749696
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004342226680040121,
      "loss": 3.0804,
      "theoretical_loss": 3.9477974841075403,
      "tokens_seen": 462815232
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004342126379137412,
      "loss": 3.0844,
      "theoretical_loss": 3.947736474037324,
      "tokens_seen": 462880768
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043420260782347044,
      "loss": 2.9364,
      "theoretical_loss": 3.9476754750227236,
      "tokens_seen": 462946304
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043419257773319957,
      "loss": 2.898,
      "theoretical_loss": 3.9476144870601706,
      "tokens_seen": 463011840
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004341825476429288,
      "loss": 2.8917,
      "theoretical_loss": 3.9475535101460997,
      "tokens_seen": 463077376
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.000434172517552658,
      "loss": 2.9134,
      "theoretical_loss": 3.947492544276945,
      "tokens_seen": 463142912
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043416248746238717,
      "loss": 3.0735,
      "theoretical_loss": 3.947431589449144,
      "tokens_seen": 463208448
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043415245737211635,
      "loss": 3.0,
      "theoretical_loss": 3.9473706456591353,
      "tokens_seen": 463273984
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043414242728184553,
      "loss": 3.08,
      "theoretical_loss": 3.947309712903359,
      "tokens_seen": 463339520
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004341323971915747,
      "loss": 2.9986,
      "theoretical_loss": 3.947248791178257,
      "tokens_seen": 463405056
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043412236710130395,
      "loss": 3.3409,
      "theoretical_loss": 3.9471878804802736,
      "tokens_seen": 463470592
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004341123370110331,
      "loss": 3.2054,
      "theoretical_loss": 3.9471269808058524,
      "tokens_seen": 463536128
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004341023069207623,
      "loss": 2.8887,
      "theoretical_loss": 3.9470660921514424,
      "tokens_seen": 463601664
    },
    {
      "epoch": 0.14,
      "objective/train/advantage_avg": 0.4931437075138092,
      "objective/train/docs_used": 274430,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.6607754230499268,
      "objective/train/original_loss": 2.6607751846313477,
      "objective/train/theoretical_loss": 3.94700521451349,
      "objective/train/tokens_used": 484127200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.243277907371521,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050550937652588,
      "objective/train/weighted_lm_loss": 2.7956254482269287,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0168063640594482,
      "theoretical_loss": 3.94700521451349,
      "tokens_seen": 463667200
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043409227683049144,
      "loss": 3.0177,
      "theoretical_loss": 3.94700521451349,
      "tokens_seen": 463667200
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043408224674022067,
      "loss": 2.92,
      "theoretical_loss": 3.9469443478884463,
      "tokens_seen": 463732736
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043407221664994985,
      "loss": 3.0363,
      "theoretical_loss": 3.9468834922727636,
      "tokens_seen": 463798272
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043406218655967903,
      "loss": 3.0359,
      "theoretical_loss": 3.946822647662894,
      "tokens_seen": 463863808
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004340521564694082,
      "loss": 2.7813,
      "theoretical_loss": 3.946761814055294,
      "tokens_seen": 463929344
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043404212637913745,
      "loss": 2.9632,
      "theoretical_loss": 3.9467009914464186,
      "tokens_seen": 463994880
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004340320962888666,
      "loss": 3.0945,
      "theoretical_loss": 3.946640179832728,
      "tokens_seen": 464060416
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004340220661985958,
      "loss": 3.0246,
      "theoretical_loss": 3.9465793792106814,
      "tokens_seen": 464125952
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043401203610832494,
      "loss": 3.0309,
      "theoretical_loss": 3.94651858957674,
      "tokens_seen": 464191488
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004340020060180542,
      "loss": 3.2073,
      "theoretical_loss": 3.9464578109273676,
      "tokens_seen": 464257024
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043399197592778336,
      "loss": 2.9279,
      "theoretical_loss": 3.9463970432590294,
      "tokens_seen": 464322560
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043398194583751254,
      "loss": 2.9548,
      "theoretical_loss": 3.9463362865681915,
      "tokens_seen": 464388096
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004339719157472417,
      "loss": 3.2488,
      "theoretical_loss": 3.946275540851322,
      "tokens_seen": 464453632
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004339618856569709,
      "loss": 2.9199,
      "theoretical_loss": 3.9462148061048907,
      "tokens_seen": 464519168
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004339518555667001,
      "loss": 3.0572,
      "theoretical_loss": 3.9461540823253696,
      "tokens_seen": 464584704
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004339418254764293,
      "loss": 3.101,
      "theoretical_loss": 3.9460933695092306,
      "tokens_seen": 464650240
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004339317953861585,
      "loss": 2.9235,
      "theoretical_loss": 3.9460326676529496,
      "tokens_seen": 464715776
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004339217652958877,
      "loss": 3.1078,
      "theoretical_loss": 3.945971976753003,
      "tokens_seen": 464781312
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004339117352056169,
      "loss": 2.823,
      "theoretical_loss": 3.9459112968058685,
      "tokens_seen": 464846848
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043390170511534604,
      "loss": 3.0161,
      "theoretical_loss": 3.945850627808025,
      "tokens_seen": 464912384
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004338916750250753,
      "loss": 3.3028,
      "theoretical_loss": 3.9457899697559546,
      "tokens_seen": 464977920
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004338816449348044,
      "loss": 3.3318,
      "theoretical_loss": 3.9457293226461396,
      "tokens_seen": 465043456
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043387161484453364,
      "loss": 3.0493,
      "theoretical_loss": 3.9456686864750647,
      "tokens_seen": 465108992
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004338615847542628,
      "loss": 3.2574,
      "theoretical_loss": 3.9456080612392164,
      "tokens_seen": 465174528
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.000433851554663992,
      "loss": 3.0967,
      "theoretical_loss": 3.945547446935082,
      "tokens_seen": 465240064
    },
    {
      "epoch": 0.14,
      "objective/train/advantage_avg": 0.49371448159217834,
      "objective/train/docs_used": 275809,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.6850204467773438,
      "objective/train/original_loss": 2.6850204467773438,
      "objective/train/theoretical_loss": 3.9454868435591504,
      "objective/train/tokens_used": 485765600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24383126199245453,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506110191345215,
      "objective/train/weighted_lm_loss": 2.821045160293579,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0363596677780151,
      "theoretical_loss": 3.9454868435591504,
      "tokens_seen": 465305600
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004338415245737212,
      "loss": 2.7727,
      "theoretical_loss": 3.9454868435591504,
      "tokens_seen": 465305600
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043383149448345036,
      "loss": 2.9426,
      "theoretical_loss": 3.9454262511079135,
      "tokens_seen": 465371136
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043382146439317954,
      "loss": 2.9392,
      "theoretical_loss": 3.9453656695778636,
      "tokens_seen": 465436672
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004338114343029088,
      "loss": 3.0441,
      "theoretical_loss": 3.9453050989654948,
      "tokens_seen": 465502208
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004338014042126379,
      "loss": 3.0096,
      "theoretical_loss": 3.945244539267303,
      "tokens_seen": 465567744
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043379137412236714,
      "loss": 2.8172,
      "theoretical_loss": 3.9451839904797854,
      "tokens_seen": 465633280
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043378134403209627,
      "loss": 3.1426,
      "theoretical_loss": 3.945123452599441,
      "tokens_seen": 465698816
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004337713139418255,
      "loss": 2.8882,
      "theoretical_loss": 3.9450629256227714,
      "tokens_seen": 465764352
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004337612838515547,
      "loss": 2.9127,
      "theoretical_loss": 3.9450024095462783,
      "tokens_seen": 465829888
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043375125376128387,
      "loss": 3.1053,
      "theoretical_loss": 3.9449419043664653,
      "tokens_seen": 465895424
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043374122367101305,
      "loss": 3.0943,
      "theoretical_loss": 3.944881410079838,
      "tokens_seen": 465960960
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004337311935807423,
      "loss": 3.2249,
      "theoretical_loss": 3.9448209266829046,
      "tokens_seen": 466026496
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004337211634904714,
      "loss": 3.1695,
      "theoretical_loss": 3.9447604541721724,
      "tokens_seen": 466092032
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043371113340020064,
      "loss": 2.9042,
      "theoretical_loss": 3.9446999925441517,
      "tokens_seen": 466157568
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043370110330992977,
      "loss": 3.0056,
      "theoretical_loss": 3.944639541795356,
      "tokens_seen": 466223104
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.000433691073219659,
      "loss": 3.1787,
      "theoretical_loss": 3.944579101922298,
      "tokens_seen": 466288640
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004336810431293882,
      "loss": 3.0948,
      "theoretical_loss": 3.944518672921493,
      "tokens_seen": 466354176
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043367101303911737,
      "loss": 3.0083,
      "theoretical_loss": 3.944458254789458,
      "tokens_seen": 466419712
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043366098294884655,
      "loss": 3.0164,
      "theoretical_loss": 3.94439784752271,
      "tokens_seen": 466485248
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043365095285857573,
      "loss": 3.0131,
      "theoretical_loss": 3.9443374511177707,
      "tokens_seen": 466550784
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004336409227683049,
      "loss": 3.0095,
      "theoretical_loss": 3.9442770655711614,
      "tokens_seen": 466616320
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043363089267803415,
      "loss": 2.8375,
      "theoretical_loss": 3.9442166908794047,
      "tokens_seen": 466681856
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004336208625877633,
      "loss": 2.9032,
      "theoretical_loss": 3.9441563270390256,
      "tokens_seen": 466747392
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004336108324974925,
      "loss": 3.1442,
      "theoretical_loss": 3.944095974046551,
      "tokens_seen": 466812928
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043360080240722164,
      "loss": 2.9714,
      "theoretical_loss": 3.9440356318985086,
      "tokens_seen": 466878464
    },
    {
      "epoch": 0.14,
      "objective/train/advantage_avg": 0.4917564392089844,
      "objective/train/docs_used": 276320,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.787079334259033,
      "objective/train/original_loss": 2.787078619003296,
      "objective/train/theoretical_loss": 3.9439753005914273,
      "objective/train/tokens_used": 487404000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24192890524864197,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0504053831100464,
      "objective/train/weighted_lm_loss": 2.9274699687957764,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0253151655197144,
      "theoretical_loss": 3.9439753005914273,
      "tokens_seen": 466944000
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043359077231695087,
      "loss": 2.9286,
      "theoretical_loss": 3.9439753005914273,
      "tokens_seen": 466944000
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043358074222668005,
      "loss": 3.258,
      "theoretical_loss": 3.9439149801218396,
      "tokens_seen": 467009536
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043357071213640923,
      "loss": 3.1042,
      "theoretical_loss": 3.9438546704862776,
      "tokens_seen": 467075072
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004335606820461384,
      "loss": 3.0845,
      "theoretical_loss": 3.9437943716812756,
      "tokens_seen": 467140608
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043355065195586765,
      "loss": 3.0601,
      "theoretical_loss": 3.9437340837033696,
      "tokens_seen": 467206144
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004335406218655968,
      "loss": 2.9732,
      "theoretical_loss": 3.9436738065490973,
      "tokens_seen": 467271680
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.000433530591775326,
      "loss": 3.3091,
      "theoretical_loss": 3.943613540214998,
      "tokens_seen": 467337216
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043352056168505514,
      "loss": 2.8343,
      "theoretical_loss": 3.9435532846976127,
      "tokens_seen": 467402752
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004335105315947844,
      "loss": 2.9827,
      "theoretical_loss": 3.943493039993483,
      "tokens_seen": 467468288
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043350050150451356,
      "loss": 3.0643,
      "theoretical_loss": 3.9434328060991533,
      "tokens_seen": 467533824
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043349047141424274,
      "loss": 3.4212,
      "theoretical_loss": 3.943372583011169,
      "tokens_seen": 467599360
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004334804413239719,
      "loss": 3.298,
      "theoretical_loss": 3.9433123707260775,
      "tokens_seen": 467664896
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004334704112337011,
      "loss": 3.0691,
      "theoretical_loss": 3.9432521692404277,
      "tokens_seen": 467730432
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004334603811434303,
      "loss": 3.3353,
      "theoretical_loss": 3.943191978550769,
      "tokens_seen": 467795968
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004334503510531595,
      "loss": 3.0149,
      "theoretical_loss": 3.943131798653654,
      "tokens_seen": 467861504
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043344032096288864,
      "loss": 2.8057,
      "theoretical_loss": 3.9430716295456363,
      "tokens_seen": 467927040
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004334302908726179,
      "loss": 3.0658,
      "theoretical_loss": 3.9430114712232704,
      "tokens_seen": 467992576
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.000433420260782347,
      "loss": 3.1649,
      "theoretical_loss": 3.942951323683113,
      "tokens_seen": 468058112
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043341023069207624,
      "loss": 3.0275,
      "theoretical_loss": 3.942891186921723,
      "tokens_seen": 468123648
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004334002006018054,
      "loss": 3.0676,
      "theoretical_loss": 3.942831060935659,
      "tokens_seen": 468189184
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004333901705115346,
      "loss": 3.0766,
      "theoretical_loss": 3.9427709457214837,
      "tokens_seen": 468254720
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004333801404212638,
      "loss": 2.9113,
      "theoretical_loss": 3.942710841275759,
      "tokens_seen": 468320256
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.000433370110330993,
      "loss": 2.9997,
      "theoretical_loss": 3.9426507475950503,
      "tokens_seen": 468385792
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043336008024072215,
      "loss": 2.919,
      "theoretical_loss": 3.9425906646759232,
      "tokens_seen": 468451328
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004333500501504514,
      "loss": 2.926,
      "theoretical_loss": 3.942530592514945,
      "tokens_seen": 468516864
    },
    {
      "epoch": 0.14,
      "objective/train/advantage_avg": 0.4949471950531006,
      "objective/train/docs_used": 277290,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.738274335861206,
      "objective/train/original_loss": 2.738274335861206,
      "objective/train/theoretical_loss": 3.9424705311086856,
      "objective/train/tokens_used": 489042400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24502809345722198,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507402420043945,
      "objective/train/weighted_lm_loss": 2.8768930435180664,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0408108234405518,
      "theoretical_loss": 3.9424705311086856,
      "tokens_seen": 468582400
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004333400200601805,
      "loss": 3.0739,
      "theoretical_loss": 3.9424705311086856,
      "tokens_seen": 468582400
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043332998996990974,
      "loss": 2.9224,
      "theoretical_loss": 3.9424104804537157,
      "tokens_seen": 468647936
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004333199598796389,
      "loss": 3.2407,
      "theoretical_loss": 3.9423504405466074,
      "tokens_seen": 468713472
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004333099297893681,
      "loss": 3.0952,
      "theoretical_loss": 3.942290411383935,
      "tokens_seen": 468779008
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004332998996990973,
      "loss": 3.1089,
      "theoretical_loss": 3.942230392962274,
      "tokens_seen": 468844544
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043328986960882647,
      "loss": 3.1931,
      "theoretical_loss": 3.942170385278202,
      "tokens_seen": 468910080
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043327983951855565,
      "loss": 3.2244,
      "theoretical_loss": 3.942110388328296,
      "tokens_seen": 468975616
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004332698094282849,
      "loss": 2.9937,
      "theoretical_loss": 3.9420504021091385,
      "tokens_seen": 469041152
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.000433259779338014,
      "loss": 2.8985,
      "theoretical_loss": 3.94199042661731,
      "tokens_seen": 469106688
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043324974924774325,
      "loss": 3.3058,
      "theoretical_loss": 3.9419304618493936,
      "tokens_seen": 469172224
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043323971915747243,
      "loss": 2.6696,
      "theoretical_loss": 3.9418705078019753,
      "tokens_seen": 469237760
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004332296890672016,
      "loss": 2.9907,
      "theoretical_loss": 3.941810564471641,
      "tokens_seen": 469303296
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004332196589769308,
      "loss": 3.1166,
      "theoretical_loss": 3.9417506318549784,
      "tokens_seen": 469368832
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043320962888665997,
      "loss": 3.1635,
      "theoretical_loss": 3.9416907099485776,
      "tokens_seen": 469434368
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043319959879638915,
      "loss": 2.8686,
      "theoretical_loss": 3.9416307987490304,
      "tokens_seen": 469499904
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004331895687061184,
      "loss": 3.0321,
      "theoretical_loss": 3.9415708982529285,
      "tokens_seen": 469565440
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043317953861584757,
      "loss": 3.0985,
      "theoretical_loss": 3.9415110084568665,
      "tokens_seen": 469630976
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043316950852557675,
      "loss": 3.0814,
      "theoretical_loss": 3.941451129357441,
      "tokens_seen": 469696512
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043315947843530593,
      "loss": 2.9201,
      "theoretical_loss": 3.9413912609512485,
      "tokens_seen": 469762048
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004331494483450351,
      "loss": 3.0263,
      "theoretical_loss": 3.9413314032348885,
      "tokens_seen": 469827584
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043313941825476435,
      "loss": 3.1559,
      "theoretical_loss": 3.9412715562049616,
      "tokens_seen": 469893120
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004331293881644935,
      "loss": 3.0128,
      "theoretical_loss": 3.94121171985807,
      "tokens_seen": 469958656
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004331193580742227,
      "loss": 2.9649,
      "theoretical_loss": 3.9411518941908166,
      "tokens_seen": 470024192
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043310932798395184,
      "loss": 3.0462,
      "theoretical_loss": 3.9410920791998074,
      "tokens_seen": 470089728
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043309929789368107,
      "loss": 3.0932,
      "theoretical_loss": 3.941032274881649,
      "tokens_seen": 470155264
    },
    {
      "epoch": 0.14,
      "objective/train/advantage_avg": 0.4935116469860077,
      "objective/train/docs_used": 277851,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.9968011379241943,
      "objective/train/original_loss": 2.996800422668457,
      "objective/train/theoretical_loss": 3.940972481232949,
      "objective/train/tokens_used": 490680800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24367982149124146,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505897998809814,
      "objective/train/weighted_lm_loss": 3.1483120918273926,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0363596677780151,
      "theoretical_loss": 3.940972481232949,
      "tokens_seen": 470220800
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043308926780341025,
      "loss": 3.0853,
      "theoretical_loss": 3.940972481232949,
      "tokens_seen": 470220800
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043307923771313943,
      "loss": 2.8127,
      "theoretical_loss": 3.9409126982503184,
      "tokens_seen": 470286336
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004330692076228686,
      "loss": 3.0069,
      "theoretical_loss": 3.9408529259303684,
      "tokens_seen": 470351872
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043305917753259785,
      "loss": 2.9575,
      "theoretical_loss": 3.940793164269711,
      "tokens_seen": 470417408
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.000433049147442327,
      "loss": 2.9122,
      "theoretical_loss": 3.9407334132649616,
      "tokens_seen": 470482944
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004330391173520562,
      "loss": 2.9779,
      "theoretical_loss": 3.9406736729127365,
      "tokens_seen": 470548480
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043302908726178534,
      "loss": 3.0337,
      "theoretical_loss": 3.9406139432096525,
      "tokens_seen": 470614016
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004330190571715146,
      "loss": 3.1882,
      "theoretical_loss": 3.940554224152329,
      "tokens_seen": 470679552
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043300902708124376,
      "loss": 3.2631,
      "theoretical_loss": 3.9404945157373867,
      "tokens_seen": 470745088
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043299899699097294,
      "loss": 3.0639,
      "theoretical_loss": 3.940434817961448,
      "tokens_seen": 470810624
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004329889669007021,
      "loss": 2.9282,
      "theoretical_loss": 3.9403751308211366,
      "tokens_seen": 470876160
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004329789368104313,
      "loss": 3.0595,
      "theoretical_loss": 3.9403154543130783,
      "tokens_seen": 470941696
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004329689067201605,
      "loss": 3.3268,
      "theoretical_loss": 3.940255788433899,
      "tokens_seen": 471007232
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004329588766298897,
      "loss": 3.151,
      "theoretical_loss": 3.940196133180228,
      "tokens_seen": 471072768
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043294884653961884,
      "loss": 3.0138,
      "theoretical_loss": 3.9401364885486942,
      "tokens_seen": 471138304
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004329388164493481,
      "loss": 3.0851,
      "theoretical_loss": 3.94007685453593,
      "tokens_seen": 471203840
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004329287863590772,
      "loss": 3.0874,
      "theoretical_loss": 3.9400172311385684,
      "tokens_seen": 471269376
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043291875626880644,
      "loss": 3.0333,
      "theoretical_loss": 3.939957618353243,
      "tokens_seen": 471334912
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004329087261785356,
      "loss": 3.0755,
      "theoretical_loss": 3.9398980161765915,
      "tokens_seen": 471400448
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004328986960882648,
      "loss": 2.9794,
      "theoretical_loss": 3.9398384246052496,
      "tokens_seen": 471465984
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.000432888665997994,
      "loss": 2.8445,
      "theoretical_loss": 3.9397788436358585,
      "tokens_seen": 471531520
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004328786359077232,
      "loss": 2.841,
      "theoretical_loss": 3.939719273265057,
      "tokens_seen": 471597056
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043286860581745235,
      "loss": 2.8945,
      "theoretical_loss": 3.9396597134894886,
      "tokens_seen": 471662592
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004328585757271816,
      "loss": 2.9967,
      "theoretical_loss": 3.939600164305796,
      "tokens_seen": 471728128
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004328485456369107,
      "loss": 3.2286,
      "theoretical_loss": 3.9395406257106256,
      "tokens_seen": 471793664
    },
    {
      "epoch": 0.14,
      "objective/train/advantage_avg": 0.49385762214660645,
      "objective/train/docs_used": 279270,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.2785885334014893,
      "objective/train/original_loss": 2.2785890102386475,
      "objective/train/theoretical_loss": 3.939481097700623,
      "objective/train/tokens_used": 492319200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24401028454303741,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506261587142944,
      "objective/train/weighted_lm_loss": 2.3939712047576904,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0373471975326538,
      "theoretical_loss": 3.939481097700623,
      "tokens_seen": 471859200
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043283851554663994,
      "loss": 2.7684,
      "theoretical_loss": 3.939481097700623,
      "tokens_seen": 471859200
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004328284854563691,
      "loss": 2.8465,
      "theoretical_loss": 3.9394215802724375,
      "tokens_seen": 471924736
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004328184553660983,
      "loss": 3.0466,
      "theoretical_loss": 3.9393620734227186,
      "tokens_seen": 471990272
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004328084252758275,
      "loss": 2.9848,
      "theoretical_loss": 3.9393025771481174,
      "tokens_seen": 472055808
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043279839518555667,
      "loss": 3.2161,
      "theoretical_loss": 3.939243091445287,
      "tokens_seen": 472121344
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043278836509528585,
      "loss": 2.8625,
      "theoretical_loss": 3.9391836163108827,
      "tokens_seen": 472186880
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004327783350050151,
      "loss": 3.1705,
      "theoretical_loss": 3.9391241517415594,
      "tokens_seen": 472252416
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004327683049147442,
      "loss": 3.1356,
      "theoretical_loss": 3.9390646977339747,
      "tokens_seen": 472317952
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043275827482447345,
      "loss": 2.9825,
      "theoretical_loss": 3.9390052542847878,
      "tokens_seen": 472383488
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043274824473420263,
      "loss": 3.0142,
      "theoretical_loss": 3.938945821390659,
      "tokens_seen": 472449024
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004327382146439318,
      "loss": 3.0682,
      "theoretical_loss": 3.9388863990482506,
      "tokens_seen": 472514560
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.000432728184553661,
      "loss": 2.8123,
      "theoretical_loss": 3.938826987254226,
      "tokens_seen": 472580096
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043271815446339017,
      "loss": 2.9657,
      "theoretical_loss": 3.9387675860052505,
      "tokens_seen": 472645632
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043270812437311935,
      "loss": 2.8364,
      "theoretical_loss": 3.9387081952979903,
      "tokens_seen": 472711168
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004326980942828486,
      "loss": 3.0393,
      "theoretical_loss": 3.938648815129114,
      "tokens_seen": 472776704
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004326880641925777,
      "loss": 3.0253,
      "theoretical_loss": 3.9385894454952908,
      "tokens_seen": 472842240
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043267803410230695,
      "loss": 2.9283,
      "theoretical_loss": 3.9385300863931914,
      "tokens_seen": 472907776
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004326680040120361,
      "loss": 3.028,
      "theoretical_loss": 3.9384707378194896,
      "tokens_seen": 472973312
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004326579739217653,
      "loss": 3.1446,
      "theoretical_loss": 3.9384113997708585,
      "tokens_seen": 473038848
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004326479438314945,
      "loss": 2.9228,
      "theoretical_loss": 3.9383520722439744,
      "tokens_seen": 473104384
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004326379137412237,
      "loss": 3.2002,
      "theoretical_loss": 3.9382927552355147,
      "tokens_seen": 473169920
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043262788365095286,
      "loss": 2.6997,
      "theoretical_loss": 3.9382334487421575,
      "tokens_seen": 473235456
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043261785356068204,
      "loss": 3.1867,
      "theoretical_loss": 3.938174152760583,
      "tokens_seen": 473300992
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004326078234704112,
      "loss": 2.8643,
      "theoretical_loss": 3.9381148672874735,
      "tokens_seen": 473366528
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043259779338014045,
      "loss": 2.9911,
      "theoretical_loss": 3.938055592319511,
      "tokens_seen": 473432064
    },
    {
      "epoch": 0.14,
      "objective/train/advantage_avg": 0.4925135672092438,
      "objective/train/docs_used": 279967,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.9032363891601562,
      "objective/train/original_loss": 2.9032368659973145,
      "objective/train/theoretical_loss": 3.937996327853382,
      "objective/train/tokens_used": 493957600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24264630675315857,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0504847764968872,
      "objective/train/weighted_lm_loss": 3.049992322921753,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0451053380966187,
      "theoretical_loss": 3.937996327853382,
      "tokens_seen": 473497600
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004325877632898696,
      "loss": 3.0291,
      "theoretical_loss": 3.937996327853382,
      "tokens_seen": 473497600
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004325777331995988,
      "loss": 3.1526,
      "theoretical_loss": 3.9379370738857706,
      "tokens_seen": 473563136
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.000432567703109328,
      "loss": 3.004,
      "theoretical_loss": 3.937877830413366,
      "tokens_seen": 473628672
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004325576730190572,
      "loss": 3.1795,
      "theoretical_loss": 3.9378185974328574,
      "tokens_seen": 473694208
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043254764292878636,
      "loss": 3.0901,
      "theoretical_loss": 3.937759374940935,
      "tokens_seen": 473759744
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043253761283851554,
      "loss": 3.1146,
      "theoretical_loss": 3.9377001629342914,
      "tokens_seen": 473825280
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004325275827482447,
      "loss": 2.9411,
      "theoretical_loss": 3.9376409614096204,
      "tokens_seen": 473890816
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043251755265797396,
      "loss": 3.2427,
      "theoretical_loss": 3.9375817703636167,
      "tokens_seen": 473956352
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004325075225677031,
      "loss": 3.2565,
      "theoretical_loss": 3.937522589792977,
      "tokens_seen": 474021888
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004324974924774323,
      "loss": 3.02,
      "theoretical_loss": 3.9374634196944003,
      "tokens_seen": 474087424
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043248746238716145,
      "loss": 3.1917,
      "theoretical_loss": 3.9374042600645858,
      "tokens_seen": 474152960
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004324774322968907,
      "loss": 3.05,
      "theoretical_loss": 3.9373451109002344,
      "tokens_seen": 474218496
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043246740220661986,
      "loss": 2.9469,
      "theoretical_loss": 3.93728597219805,
      "tokens_seen": 474284032
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043245737211634904,
      "loss": 3.1815,
      "theoretical_loss": 3.9372268439547353,
      "tokens_seen": 474349568
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004324473420260782,
      "loss": 2.9465,
      "theoretical_loss": 3.9371677261669973,
      "tokens_seen": 474415104
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004324373119358074,
      "loss": 3.1666,
      "theoretical_loss": 3.9371086188315427,
      "tokens_seen": 474480640
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043242728184553664,
      "loss": 3.118,
      "theoretical_loss": 3.93704952194508,
      "tokens_seen": 474546176
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004324172517552658,
      "loss": 3.1717,
      "theoretical_loss": 3.936990435504319,
      "tokens_seen": 474611712
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.000432407221664995,
      "loss": 3.0663,
      "theoretical_loss": 3.936931359505972,
      "tokens_seen": 474677248
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004323971915747242,
      "loss": 3.1366,
      "theoretical_loss": 3.9368722939467524,
      "tokens_seen": 474742784
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004323871614844534,
      "loss": 3.1359,
      "theoretical_loss": 3.9368132388233743,
      "tokens_seen": 474808320
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043237713139418255,
      "loss": 3.0114,
      "theoretical_loss": 3.936754194132554,
      "tokens_seen": 474873856
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004323671013039118,
      "loss": 3.0154,
      "theoretical_loss": 3.936695159871009,
      "tokens_seen": 474939392
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004323570712136409,
      "loss": 3.1325,
      "theoretical_loss": 3.9366361360354585,
      "tokens_seen": 475004928
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043234704112337014,
      "loss": 2.9123,
      "theoretical_loss": 3.936577122622624,
      "tokens_seen": 475070464
    },
    {
      "epoch": 0.14,
      "objective/train/advantage_avg": 0.4944530129432678,
      "objective/train/docs_used": 280356,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.0068039894104004,
      "objective/train/original_loss": 3.0068042278289795,
      "objective/train/theoretical_loss": 3.936518119629225,
      "objective/train/tokens_used": 495596000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24453407526016235,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506882667541504,
      "objective/train/weighted_lm_loss": 3.1591176986694336,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.044721245765686,
      "theoretical_loss": 3.936518119629225,
      "tokens_seen": 475136000
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004323370110330993,
      "loss": 3.0995,
      "theoretical_loss": 3.936518119629225,
      "tokens_seen": 475136000
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004323269809428285,
      "loss": 3.257,
      "theoretical_loss": 3.9364591270519877,
      "tokens_seen": 475201536
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004323169508525577,
      "loss": 3.2367,
      "theoretical_loss": 3.936400144887637,
      "tokens_seen": 475267072
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043230692076228687,
      "loss": 2.9676,
      "theoretical_loss": 3.9363411731328974,
      "tokens_seen": 475332608
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043229689067201605,
      "loss": 2.9631,
      "theoretical_loss": 3.9362822117844987,
      "tokens_seen": 475398144
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004322868605817453,
      "loss": 3.1164,
      "theoretical_loss": 3.936223260839169,
      "tokens_seen": 475463680
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004322768304914744,
      "loss": 2.7645,
      "theoretical_loss": 3.936164320293641,
      "tokens_seen": 475529216
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043226680040120365,
      "loss": 2.9982,
      "theoretical_loss": 3.9361053901446454,
      "tokens_seen": 475594752
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043225677031093283,
      "loss": 3.0555,
      "theoretical_loss": 3.9360464703889173,
      "tokens_seen": 475660288
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.000432246740220662,
      "loss": 2.9178,
      "theoretical_loss": 3.935987561023192,
      "tokens_seen": 475725824
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004322367101303912,
      "loss": 2.9312,
      "theoretical_loss": 3.9359286620442058,
      "tokens_seen": 475791360
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043222668004012037,
      "loss": 3.0464,
      "theoretical_loss": 3.9358697734486965,
      "tokens_seen": 475856896
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043221664994984955,
      "loss": 3.1983,
      "theoretical_loss": 3.9358108952334057,
      "tokens_seen": 475922432
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004322066198595788,
      "loss": 3.0833,
      "theoretical_loss": 3.9357520273950732,
      "tokens_seen": 475987968
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004321965897693079,
      "loss": 2.9502,
      "theoretical_loss": 3.9356931699304427,
      "tokens_seen": 476053504
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043218655967903715,
      "loss": 3.0765,
      "theoretical_loss": 3.9356343228362567,
      "tokens_seen": 476119040
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004321765295887663,
      "loss": 2.9977,
      "theoretical_loss": 3.935575486109263,
      "tokens_seen": 476184576
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004321664994984955,
      "loss": 2.9737,
      "theoretical_loss": 3.9355166597462086,
      "tokens_seen": 476250112
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004321564694082247,
      "loss": 2.7782,
      "theoretical_loss": 3.9354578437438406,
      "tokens_seen": 476315648
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004321464393179539,
      "loss": 2.9228,
      "theoretical_loss": 3.9353990380989097,
      "tokens_seen": 476381184
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043213640922768306,
      "loss": 2.7757,
      "theoretical_loss": 3.935340242808168,
      "tokens_seen": 476446720
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043212637913741224,
      "loss": 3.0632,
      "theoretical_loss": 3.935281457868368,
      "tokens_seen": 476512256
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004321163490471414,
      "loss": 2.8666,
      "theoretical_loss": 3.9352226832762645,
      "tokens_seen": 476577792
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043210631895687065,
      "loss": 3.3191,
      "theoretical_loss": 3.9351639190286134,
      "tokens_seen": 476643328
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004320962888665998,
      "loss": 3.2801,
      "theoretical_loss": 3.935105165122172,
      "tokens_seen": 476708864
    },
    {
      "epoch": 0.14,
      "objective/train/advantage_avg": 0.49446460604667664,
      "objective/train/docs_used": 281398,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.231257438659668,
      "objective/train/original_loss": 3.231257915496826,
      "objective/train/theoretical_loss": 3.9350464215537,
      "objective/train/tokens_used": 497234400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24456366896629333,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506895780563354,
      "objective/train/weighted_lm_loss": 3.3949594497680664,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0442860126495361,
      "theoretical_loss": 3.9350464215537,
      "tokens_seen": 476774400
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.000432086258776329,
      "loss": 2.9331,
      "theoretical_loss": 3.9350464215537,
      "tokens_seen": 476774400
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004320762286860582,
      "loss": 3.0837,
      "theoretical_loss": 3.934987688319956,
      "tokens_seen": 476839936
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004320661985957874,
      "loss": 2.9952,
      "theoretical_loss": 3.9349289654177033,
      "tokens_seen": 476905472
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043205616850551656,
      "loss": 3.0628,
      "theoretical_loss": 3.934870252843705,
      "tokens_seen": 476971008
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043204613841524574,
      "loss": 3.0107,
      "theoretical_loss": 3.934811550594725,
      "tokens_seen": 477036544
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004320361083249749,
      "loss": 3.0106,
      "theoretical_loss": 3.9347528586675304,
      "tokens_seen": 477102080
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043202607823470416,
      "loss": 3.3245,
      "theoretical_loss": 3.934694177058888,
      "tokens_seen": 477167616
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004320160481444333,
      "loss": 3.1698,
      "theoretical_loss": 3.9346355057655673,
      "tokens_seen": 477233152
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004320060180541625,
      "loss": 2.7398,
      "theoretical_loss": 3.9345768447843392,
      "tokens_seen": 477298688
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043199598796389165,
      "loss": 3.2594,
      "theoretical_loss": 3.934518194111975,
      "tokens_seen": 477364224
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004319859578736209,
      "loss": 3.1564,
      "theoretical_loss": 3.934459553745249,
      "tokens_seen": 477429760
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043197592778335006,
      "loss": 3.1759,
      "theoretical_loss": 3.9344009236809354,
      "tokens_seen": 477495296
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043196589769307924,
      "loss": 3.2076,
      "theoretical_loss": 3.9343423039158107,
      "tokens_seen": 477560832
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004319558676028084,
      "loss": 3.3597,
      "theoretical_loss": 3.9342836944466533,
      "tokens_seen": 477626368
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004319458375125376,
      "loss": 2.9235,
      "theoretical_loss": 3.9342250952702416,
      "tokens_seen": 477691904
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004319358074222668,
      "loss": 3.0863,
      "theoretical_loss": 3.934166506383357,
      "tokens_seen": 477757440
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.000431925777331996,
      "loss": 3.0767,
      "theoretical_loss": 3.934107927782781,
      "tokens_seen": 477822976
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043191574724172515,
      "loss": 3.1171,
      "theoretical_loss": 3.934049359465297,
      "tokens_seen": 477888512
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004319057171514544,
      "loss": 3.0928,
      "theoretical_loss": 3.933990801427691,
      "tokens_seen": 477954048
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043189568706118357,
      "loss": 3.23,
      "theoretical_loss": 3.9339322536667494,
      "tokens_seen": 478019584
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043188565697091275,
      "loss": 3.1435,
      "theoretical_loss": 3.93387371617926,
      "tokens_seen": 478085120
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043187562688064193,
      "loss": 3.0137,
      "theoretical_loss": 3.9338151889620114,
      "tokens_seen": 478150656
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004318655967903711,
      "loss": 3.1378,
      "theoretical_loss": 3.933756672011796,
      "tokens_seen": 478216192
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004318555667001003,
      "loss": 3.122,
      "theoretical_loss": 3.933698165325404,
      "tokens_seen": 478281728
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004318455366098295,
      "loss": 3.0983,
      "theoretical_loss": 3.933639668899631,
      "tokens_seen": 478347264
    },
    {
      "epoch": 0.14,
      "objective/train/advantage_avg": 0.49408531188964844,
      "objective/train/docs_used": 281924,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.797677993774414,
      "objective/train/original_loss": 2.797678232192993,
      "objective/train/theoretical_loss": 3.933581182731271,
      "objective/train/tokens_used": 498872800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.244193896651268,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050649881362915,
      "objective/train/weighted_lm_loss": 2.9392759799957275,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0338950157165527,
      "theoretical_loss": 3.933581182731271,
      "tokens_seen": 478412800
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.00043183550651955865,
      "loss": 2.9643,
      "theoretical_loss": 3.933581182731271,
      "tokens_seen": 478412800
    },
    {
      "epoch": 0.14,
      "learning_rate": 0.0004318254764292879,
      "loss": 3.3039,
      "theoretical_loss": 3.9335227068171212,
      "tokens_seen": 478478336
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.000431815446339017,
      "loss": 2.8884,
      "theoretical_loss": 3.9334642411539793,
      "tokens_seen": 478543872
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00043180541624874625,
      "loss": 2.9931,
      "theoretical_loss": 3.9334057857386453,
      "tokens_seen": 478609408
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00043179538615847543,
      "loss": 2.7981,
      "theoretical_loss": 3.933347340567919,
      "tokens_seen": 478674944
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004317853560682046,
      "loss": 2.9827,
      "theoretical_loss": 3.9332889056386042,
      "tokens_seen": 478740480
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004317753259779338,
      "loss": 2.933,
      "theoretical_loss": 3.933230480947504,
      "tokens_seen": 478806016
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00043176529588766303,
      "loss": 3.013,
      "theoretical_loss": 3.933172066491423,
      "tokens_seen": 478871552
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00043175526579739215,
      "loss": 3.1053,
      "theoretical_loss": 3.9331136622671687,
      "tokens_seen": 478937088
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004317452357071214,
      "loss": 2.9711,
      "theoretical_loss": 3.9330552682715485,
      "tokens_seen": 479002624
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004317352056168505,
      "loss": 2.8828,
      "theoretical_loss": 3.9329968845013727,
      "tokens_seen": 479068160
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00043172517552657975,
      "loss": 3.0812,
      "theoretical_loss": 3.932938510953452,
      "tokens_seen": 479133696
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00043171514543630893,
      "loss": 3.178,
      "theoretical_loss": 3.9328801476245987,
      "tokens_seen": 479199232
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004317051153460381,
      "loss": 2.8683,
      "theoretical_loss": 3.932821794511626,
      "tokens_seen": 479264768
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004316950852557673,
      "loss": 3.0418,
      "theoretical_loss": 3.9327634516113497,
      "tokens_seen": 479330304
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004316850551654965,
      "loss": 3.1097,
      "theoretical_loss": 3.932705118920587,
      "tokens_seen": 479395840
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004316750250752257,
      "loss": 3.2029,
      "theoretical_loss": 3.9326467964361553,
      "tokens_seen": 479461376
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004316649949849549,
      "loss": 3.2551,
      "theoretical_loss": 3.9325884841548735,
      "tokens_seen": 479526912
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004316549648946841,
      "loss": 2.7585,
      "theoretical_loss": 3.932530182073564,
      "tokens_seen": 479592448
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00043164493480441326,
      "loss": 3.0977,
      "theoretical_loss": 3.9324718901890487,
      "tokens_seen": 479657984
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00043163490471414244,
      "loss": 3.0461,
      "theoretical_loss": 3.932413608498151,
      "tokens_seen": 479723520
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004316248746238716,
      "loss": 2.9119,
      "theoretical_loss": 3.932355336997696,
      "tokens_seen": 479789056
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00043161484453360085,
      "loss": 2.972,
      "theoretical_loss": 3.9322970756845113,
      "tokens_seen": 479854592
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00043160481444333,
      "loss": 3.168,
      "theoretical_loss": 3.9322388245554234,
      "tokens_seen": 479920128
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004315947843530592,
      "loss": 3.0179,
      "theoretical_loss": 3.9321805836072627,
      "tokens_seen": 479985664
    },
    {
      "epoch": 0.15,
      "objective/train/advantage_avg": 0.4941039979457855,
      "objective/train/docs_used": 283309,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.4523379802703857,
      "objective/train/original_loss": 3.4523375034332275,
      "objective/train/theoretical_loss": 3.9321223528368607,
      "objective/train/tokens_used": 500511200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.244218647480011,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506519079208374,
      "objective/train/weighted_lm_loss": 3.6272056102752686,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0417574644088745,
      "theoretical_loss": 3.9321223528368607,
      "tokens_seen": 480051200
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004315847542627884,
      "loss": 3.1433,
      "theoretical_loss": 3.9321223528368607,
      "tokens_seen": 480051200
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004315747241725176,
      "loss": 3.0907,
      "theoretical_loss": 3.932064132241049,
      "tokens_seen": 480116736
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00043156469408224676,
      "loss": 3.1114,
      "theoretical_loss": 3.932005921816661,
      "tokens_seen": 480182272
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00043155466399197594,
      "loss": 3.0695,
      "theoretical_loss": 3.9319477215605323,
      "tokens_seen": 480247808
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004315446339017051,
      "loss": 3.0719,
      "theoretical_loss": 3.9318895314694995,
      "tokens_seen": 480313344
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00043153460381143436,
      "loss": 3.0657,
      "theoretical_loss": 3.931831351540401,
      "tokens_seen": 480378880
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004315245737211635,
      "loss": 2.8101,
      "theoretical_loss": 3.931773181770075,
      "tokens_seen": 480444416
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004315145436308927,
      "loss": 2.9361,
      "theoretical_loss": 3.931715022155364,
      "tokens_seen": 480509952
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00043150451354062185,
      "loss": 2.8826,
      "theoretical_loss": 3.9316568726931083,
      "tokens_seen": 480575488
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004314944834503511,
      "loss": 2.9489,
      "theoretical_loss": 3.931598733380153,
      "tokens_seen": 480641024
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00043148445336008026,
      "loss": 3.2691,
      "theoretical_loss": 3.9315406042133425,
      "tokens_seen": 480706560
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00043147442326980944,
      "loss": 2.9792,
      "theoretical_loss": 3.931482485189524,
      "tokens_seen": 480772096
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004314643931795386,
      "loss": 2.8329,
      "theoretical_loss": 3.9314243763055443,
      "tokens_seen": 480837632
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004314543630892678,
      "loss": 3.1068,
      "theoretical_loss": 3.9313662775582534,
      "tokens_seen": 480903168
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.000431444332998997,
      "loss": 3.0114,
      "theoretical_loss": 3.9313081889445014,
      "tokens_seen": 480968704
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004314343029087262,
      "loss": 3.106,
      "theoretical_loss": 3.931250110461141,
      "tokens_seen": 481034240
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00043142427281845535,
      "loss": 3.1156,
      "theoretical_loss": 3.9311920421050255,
      "tokens_seen": 481099776
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004314142427281846,
      "loss": 2.8787,
      "theoretical_loss": 3.9311339838730097,
      "tokens_seen": 481165312
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00043140421263791377,
      "loss": 2.8752,
      "theoretical_loss": 3.9310759357619505,
      "tokens_seen": 481230848
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00043139418254764295,
      "loss": 3.0991,
      "theoretical_loss": 3.9310178977687045,
      "tokens_seen": 481296384
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00043138415245737213,
      "loss": 2.9701,
      "theoretical_loss": 3.930959869890132,
      "tokens_seen": 481361920
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004313741223671013,
      "loss": 2.902,
      "theoretical_loss": 3.9309018521230925,
      "tokens_seen": 481427456
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004313640922768305,
      "loss": 2.9987,
      "theoretical_loss": 3.930843844464449,
      "tokens_seen": 481492992
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004313540621865597,
      "loss": 3.0723,
      "theoretical_loss": 3.9307858469110633,
      "tokens_seen": 481558528
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00043134403209628885,
      "loss": 2.9453,
      "theoretical_loss": 3.9307278594598016,
      "tokens_seen": 481624064
    },
    {
      "epoch": 0.15,
      "objective/train/advantage_avg": 0.49569082260131836,
      "objective/train/docs_used": 283520,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.126472234725952,
      "objective/train/original_loss": 3.1264724731445312,
      "objective/train/theoretical_loss": 3.930669882107529,
      "objective/train/tokens_used": 502149600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24575699865818024,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0508184432983398,
      "objective/train/weighted_lm_loss": 3.2853171825408936,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0417574644088745,
      "theoretical_loss": 3.930669882107529,
      "tokens_seen": 481689600
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004313340020060181,
      "loss": 3.1707,
      "theoretical_loss": 3.930669882107529,
      "tokens_seen": 481689600
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004313239719157472,
      "loss": 2.916,
      "theoretical_loss": 3.9306119148511147,
      "tokens_seen": 481755136
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00043131394182547645,
      "loss": 3.0722,
      "theoretical_loss": 3.930553957687426,
      "tokens_seen": 481820672
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00043130391173520563,
      "loss": 3.1177,
      "theoretical_loss": 3.930496010613333,
      "tokens_seen": 481886208
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004312938816449348,
      "loss": 3.1653,
      "theoretical_loss": 3.9304380736257087,
      "tokens_seen": 481951744
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.000431283851554664,
      "loss": 3.159,
      "theoretical_loss": 3.9303801467214248,
      "tokens_seen": 482017280
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00043127382146439323,
      "loss": 3.0825,
      "theoretical_loss": 3.9303222298973575,
      "tokens_seen": 482082816
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00043126379137412236,
      "loss": 3.2288,
      "theoretical_loss": 3.930264323150381,
      "tokens_seen": 482148352
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004312537612838516,
      "loss": 3.2629,
      "theoretical_loss": 3.9302064264773735,
      "tokens_seen": 482213888
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004312437311935807,
      "loss": 3.3477,
      "theoretical_loss": 3.930148539875214,
      "tokens_seen": 482279424
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00043123370110330995,
      "loss": 3.2284,
      "theoretical_loss": 3.930090663340782,
      "tokens_seen": 482344960
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00043122367101303913,
      "loss": 2.9519,
      "theoretical_loss": 3.9300327968709587,
      "tokens_seen": 482410496
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004312136409227683,
      "loss": 3.0068,
      "theoretical_loss": 3.9299749404626274,
      "tokens_seen": 482476032
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004312036108324975,
      "loss": 3.1976,
      "theoretical_loss": 3.929917094112673,
      "tokens_seen": 482541568
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004311935807422267,
      "loss": 2.7682,
      "theoretical_loss": 3.92985925781798,
      "tokens_seen": 482607104
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00043118355065195586,
      "loss": 2.6728,
      "theoretical_loss": 3.9298014315754357,
      "tokens_seen": 482672640
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004311735205616851,
      "loss": 3.0331,
      "theoretical_loss": 3.9297436153819287,
      "tokens_seen": 482738176
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004311634904714142,
      "loss": 3.2047,
      "theoretical_loss": 3.929685809234349,
      "tokens_seen": 482803712
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00043115346038114346,
      "loss": 2.9986,
      "theoretical_loss": 3.929628013129587,
      "tokens_seen": 482869248
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004311434302908726,
      "loss": 3.195,
      "theoretical_loss": 3.929570227064536,
      "tokens_seen": 482934784
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004311334002006018,
      "loss": 3.1813,
      "theoretical_loss": 3.9295124510360893,
      "tokens_seen": 483000320
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.000431123370110331,
      "loss": 2.7921,
      "theoretical_loss": 3.929454685041143,
      "tokens_seen": 483065856
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004311133400200602,
      "loss": 2.9955,
      "theoretical_loss": 3.929396929076594,
      "tokens_seen": 483131392
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00043110330992978936,
      "loss": 2.9694,
      "theoretical_loss": 3.929339183139339,
      "tokens_seen": 483196928
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004310932798395186,
      "loss": 3.1768,
      "theoretical_loss": 3.9292814472262787,
      "tokens_seen": 483262464
    },
    {
      "epoch": 0.15,
      "objective/train/advantage_avg": 0.49459001421928406,
      "objective/train/docs_used": 284734,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.8600213527679443,
      "objective/train/original_loss": 2.8600211143493652,
      "objective/train/theoretical_loss": 3.9292237213343135,
      "objective/train/tokens_used": 503788000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24467338621616364,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050702691078186,
      "objective/train/weighted_lm_loss": 3.004897356033325,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0437887907028198,
      "theoretical_loss": 3.9292237213343135,
      "tokens_seen": 483328000
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004310832497492477,
      "loss": 3.1423,
      "theoretical_loss": 3.9292237213343135,
      "tokens_seen": 483328000
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00043107321965897696,
      "loss": 3.3369,
      "theoretical_loss": 3.9291660054603454,
      "tokens_seen": 483393536
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004310631895687061,
      "loss": 2.9209,
      "theoretical_loss": 3.9291082996012783,
      "tokens_seen": 483459072
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004310531594784353,
      "loss": 3.0719,
      "theoretical_loss": 3.9290506037540167,
      "tokens_seen": 483524608
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004310431293881645,
      "loss": 3.0531,
      "theoretical_loss": 3.928992917915468,
      "tokens_seen": 483590144
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004310330992978937,
      "loss": 3.1544,
      "theoretical_loss": 3.928935242082539,
      "tokens_seen": 483655680
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00043102306920762286,
      "loss": 2.9613,
      "theoretical_loss": 3.9288775762521393,
      "tokens_seen": 483721216
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00043101303911735205,
      "loss": 2.9657,
      "theoretical_loss": 3.928819920421179,
      "tokens_seen": 483786752
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004310030090270812,
      "loss": 2.9865,
      "theoretical_loss": 3.9287622745865702,
      "tokens_seen": 483852288
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00043099297893681046,
      "loss": 3.2067,
      "theoretical_loss": 3.9287046387452262,
      "tokens_seen": 483917824
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004309829488465396,
      "loss": 3.1887,
      "theoretical_loss": 3.928647012894061,
      "tokens_seen": 483983360
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004309729187562688,
      "loss": 2.8886,
      "theoretical_loss": 3.9285893970299917,
      "tokens_seen": 484048896
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00043096288866599795,
      "loss": 3.2256,
      "theoretical_loss": 3.928531791149934,
      "tokens_seen": 484114432
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004309528585757272,
      "loss": 3.0675,
      "theoretical_loss": 3.928474195250808,
      "tokens_seen": 484179968
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00043094282848545637,
      "loss": 3.1613,
      "theoretical_loss": 3.928416609329533,
      "tokens_seen": 484245504
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00043093279839518555,
      "loss": 3.2329,
      "theoretical_loss": 3.928359033383031,
      "tokens_seen": 484311040
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004309227683049148,
      "loss": 3.2725,
      "theoretical_loss": 3.9283014674082244,
      "tokens_seen": 484376576
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00043091273821464397,
      "loss": 3.0497,
      "theoretical_loss": 3.9282439114020375,
      "tokens_seen": 484442112
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00043090270812437315,
      "loss": 3.1065,
      "theoretical_loss": 3.9281863653613964,
      "tokens_seen": 484507648
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00043089267803410233,
      "loss": 3.3226,
      "theoretical_loss": 3.928128829283226,
      "tokens_seen": 484573184
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004308826479438315,
      "loss": 3.342,
      "theoretical_loss": 3.928071303164457,
      "tokens_seen": 484638720
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004308726178535607,
      "loss": 2.8722,
      "theoretical_loss": 3.928013787002018,
      "tokens_seen": 484704256
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004308625877632899,
      "loss": 3.0441,
      "theoretical_loss": 3.9279562807928396,
      "tokens_seen": 484769792
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00043085255767301905,
      "loss": 3.0805,
      "theoretical_loss": 3.9278987845338547,
      "tokens_seen": 484835328
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004308425275827483,
      "loss": 2.9666,
      "theoretical_loss": 3.9278412982219963,
      "tokens_seen": 484900864
    },
    {
      "epoch": 0.15,
      "objective/train/advantage_avg": 0.495425283908844,
      "objective/train/docs_used": 285372,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.152074098587036,
      "objective/train/original_loss": 3.1520743370056152,
      "objective/train/theoretical_loss": 3.927783821854201,
      "objective/train/tokens_used": 505426400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24551770091056824,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050790548324585,
      "objective/train/weighted_lm_loss": 3.3121497631073,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0338950157165527,
      "theoretical_loss": 3.927783821854201,
      "tokens_seen": 484966400
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004308324974924774,
      "loss": 3.1198,
      "theoretical_loss": 3.927783821854201,
      "tokens_seen": 484966400
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00043082246740220665,
      "loss": 3.1853,
      "theoretical_loss": 3.9277263554274033,
      "tokens_seen": 485031936
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00043081243731193583,
      "loss": 2.9648,
      "theoretical_loss": 3.927668898938542,
      "tokens_seen": 485097472
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.000430802407221665,
      "loss": 3.1781,
      "theoretical_loss": 3.9276114523845562,
      "tokens_seen": 485163008
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004307923771313942,
      "loss": 2.8746,
      "theoretical_loss": 3.9275540157623867,
      "tokens_seen": 485228544
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00043078234704112343,
      "loss": 3.1823,
      "theoretical_loss": 3.927496589068974,
      "tokens_seen": 485294080
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00043077231695085256,
      "loss": 3.0649,
      "theoretical_loss": 3.9274391723012627,
      "tokens_seen": 485359616
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004307622868605818,
      "loss": 3.0562,
      "theoretical_loss": 3.9273817654561967,
      "tokens_seen": 485425152
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004307522567703109,
      "loss": 3.1606,
      "theoretical_loss": 3.927324368530723,
      "tokens_seen": 485490688
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00043074222668004015,
      "loss": 2.8455,
      "theoretical_loss": 3.927266981521787,
      "tokens_seen": 485556224
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00043073219658976933,
      "loss": 3.0817,
      "theoretical_loss": 3.927209604426338,
      "tokens_seen": 485621760
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004307221664994985,
      "loss": 3.0523,
      "theoretical_loss": 3.9271522372413266,
      "tokens_seen": 485687296
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004307121364092277,
      "loss": 3.2755,
      "theoretical_loss": 3.927094879963704,
      "tokens_seen": 485752832
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004307021063189569,
      "loss": 3.2691,
      "theoretical_loss": 3.927037532590422,
      "tokens_seen": 485818368
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00043069207622868606,
      "loss": 3.1635,
      "theoretical_loss": 3.9269801951184355,
      "tokens_seen": 485883904
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004306820461384153,
      "loss": 3.2541,
      "theoretical_loss": 3.9269228675446994,
      "tokens_seen": 485949440
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004306720160481444,
      "loss": 3.024,
      "theoretical_loss": 3.9268655498661706,
      "tokens_seen": 486014976
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00043066198595787366,
      "loss": 2.9739,
      "theoretical_loss": 3.926808242079807,
      "tokens_seen": 486080512
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004306519558676028,
      "loss": 3.1989,
      "theoretical_loss": 3.9267509441825688,
      "tokens_seen": 486146048
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.000430641925777332,
      "loss": 3.0271,
      "theoretical_loss": 3.926693656171415,
      "tokens_seen": 486211584
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004306318956870612,
      "loss": 3.2498,
      "theoretical_loss": 3.9266363780433093,
      "tokens_seen": 486277120
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004306218655967904,
      "loss": 3.0299,
      "theoretical_loss": 3.926579109795214,
      "tokens_seen": 486342656
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00043061183550651956,
      "loss": 3.0142,
      "theoretical_loss": 3.926521851424095,
      "tokens_seen": 486408192
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004306018054162488,
      "loss": 3.0376,
      "theoretical_loss": 3.9264646029269175,
      "tokens_seen": 486473728
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004305917753259779,
      "loss": 2.8882,
      "theoretical_loss": 3.926407364300649,
      "tokens_seen": 486539264
    },
    {
      "epoch": 0.15,
      "objective/train/advantage_avg": 0.4927351474761963,
      "objective/train/docs_used": 285999,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.0994911193847656,
      "objective/train/original_loss": 3.0994908809661865,
      "objective/train/theoretical_loss": 3.9263501355422585,
      "objective/train/tokens_used": 507064800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24288056790828705,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505081415176392,
      "objective/train/weighted_lm_loss": 3.2561938762664795,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0417574644088745,
      "theoretical_loss": 3.9263501355422585,
      "tokens_seen": 486604800
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00043058174523570716,
      "loss": 3.2387,
      "theoretical_loss": 3.9263501355422585,
      "tokens_seen": 486604800
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004305717151454363,
      "loss": 2.9587,
      "theoretical_loss": 3.926292916648716,
      "tokens_seen": 486670336
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004305616850551655,
      "loss": 2.9779,
      "theoretical_loss": 3.926235707616994,
      "tokens_seen": 486735872
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004305516549648947,
      "loss": 2.9733,
      "theoretical_loss": 3.926178508444064,
      "tokens_seen": 486801408
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004305416248746239,
      "loss": 3.0478,
      "theoretical_loss": 3.9261213191269,
      "tokens_seen": 486866944
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00043053159478435306,
      "loss": 3.0301,
      "theoretical_loss": 3.9260641396624787,
      "tokens_seen": 486932480
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00043052156469408225,
      "loss": 3.1622,
      "theoretical_loss": 3.9260069700477755,
      "tokens_seen": 486998016
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004305115346038114,
      "loss": 3.0496,
      "theoretical_loss": 3.9259498102797696,
      "tokens_seen": 487063552
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00043050150451354066,
      "loss": 3.0221,
      "theoretical_loss": 3.9258926603554407,
      "tokens_seen": 487129088
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004304914744232698,
      "loss": 2.8636,
      "theoretical_loss": 3.925835520271768,
      "tokens_seen": 487194624
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.000430481444332999,
      "loss": 2.9064,
      "theoretical_loss": 3.9257783900257355,
      "tokens_seen": 487260160
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00043047141424272815,
      "loss": 2.9742,
      "theoretical_loss": 3.9257212696143258,
      "tokens_seen": 487325696
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004304613841524574,
      "loss": 3.0105,
      "theoretical_loss": 3.9256641590345236,
      "tokens_seen": 487391232
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00043045135406218657,
      "loss": 3.193,
      "theoretical_loss": 3.9256070582833145,
      "tokens_seen": 487456768
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00043044132397191575,
      "loss": 3.1508,
      "theoretical_loss": 3.9255499673576875,
      "tokens_seen": 487522304
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00043043129388164493,
      "loss": 3.0368,
      "theoretical_loss": 3.9254928862546303,
      "tokens_seen": 487587840
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00043042126379137417,
      "loss": 3.0526,
      "theoretical_loss": 3.9254358149711335,
      "tokens_seen": 487653376
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004304112337011033,
      "loss": 3.2948,
      "theoretical_loss": 3.925378753504188,
      "tokens_seen": 487718912
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00043040120361083253,
      "loss": 2.8635,
      "theoretical_loss": 3.925321701850787,
      "tokens_seen": 487784448
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00043039117352056165,
      "loss": 3.1259,
      "theoretical_loss": 3.925264660007924,
      "tokens_seen": 487849984
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004303811434302909,
      "loss": 3.213,
      "theoretical_loss": 3.925207627972596,
      "tokens_seen": 487915520
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00043037111334002007,
      "loss": 3.0189,
      "theoretical_loss": 3.9251506057417975,
      "tokens_seen": 487981056
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00043036108324974925,
      "loss": 3.1772,
      "theoretical_loss": 3.9250935933125284,
      "tokens_seen": 488046592
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00043035105315947843,
      "loss": 3.2011,
      "theoretical_loss": 3.925036590681787,
      "tokens_seen": 488112128
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004303410230692076,
      "loss": 2.8267,
      "theoretical_loss": 3.9249795978465745,
      "tokens_seen": 488177664
    },
    {
      "epoch": 0.15,
      "objective/train/advantage_avg": 0.4929489195346832,
      "objective/train/docs_used": 287428,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.9665658473968506,
      "objective/train/original_loss": 2.966566324234009,
      "objective/train/theoretical_loss": 3.9249226148038927,
      "objective/train/tokens_used": 508703200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24314060807228088,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505309104919434,
      "objective/train/weighted_lm_loss": 3.116513967514038,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 3.9249226148038927,
      "tokens_seen": 488243200
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004303309929789368,
      "loss": 3.0953,
      "theoretical_loss": 3.9249226148038927,
      "tokens_seen": 488243200
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00043032096288866603,
      "loss": 3.0286,
      "theoretical_loss": 3.924865641550745,
      "tokens_seen": 488308736
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00043031093279839516,
      "loss": 3.2169,
      "theoretical_loss": 3.9248086780841365,
      "tokens_seen": 488374272
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004303009027081244,
      "loss": 3.093,
      "theoretical_loss": 3.924751724401073,
      "tokens_seen": 488439808
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004302908726178535,
      "loss": 3.0577,
      "theoretical_loss": 3.9246947804985606,
      "tokens_seen": 488505344
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00043028084252758276,
      "loss": 2.9912,
      "theoretical_loss": 3.9246378463736096,
      "tokens_seen": 488570880
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00043027081243731194,
      "loss": 2.8428,
      "theoretical_loss": 3.9245809220232295,
      "tokens_seen": 488636416
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004302607823470411,
      "loss": 2.9819,
      "theoretical_loss": 3.9245240074444308,
      "tokens_seen": 488701952
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004302507522567703,
      "loss": 3.1941,
      "theoretical_loss": 3.9244671026342273,
      "tokens_seen": 488767488
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00043024072216649953,
      "loss": 3.196,
      "theoretical_loss": 3.9244102075896317,
      "tokens_seen": 488833024
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00043023069207622866,
      "loss": 3.1792,
      "theoretical_loss": 3.924353322307659,
      "tokens_seen": 488898560
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004302206619859579,
      "loss": 3.0277,
      "theoretical_loss": 3.924296446785328,
      "tokens_seen": 488964096
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.000430210631895687,
      "loss": 3.2441,
      "theoretical_loss": 3.9242395810196538,
      "tokens_seen": 489029632
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00043020060180541626,
      "loss": 2.7819,
      "theoretical_loss": 3.9241827250076566,
      "tokens_seen": 489095168
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00043019057171514544,
      "loss": 2.9077,
      "theoretical_loss": 3.9241258787463567,
      "tokens_seen": 489160704
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004301805416248746,
      "loss": 2.8301,
      "theoretical_loss": 3.9240690422327766,
      "tokens_seen": 489226240
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00043017051153460386,
      "loss": 2.9224,
      "theoretical_loss": 3.9240122154639385,
      "tokens_seen": 489291776
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.000430160481444333,
      "loss": 3.0911,
      "theoretical_loss": 3.923955398436867,
      "tokens_seen": 489357312
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004301504513540622,
      "loss": 3.1421,
      "theoretical_loss": 3.9238985911485873,
      "tokens_seen": 489422848
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004301404212637914,
      "loss": 2.7554,
      "theoretical_loss": 3.923841793596128,
      "tokens_seen": 489488384
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004301303911735206,
      "loss": 2.9636,
      "theoretical_loss": 3.923785005776515,
      "tokens_seen": 489553920
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00043012036108324976,
      "loss": 2.9689,
      "theoretical_loss": 3.92372822768678,
      "tokens_seen": 489619456
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.000430110330992979,
      "loss": 3.0284,
      "theoretical_loss": 3.923671459323953,
      "tokens_seen": 489684992
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004301003009027081,
      "loss": 3.1219,
      "theoretical_loss": 3.923614700685066,
      "tokens_seen": 489750528
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00043009027081243736,
      "loss": 3.1596,
      "theoretical_loss": 3.9235579517671524,
      "tokens_seen": 489816064
    },
    {
      "epoch": 0.15,
      "objective/train/advantage_avg": 0.49667081236839294,
      "objective/train/docs_used": 288116,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.9261648654937744,
      "objective/train/original_loss": 2.9261648654937744,
      "objective/train/theoretical_loss": 3.9235012125672473,
      "objective/train/tokens_used": 510341600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2467307448387146,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0509213209152222,
      "objective/train/weighted_lm_loss": 3.0750391483306885,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0408108234405518,
      "theoretical_loss": 3.9235012125672473,
      "tokens_seen": 489881600
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004300802407221665,
      "loss": 3.1459,
      "theoretical_loss": 3.9235012125672473,
      "tokens_seen": 489881600
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004300702106318957,
      "loss": 2.9628,
      "theoretical_loss": 3.9234444830823874,
      "tokens_seen": 489947136
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004300601805416249,
      "loss": 3.0177,
      "theoretical_loss": 3.9233877633096093,
      "tokens_seen": 490012672
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004300501504513541,
      "loss": 3.1439,
      "theoretical_loss": 3.9233310532459518,
      "tokens_seen": 490078208
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00043004012036108326,
      "loss": 2.9084,
      "theoretical_loss": 3.9232743528884546,
      "tokens_seen": 490143744
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00043003009027081245,
      "loss": 3.224,
      "theoretical_loss": 3.92321766223416,
      "tokens_seen": 490209280
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00043002006018054163,
      "loss": 3.0177,
      "theoretical_loss": 3.92316098128011,
      "tokens_seen": 490274816
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00043001003009027086,
      "loss": 3.0836,
      "theoretical_loss": 3.923104310023348,
      "tokens_seen": 490340352
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00043,
      "loss": 3.2779,
      "theoretical_loss": 3.92304764846092,
      "tokens_seen": 490405888
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004299899699097292,
      "loss": 3.1012,
      "theoretical_loss": 3.9229909965898715,
      "tokens_seen": 490471424
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042997993981945835,
      "loss": 3.0244,
      "theoretical_loss": 3.9229343544072517,
      "tokens_seen": 490536960
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004299699097291876,
      "loss": 2.8935,
      "theoretical_loss": 3.9228777219101088,
      "tokens_seen": 490602496
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042995987963891677,
      "loss": 3.0597,
      "theoretical_loss": 3.9228210990954935,
      "tokens_seen": 490668032
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042994984954864595,
      "loss": 3.1263,
      "theoretical_loss": 3.9227644859604562,
      "tokens_seen": 490733568
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042993981945837513,
      "loss": 3.0451,
      "theoretical_loss": 3.922707882502052,
      "tokens_seen": 490799104
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042992978936810437,
      "loss": 3.1519,
      "theoretical_loss": 3.922651288717333,
      "tokens_seen": 490864640
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004299197592778335,
      "loss": 2.9534,
      "theoretical_loss": 3.9225947046033562,
      "tokens_seen": 490930176
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042990972918756273,
      "loss": 3.0562,
      "theoretical_loss": 3.922538130157178,
      "tokens_seen": 490995712
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042989969909729185,
      "loss": 3.0875,
      "theoretical_loss": 3.922481565375856,
      "tokens_seen": 491061248
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004298896690070211,
      "loss": 2.9633,
      "theoretical_loss": 3.9224250102564504,
      "tokens_seen": 491126784
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042987963891675027,
      "loss": 3.133,
      "theoretical_loss": 3.922368464796021,
      "tokens_seen": 491192320
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042986960882647945,
      "loss": 3.1952,
      "theoretical_loss": 3.922311928991631,
      "tokens_seen": 491257856
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042985957873620863,
      "loss": 3.0355,
      "theoretical_loss": 3.9222554028403422,
      "tokens_seen": 491323392
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004298495486459378,
      "loss": 3.0389,
      "theoretical_loss": 3.9221988863392205,
      "tokens_seen": 491388928
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.000429839518555667,
      "loss": 3.2253,
      "theoretical_loss": 3.9221423794853303,
      "tokens_seen": 491454464
    },
    {
      "debugging/Self-BLEU-5": 0.31008426371372083,
      "debugging/distinct-1-grams": 0.7945037853782116,
      "debugging/distinct-2-grams": 0.9759782108419659,
      "debugging/entropy-1-grams": 5.529295001026804,
      "debugging/entropy-2-grams": 6.1656973428761805,
      "debugging/length": 452.7142857142857,
      "debugging/num_segments": 7,
      "debugging/raw_token_scores_avg": 0.007705323863774538,
      "debugging/raw_token_scores_std": 0.011359859257936478,
      "debugging/score": 0.007122610434625429,
      "debugging/score_std": 0.004618225724562532,
      "epoch": 0.15,
      "objective/train/advantage_avg": 0.4922936260700226,
      "objective/train/docs_used": 289197,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.547363758087158,
      "objective/train/original_loss": 3.547363519668579,
      "objective/train/theoretical_loss": 3.9220858822757396,
      "objective/train/tokens_used": 511980000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24248208105564117,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0504618883132935,
      "objective/train/weighted_lm_loss": 3.7262022495269775,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0396549701690674,
      "theoretical_loss": 3.9220858822757396,
      "tokens_seen": 491520000
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042982948846539623,
      "loss": 3.0514,
      "theoretical_loss": 3.9220858822757396,
      "tokens_seen": 491520000
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042981945837512536,
      "loss": 3.2312,
      "theoretical_loss": 3.9220293947075167,
      "tokens_seen": 491585536
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004298094282848546,
      "loss": 2.9753,
      "theoretical_loss": 3.9219729167777313,
      "tokens_seen": 491651072
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004297993981945837,
      "loss": 3.0668,
      "theoretical_loss": 3.9219164484834543,
      "tokens_seen": 491716608
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042978936810431296,
      "loss": 2.9973,
      "theoretical_loss": 3.9218599898217583,
      "tokens_seen": 491782144
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042977933801404214,
      "loss": 3.1284,
      "theoretical_loss": 3.921803540789716,
      "tokens_seen": 491847680
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004297693079237713,
      "loss": 3.2701,
      "theoretical_loss": 3.921747101384402,
      "tokens_seen": 491913216
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004297592778335005,
      "loss": 3.096,
      "theoretical_loss": 3.9216906716028936,
      "tokens_seen": 491978752
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004297592778335005,
      "loss": 3.4698,
      "theoretical_loss": 3.921634251442267,
      "tokens_seen": 492044288
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042974924774322973,
      "loss": 3.4987,
      "theoretical_loss": 3.9215778408996016,
      "tokens_seen": 492109824
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042973921765295886,
      "loss": 3.1385,
      "theoretical_loss": 3.9215214399719773,
      "tokens_seen": 492175360
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004297291875626881,
      "loss": 3.0464,
      "theoretical_loss": 3.9214650486564744,
      "tokens_seen": 492240896
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004297191574724172,
      "loss": 3.2265,
      "theoretical_loss": 3.9214086669501755,
      "tokens_seen": 492306432
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042970912738214646,
      "loss": 3.1607,
      "theoretical_loss": 3.921352294850166,
      "tokens_seen": 492371968
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042969909729187564,
      "loss": 3.0351,
      "theoretical_loss": 3.9212959323535284,
      "tokens_seen": 492437504
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004296890672016048,
      "loss": 3.2072,
      "theoretical_loss": 3.9212395794573505,
      "tokens_seen": 492503040
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.000429679037111334,
      "loss": 3.2537,
      "theoretical_loss": 3.9211832361587193,
      "tokens_seen": 492568576
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004296690070210632,
      "loss": 3.288,
      "theoretical_loss": 3.9211269024547244,
      "tokens_seen": 492634112
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042965897693079236,
      "loss": 3.2338,
      "theoretical_loss": 3.921070578342455,
      "tokens_seen": 492699648
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004296489468405216,
      "loss": 3.3061,
      "theoretical_loss": 3.9210142638190026,
      "tokens_seen": 492765184
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004296389167502507,
      "loss": 3.2624,
      "theoretical_loss": 3.92095795888146,
      "tokens_seen": 492830720
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042962888665997996,
      "loss": 3.0553,
      "theoretical_loss": 3.9209016635269207,
      "tokens_seen": 492896256
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004296188565697091,
      "loss": 3.3633,
      "theoretical_loss": 3.9208453777524808,
      "tokens_seen": 492961792
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004296088264794383,
      "loss": 3.306,
      "theoretical_loss": 3.9207891015552354,
      "tokens_seen": 493027328
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004295987963891675,
      "loss": 3.1271,
      "theoretical_loss": 3.920732834932283,
      "tokens_seen": 493092864
    },
    {
      "epoch": 0.15,
      "objective/train/advantage_avg": 0.49589046835899353,
      "objective/train/docs_used": 289918,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.221669912338257,
      "objective/train/original_loss": 3.221669912338257,
      "objective/train/theoretical_loss": 3.9206765778807227,
      "objective/train/tokens_used": 513618400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24595336616039276,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0508393049240112,
      "objective/train/weighted_lm_loss": 3.3854315280914307,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0442860126495361,
      "theoretical_loss": 3.9206765778807227,
      "tokens_seen": 493158400
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004295887662988967,
      "loss": 3.0381,
      "theoretical_loss": 3.9206765778807227,
      "tokens_seen": 493158400
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042957873620862587,
      "loss": 2.9746,
      "theoretical_loss": 3.920620330397654,
      "tokens_seen": 493223936
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004295687061183551,
      "loss": 3.0675,
      "theoretical_loss": 3.920564092480179,
      "tokens_seen": 493289472
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042955867602808423,
      "loss": 3.158,
      "theoretical_loss": 3.9205078641254003,
      "tokens_seen": 493355008
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042954864593781347,
      "loss": 3.0312,
      "theoretical_loss": 3.920451645330422,
      "tokens_seen": 493420544
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004295386158475426,
      "loss": 3.0334,
      "theoretical_loss": 3.920395436092349,
      "tokens_seen": 493486080
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042952858575727183,
      "loss": 3.2755,
      "theoretical_loss": 3.920339236408288,
      "tokens_seen": 493551616
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.000429518555667001,
      "loss": 3.126,
      "theoretical_loss": 3.920283046275346,
      "tokens_seen": 493617152
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004295085255767302,
      "loss": 3.1656,
      "theoretical_loss": 3.920226865690634,
      "tokens_seen": 493682688
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042949849548645937,
      "loss": 2.9406,
      "theoretical_loss": 3.9201706946512607,
      "tokens_seen": 493748224
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042948846539618855,
      "loss": 3.1786,
      "theoretical_loss": 3.920114533154338,
      "tokens_seen": 493813760
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042947843530591773,
      "loss": 3.0495,
      "theoretical_loss": 3.9200583811969785,
      "tokens_seen": 493879296
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042946840521564697,
      "loss": 3.3595,
      "theoretical_loss": 3.9200022387762967,
      "tokens_seen": 493944832
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004294583751253761,
      "loss": 3.0538,
      "theoretical_loss": 3.919946105889408,
      "tokens_seen": 494010368
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042944834503510533,
      "loss": 3.126,
      "theoretical_loss": 3.919889982533429,
      "tokens_seen": 494075904
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004294383149448345,
      "loss": 3.2459,
      "theoretical_loss": 3.919833868705477,
      "tokens_seen": 494141440
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004294282848545637,
      "loss": 2.9219,
      "theoretical_loss": 3.9197777644026717,
      "tokens_seen": 494206976
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042941825476429293,
      "loss": 3.3557,
      "theoretical_loss": 3.9197216696221324,
      "tokens_seen": 494272512
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042940822467402206,
      "loss": 3.2432,
      "theoretical_loss": 3.919665584360982,
      "tokens_seen": 494338048
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004293981945837513,
      "loss": 3.3128,
      "theoretical_loss": 3.9196095086163427,
      "tokens_seen": 494403584
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042938816449348047,
      "loss": 3.02,
      "theoretical_loss": 3.9195534423853386,
      "tokens_seen": 494469120
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042937813440320965,
      "loss": 3.135,
      "theoretical_loss": 3.919497385665095,
      "tokens_seen": 494534656
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042936810431293883,
      "loss": 3.0354,
      "theoretical_loss": 3.9194413384527387,
      "tokens_seen": 494600192
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.000429358074222668,
      "loss": 3.0193,
      "theoretical_loss": 3.9193853007453976,
      "tokens_seen": 494665728
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004293480441323972,
      "loss": 3.267,
      "theoretical_loss": 3.9193292725402014,
      "tokens_seen": 494731264
    },
    {
      "epoch": 0.15,
      "objective/train/advantage_avg": 0.49541202187538147,
      "objective/train/docs_used": 291263,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.143934488296509,
      "objective/train/original_loss": 3.143934488296509,
      "objective/train/theoretical_loss": 3.9192732538342785,
      "objective/train/tokens_used": 515256800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24548456072807312,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507891178131104,
      "objective/train/weighted_lm_loss": 3.3035683631896973,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0437887907028198,
      "theoretical_loss": 3.9192732538342785,
      "tokens_seen": 494796800
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042933801404212643,
      "loss": 3.1192,
      "theoretical_loss": 3.9192732538342785,
      "tokens_seen": 494796800
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042932798395185556,
      "loss": 3.2972,
      "theoretical_loss": 3.919217244624762,
      "tokens_seen": 494862336
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004293179538615848,
      "loss": 2.995,
      "theoretical_loss": 3.919161244908785,
      "tokens_seen": 494927872
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004293079237713139,
      "loss": 3.1508,
      "theoretical_loss": 3.9191052546834806,
      "tokens_seen": 494993408
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042929789368104316,
      "loss": 2.8719,
      "theoretical_loss": 3.9190492739459843,
      "tokens_seen": 495058944
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042928786359077234,
      "loss": 3.2228,
      "theoretical_loss": 3.9189933026934334,
      "tokens_seen": 495124480
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004292778335005015,
      "loss": 3.2502,
      "theoretical_loss": 3.918937340922965,
      "tokens_seen": 495190016
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004292678034102307,
      "loss": 3.2388,
      "theoretical_loss": 3.9188813886317186,
      "tokens_seen": 495255552
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042925777331995993,
      "loss": 3.2952,
      "theoretical_loss": 3.918825445816834,
      "tokens_seen": 495321088
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042924774322968906,
      "loss": 2.9575,
      "theoretical_loss": 3.918769512475453,
      "tokens_seen": 495386624
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004292377131394183,
      "loss": 3.226,
      "theoretical_loss": 3.918713588604718,
      "tokens_seen": 495452160
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004292276830491474,
      "loss": 3.295,
      "theoretical_loss": 3.918657674201774,
      "tokens_seen": 495517696
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042921765295887666,
      "loss": 3.1151,
      "theoretical_loss": 3.9186017692637654,
      "tokens_seen": 495583232
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042920762286860584,
      "loss": 3.2166,
      "theoretical_loss": 3.918545873787839,
      "tokens_seen": 495648768
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.000429197592778335,
      "loss": 2.926,
      "theoretical_loss": 3.9184899877711423,
      "tokens_seen": 495714304
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004291875626880642,
      "loss": 3.0615,
      "theoretical_loss": 3.918434111210825,
      "tokens_seen": 495779840
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004291775325977934,
      "loss": 3.1201,
      "theoretical_loss": 3.918378244104036,
      "tokens_seen": 495845376
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042916750250752256,
      "loss": 3.0104,
      "theoretical_loss": 3.9183223864479277,
      "tokens_seen": 495910912
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004291574724172518,
      "loss": 3.217,
      "theoretical_loss": 3.918266538239653,
      "tokens_seen": 495976448
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004291474423269809,
      "loss": 3.2757,
      "theoretical_loss": 3.9182106994763646,
      "tokens_seen": 496041984
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042913741223671016,
      "loss": 3.0151,
      "theoretical_loss": 3.918154870155219,
      "tokens_seen": 496107520
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004291273821464393,
      "loss": 3.1918,
      "theoretical_loss": 3.9180990502733715,
      "tokens_seen": 496173056
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004291173520561685,
      "loss": 3.063,
      "theoretical_loss": 3.91804323982798,
      "tokens_seen": 496238592
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004291073219658977,
      "loss": 3.287,
      "theoretical_loss": 3.917987438816204,
      "tokens_seen": 496304128
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004290972918756269,
      "loss": 3.3601,
      "theoretical_loss": 3.9179316472352026,
      "tokens_seen": 496369664
    },
    {
      "epoch": 0.15,
      "objective/train/advantage_avg": 0.4945076107978821,
      "objective/train/docs_used": 291574,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.281736373901367,
      "objective/train/original_loss": 3.281736373901367,
      "objective/train/theoretical_loss": 3.917875865082138,
      "objective/train/tokens_used": 516895200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24459809064865112,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506941080093384,
      "objective/train/weighted_lm_loss": 3.448192596435547,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0451053380966187,
      "theoretical_loss": 3.917875865082138,
      "tokens_seen": 496435200
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042908726178535607,
      "loss": 3.3572,
      "theoretical_loss": 3.917875865082138,
      "tokens_seen": 496435200
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004290772316950853,
      "loss": 3.1008,
      "theoretical_loss": 3.917820092354172,
      "tokens_seen": 496500736
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042906720160481443,
      "loss": 3.1412,
      "theoretical_loss": 3.9177643290484694,
      "tokens_seen": 496566272
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042905717151454367,
      "loss": 3.2313,
      "theoretical_loss": 3.9177085751621936,
      "tokens_seen": 496631808
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004290471414242728,
      "loss": 3.2083,
      "theoretical_loss": 3.9176528306925116,
      "tokens_seen": 496697344
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042903711133400203,
      "loss": 3.3772,
      "theoretical_loss": 3.917597095636591,
      "tokens_seen": 496762880
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004290270812437312,
      "loss": 2.9538,
      "theoretical_loss": 3.9175413699916004,
      "tokens_seen": 496828416
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004290170511534604,
      "loss": 3.0785,
      "theoretical_loss": 3.91748565375471,
      "tokens_seen": 496893952
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042900702106318957,
      "loss": 3.1413,
      "theoretical_loss": 3.91742994692309,
      "tokens_seen": 496959488
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042899699097291875,
      "loss": 2.9285,
      "theoretical_loss": 3.917374249493913,
      "tokens_seen": 497025024
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042898696088264793,
      "loss": 3.4339,
      "theoretical_loss": 3.9173185614643535,
      "tokens_seen": 497090560
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042897693079237717,
      "loss": 3.2153,
      "theoretical_loss": 3.9172628828315847,
      "tokens_seen": 497156096
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004289669007021063,
      "loss": 2.9731,
      "theoretical_loss": 3.9172072135927842,
      "tokens_seen": 497221632
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042895687061183553,
      "loss": 3.286,
      "theoretical_loss": 3.9171515537451285,
      "tokens_seen": 497287168
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004289468405215647,
      "loss": 3.0873,
      "theoretical_loss": 3.9170959032857953,
      "tokens_seen": 497352704
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004289368104312939,
      "loss": 3.4126,
      "theoretical_loss": 3.9170402622119656,
      "tokens_seen": 497418240
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004289267803410231,
      "loss": 3.3487,
      "theoretical_loss": 3.91698463052082,
      "tokens_seen": 497483776
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042891675025075226,
      "loss": 2.7877,
      "theoretical_loss": 3.9169290082095394,
      "tokens_seen": 497549312
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042890672016048144,
      "loss": 3.0985,
      "theoretical_loss": 3.9168733952753083,
      "tokens_seen": 497614848
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042889669007021067,
      "loss": 3.0787,
      "theoretical_loss": 3.916817791715311,
      "tokens_seen": 497680384
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004288866599799398,
      "loss": 2.9007,
      "theoretical_loss": 3.916762197526733,
      "tokens_seen": 497745920
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042887662988966903,
      "loss": 3.2208,
      "theoretical_loss": 3.916706612706762,
      "tokens_seen": 497811456
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042886659979939816,
      "loss": 3.4166,
      "theoretical_loss": 3.916651037252585,
      "tokens_seen": 497876992
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004288565697091274,
      "loss": 2.8781,
      "theoretical_loss": 3.9165954711613917,
      "tokens_seen": 497942528
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004288465396188566,
      "loss": 3.3559,
      "theoretical_loss": 3.9165399144303734,
      "tokens_seen": 498008064
    },
    {
      "epoch": 0.15,
      "objective/train/advantage_avg": 0.4942743182182312,
      "objective/train/docs_used": 293024,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.579573154449463,
      "objective/train/original_loss": 2.579573392868042,
      "objective/train/theoretical_loss": 3.9164843670567215,
      "objective/train/tokens_used": 518533600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24438542127609253,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.05066978931427,
      "objective/train/weighted_lm_loss": 2.7103893756866455,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0338950157165527,
      "theoretical_loss": 3.9164843670567215,
      "tokens_seen": 498073600
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042883650952858576,
      "loss": 2.8539,
      "theoretical_loss": 3.9164843670567215,
      "tokens_seen": 498073600
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042882647943831494,
      "loss": 3.1981,
      "theoretical_loss": 3.9164288290376286,
      "tokens_seen": 498139136
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004288164493480441,
      "loss": 3.0434,
      "theoretical_loss": 3.9163733003702896,
      "tokens_seen": 498204672
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004288064192577733,
      "loss": 2.8463,
      "theoretical_loss": 3.9163177810519,
      "tokens_seen": 498270208
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042879638916750254,
      "loss": 2.9799,
      "theoretical_loss": 3.916262271079656,
      "tokens_seen": 498335744
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042878635907723166,
      "loss": 3.0548,
      "theoretical_loss": 3.9162067704507555,
      "tokens_seen": 498401280
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004287763289869609,
      "loss": 3.0006,
      "theoretical_loss": 3.9161512791623974,
      "tokens_seen": 498466816
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004287662988966901,
      "loss": 3.2053,
      "theoretical_loss": 3.9160957972117827,
      "tokens_seen": 498532352
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042875626880641926,
      "loss": 3.1095,
      "theoretical_loss": 3.9160403245961124,
      "tokens_seen": 498597888
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042874623871614844,
      "loss": 3.0809,
      "theoretical_loss": 3.915984861312589,
      "tokens_seen": 498663424
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004287362086258776,
      "loss": 2.9572,
      "theoretical_loss": 3.9159294073584174,
      "tokens_seen": 498728960
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004287261785356068,
      "loss": 3.0831,
      "theoretical_loss": 3.9158739627308012,
      "tokens_seen": 498794496
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042871614844533604,
      "loss": 2.971,
      "theoretical_loss": 3.9158185274269477,
      "tokens_seen": 498860032
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042870611835506517,
      "loss": 3.4062,
      "theoretical_loss": 3.915763101444065,
      "tokens_seen": 498925568
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004286960882647944,
      "loss": 3.0148,
      "theoretical_loss": 3.9157076847793597,
      "tokens_seen": 498991104
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042868605817452353,
      "loss": 3.101,
      "theoretical_loss": 3.915652277430044,
      "tokens_seen": 499056640
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042867602808425276,
      "loss": 3.1042,
      "theoretical_loss": 3.9155968793933273,
      "tokens_seen": 499122176
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.000428665997993982,
      "loss": 3.2172,
      "theoretical_loss": 3.915541490666423,
      "tokens_seen": 499187712
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004286559679037111,
      "loss": 2.9653,
      "theoretical_loss": 3.915486111246545,
      "tokens_seen": 499253248
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042864593781344036,
      "loss": 2.9273,
      "theoretical_loss": 3.9154307411309066,
      "tokens_seen": 499318784
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004286359077231695,
      "loss": 3.0299,
      "theoretical_loss": 3.915375380316725,
      "tokens_seen": 499384320
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004286258776328987,
      "loss": 3.244,
      "theoretical_loss": 3.915320028801216,
      "tokens_seen": 499449856
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004286158475426279,
      "loss": 3.0382,
      "theoretical_loss": 3.915264686581599,
      "tokens_seen": 499515392
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004286058174523571,
      "loss": 3.0372,
      "theoretical_loss": 3.9152093536550936,
      "tokens_seen": 499580928
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042859578736208627,
      "loss": 3.0271,
      "theoretical_loss": 3.9151540300189196,
      "tokens_seen": 499646464
    },
    {
      "epoch": 0.15,
      "objective/train/advantage_avg": 0.4922420382499695,
      "objective/train/docs_used": 293514,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.983635902404785,
      "objective/train/original_loss": 2.983635663986206,
      "objective/train/theoretical_loss": 3.9150987156702994,
      "objective/train/tokens_used": 520172000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2423778623342514,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050456166267395,
      "objective/train/weighted_lm_loss": 3.1342062950134277,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0396549701690674,
      "theoretical_loss": 3.9150987156702994,
      "tokens_seen": 499712000
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004285857572718155,
      "loss": 3.0679,
      "theoretical_loss": 3.9150987156702994,
      "tokens_seen": 499712000
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042857572718154463,
      "loss": 3.1851,
      "theoretical_loss": 3.915043410606457,
      "tokens_seen": 499777536
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042856569709127387,
      "loss": 3.0549,
      "theoretical_loss": 3.914988114824615,
      "tokens_seen": 499843072
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.000428555667001003,
      "loss": 2.9514,
      "theoretical_loss": 3.914932828322,
      "tokens_seen": 499908608
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042854563691073223,
      "loss": 2.8041,
      "theoretical_loss": 3.914877551095838,
      "tokens_seen": 499974144
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004285356068204614,
      "loss": 3.0606,
      "theoretical_loss": 3.9148222831433577,
      "tokens_seen": 500039680
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004285255767301906,
      "loss": 3.179,
      "theoretical_loss": 3.914767024461788,
      "tokens_seen": 500105216
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042851554663991977,
      "loss": 3.3372,
      "theoretical_loss": 3.9147117750483584,
      "tokens_seen": 500170752
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042850551654964895,
      "loss": 3.2605,
      "theoretical_loss": 3.9146565349003017,
      "tokens_seen": 500236288
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042849548645937813,
      "loss": 3.2681,
      "theoretical_loss": 3.9146013040148486,
      "tokens_seen": 500301824
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042848545636910737,
      "loss": 3.2748,
      "theoretical_loss": 3.9145460823892346,
      "tokens_seen": 500367360
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004284754262788365,
      "loss": 3.2548,
      "theoretical_loss": 3.9144908700206944,
      "tokens_seen": 500432896
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042846539618856573,
      "loss": 3.191,
      "theoretical_loss": 3.9144356669064635,
      "tokens_seen": 500498432
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004284553660982949,
      "loss": 3.0463,
      "theoretical_loss": 3.9143804730437797,
      "tokens_seen": 500563968
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004284453360080241,
      "loss": 3.0559,
      "theoretical_loss": 3.9143252884298816,
      "tokens_seen": 500629504
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004284353059177533,
      "loss": 2.7662,
      "theoretical_loss": 3.914270113062009,
      "tokens_seen": 500695040
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042842527582748246,
      "loss": 3.1498,
      "theoretical_loss": 3.9142149469374026,
      "tokens_seen": 500760576
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042841524573721164,
      "loss": 3.1166,
      "theoretical_loss": 3.914159790053305,
      "tokens_seen": 500826112
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042840521564694087,
      "loss": 3.0013,
      "theoretical_loss": 3.9141046424069597,
      "tokens_seen": 500891648
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042839518555667,
      "loss": 3.0596,
      "theoretical_loss": 3.91404950399561,
      "tokens_seen": 500957184
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042838515546639923,
      "loss": 2.9514,
      "theoretical_loss": 3.913994374816503,
      "tokens_seen": 501022720
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042837512537612836,
      "loss": 3.3005,
      "theoretical_loss": 3.913939254866884,
      "tokens_seen": 501088256
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004283650952858576,
      "loss": 3.1889,
      "theoretical_loss": 3.9138841441440024,
      "tokens_seen": 501153792
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004283550651955868,
      "loss": 3.3556,
      "theoretical_loss": 3.913829042645107,
      "tokens_seen": 501219328
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042834503510531596,
      "loss": 3.0235,
      "theoretical_loss": 3.9137739503674482,
      "tokens_seen": 501284864
    },
    {
      "epoch": 0.15,
      "objective/train/advantage_avg": 0.4955374598503113,
      "objective/train/docs_used": 294722,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.2794601917266846,
      "objective/train/original_loss": 3.2794606685638428,
      "objective/train/theoretical_loss": 3.913718867308278,
      "objective/train/tokens_used": 521810400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24562107026576996,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0508023500442505,
      "objective/train/weighted_lm_loss": 3.4459545612335205,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.032329797744751,
      "theoretical_loss": 3.913718867308278,
      "tokens_seen": 501350400
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042833500501504514,
      "loss": 3.2422,
      "theoretical_loss": 3.913718867308278,
      "tokens_seen": 501350400
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004283249749247743,
      "loss": 2.9118,
      "theoretical_loss": 3.9136637934648477,
      "tokens_seen": 501415936
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004283149448345035,
      "loss": 3.263,
      "theoretical_loss": 3.9136087288344124,
      "tokens_seen": 501481472
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042830491474423274,
      "loss": 3.0198,
      "theoretical_loss": 3.9135536734142278,
      "tokens_seen": 501547008
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042829488465396186,
      "loss": 2.9408,
      "theoretical_loss": 3.913498627201549,
      "tokens_seen": 501612544
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004282848545636911,
      "loss": 3.0204,
      "theoretical_loss": 3.9134435901936335,
      "tokens_seen": 501678080
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004282748244734203,
      "loss": 3.1125,
      "theoretical_loss": 3.9133885623877407,
      "tokens_seen": 501743616
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042826479438314946,
      "loss": 2.9256,
      "theoretical_loss": 3.91333354378113,
      "tokens_seen": 501809152
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042825476429287864,
      "loss": 3.3554,
      "theoretical_loss": 3.9132785343710625,
      "tokens_seen": 501874688
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004282447342026078,
      "loss": 3.2883,
      "theoretical_loss": 3.9132235341548007,
      "tokens_seen": 501940224
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.000428234704112337,
      "loss": 3.1483,
      "theoretical_loss": 3.913168543129607,
      "tokens_seen": 502005760
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042822467402206624,
      "loss": 3.2737,
      "theoretical_loss": 3.9131135612927466,
      "tokens_seen": 502071296
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042821464393179537,
      "loss": 3.3591,
      "theoretical_loss": 3.913058588641485,
      "tokens_seen": 502136832
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004282046138415246,
      "loss": 3.2939,
      "theoretical_loss": 3.913003625173089,
      "tokens_seen": 502202368
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042819458375125373,
      "loss": 3.1195,
      "theoretical_loss": 3.912948670884827,
      "tokens_seen": 502267904
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042818455366098296,
      "loss": 3.3624,
      "theoretical_loss": 3.9128937257739675,
      "tokens_seen": 502333440
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042817452357071215,
      "loss": 3.1315,
      "theoretical_loss": 3.912838789837781,
      "tokens_seen": 502398976
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004281644934804413,
      "loss": 3.1055,
      "theoretical_loss": 3.91278386307354,
      "tokens_seen": 502464512
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004281544633901705,
      "loss": 3.1306,
      "theoretical_loss": 3.912728945478516,
      "tokens_seen": 502530048
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004281444332998997,
      "loss": 3.2406,
      "theoretical_loss": 3.9126740370499835,
      "tokens_seen": 502595584
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042813440320962887,
      "loss": 3.2089,
      "theoretical_loss": 3.912619137785217,
      "tokens_seen": 502661120
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004281243731193581,
      "loss": 3.1427,
      "theoretical_loss": 3.912564247681493,
      "tokens_seen": 502726656
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042811434302908723,
      "loss": 3.1436,
      "theoretical_loss": 3.912509366736089,
      "tokens_seen": 502792192
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042810431293881647,
      "loss": 3.0135,
      "theoretical_loss": 3.9124544949462834,
      "tokens_seen": 502857728
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042809428284854565,
      "loss": 3.1075,
      "theoretical_loss": 3.9123996323093557,
      "tokens_seen": 502923264
    },
    {
      "epoch": 0.15,
      "objective/train/advantage_avg": 0.49610692262649536,
      "objective/train/docs_used": 295112,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.2018768787384033,
      "objective/train/original_loss": 3.201876401901245,
      "objective/train/theoretical_loss": 3.912344778822587,
      "objective/train/tokens_used": 523448800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24618685245513916,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050862193107605,
      "objective/train/weighted_lm_loss": 3.364605665206909,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0396549701690674,
      "theoretical_loss": 3.912344778822587,
      "tokens_seen": 502988800
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042808425275827483,
      "loss": 3.059,
      "theoretical_loss": 3.912344778822587,
      "tokens_seen": 502988800
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.000428074222668004,
      "loss": 3.1105,
      "theoretical_loss": 3.912289934483259,
      "tokens_seen": 503054336
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004280641925777332,
      "loss": 2.9523,
      "theoretical_loss": 3.9122350992886554,
      "tokens_seen": 503119872
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004280541624874624,
      "loss": 3.1513,
      "theoretical_loss": 3.91218027323606,
      "tokens_seen": 503185408
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004280441323971916,
      "loss": 3.2973,
      "theoretical_loss": 3.9121254563227588,
      "tokens_seen": 503250944
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042803410230692074,
      "loss": 3.0634,
      "theoretical_loss": 3.912070648546038,
      "tokens_seen": 503316480
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042802407221664997,
      "loss": 2.9101,
      "theoretical_loss": 3.9120158499031854,
      "tokens_seen": 503382016
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004280140421263791,
      "loss": 3.1946,
      "theoretical_loss": 3.91196106039149,
      "tokens_seen": 503447552
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042800401203610833,
      "loss": 3.2672,
      "theoretical_loss": 3.9119062800082416,
      "tokens_seen": 503513088
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004279939819458375,
      "loss": 3.1869,
      "theoretical_loss": 3.9118515087507326,
      "tokens_seen": 503578624
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004279839518555667,
      "loss": 3.1615,
      "theoretical_loss": 3.911796746616254,
      "tokens_seen": 503644160
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004279739217652959,
      "loss": 3.0321,
      "theoretical_loss": 3.911741993602101,
      "tokens_seen": 503709696
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004279638916750251,
      "loss": 3.0522,
      "theoretical_loss": 3.911687249705567,
      "tokens_seen": 503775232
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042795386158475424,
      "loss": 3.0712,
      "theoretical_loss": 3.911632514923948,
      "tokens_seen": 503840768
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004279438314944835,
      "loss": 3.1044,
      "theoretical_loss": 3.911577789254542,
      "tokens_seen": 503906304
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004279338014042126,
      "loss": 3.1459,
      "theoretical_loss": 3.9115230726946457,
      "tokens_seen": 503971840
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042792377131394184,
      "loss": 2.9229,
      "theoretical_loss": 3.9114683652415603,
      "tokens_seen": 504037376
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042791374122367107,
      "loss": 2.9946,
      "theoretical_loss": 3.911413666892585,
      "tokens_seen": 504102912
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004279037111334002,
      "loss": 3.1146,
      "theoretical_loss": 3.9113589776450217,
      "tokens_seen": 504168448
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042789368104312943,
      "loss": 2.9874,
      "theoretical_loss": 3.9113042974961734,
      "tokens_seen": 504233984
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042788365095285856,
      "loss": 3.131,
      "theoretical_loss": 3.911249626443344,
      "tokens_seen": 504299520
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004278736208625878,
      "loss": 3.1798,
      "theoretical_loss": 3.9111949644838386,
      "tokens_seen": 504365056
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.000427863590772317,
      "loss": 2.895,
      "theoretical_loss": 3.9111403116149637,
      "tokens_seen": 504430592
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042785356068204616,
      "loss": 3.3931,
      "theoretical_loss": 3.9110856678340253,
      "tokens_seen": 504496128
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042784353059177534,
      "loss": 3.0713,
      "theoretical_loss": 3.9110310331383342,
      "tokens_seen": 504561664
    },
    {
      "epoch": 0.15,
      "objective/train/advantage_avg": 0.49453234672546387,
      "objective/train/docs_used": 296352,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.300678253173828,
      "objective/train/original_loss": 3.300678014755249,
      "objective/train/theoretical_loss": 3.910976407525199,
      "objective/train/tokens_used": 525087200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2446424961090088,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506967306137085,
      "objective/train/weighted_lm_loss": 3.4677445888519287,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 3.910976407525199,
      "tokens_seen": 504627200
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004278335005015045,
      "loss": 3.0847,
      "theoretical_loss": 3.910976407525199,
      "tokens_seen": 504627200
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004278234704112337,
      "loss": 3.3497,
      "theoretical_loss": 3.9109217909919303,
      "tokens_seen": 504692736
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042781344032096294,
      "loss": 3.1578,
      "theoretical_loss": 3.9108671835358404,
      "tokens_seen": 504758272
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042780341023069206,
      "loss": 3.0226,
      "theoretical_loss": 3.910812585154242,
      "tokens_seen": 504823808
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004277933801404213,
      "loss": 3.2046,
      "theoretical_loss": 3.91075799584445,
      "tokens_seen": 504889344
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004277833500501505,
      "loss": 3.184,
      "theoretical_loss": 3.9107034156037797,
      "tokens_seen": 504954880
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042777331995987966,
      "loss": 3.1815,
      "theoretical_loss": 3.9106488444295473,
      "tokens_seen": 505020416
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042776328986960884,
      "loss": 3.1685,
      "theoretical_loss": 3.9105942823190705,
      "tokens_seen": 505085952
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.000427753259779338,
      "loss": 3.0723,
      "theoretical_loss": 3.9105397292696686,
      "tokens_seen": 505151488
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004277432296890672,
      "loss": 2.9815,
      "theoretical_loss": 3.9104851852786617,
      "tokens_seen": 505217024
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042773319959879644,
      "loss": 2.9293,
      "theoretical_loss": 3.91043065034337,
      "tokens_seen": 505282560
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042772316950852557,
      "loss": 3.138,
      "theoretical_loss": 3.9103761244611164,
      "tokens_seen": 505348096
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004277131394182548,
      "loss": 3.0897,
      "theoretical_loss": 3.910321607629225,
      "tokens_seen": 505413632
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042770310932798393,
      "loss": 2.9952,
      "theoretical_loss": 3.9102670998450186,
      "tokens_seen": 505479168
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042769307923771316,
      "loss": 2.9723,
      "theoretical_loss": 3.9102126011058247,
      "tokens_seen": 505544704
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042768304914744235,
      "loss": 3.1233,
      "theoretical_loss": 3.910158111408969,
      "tokens_seen": 505610240
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042767301905717153,
      "loss": 3.1334,
      "theoretical_loss": 3.91010363075178,
      "tokens_seen": 505675776
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004276629889669007,
      "loss": 2.9281,
      "theoretical_loss": 3.910049159131586,
      "tokens_seen": 505741312
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004276529588766299,
      "loss": 3.1295,
      "theoretical_loss": 3.909994696545718,
      "tokens_seen": 505806848
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042764292878635907,
      "loss": 2.8293,
      "theoretical_loss": 3.9099402429915076,
      "tokens_seen": 505872384
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004276328986960883,
      "loss": 3.2094,
      "theoretical_loss": 3.909885798466286,
      "tokens_seen": 505937920
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042762286860581743,
      "loss": 3.243,
      "theoretical_loss": 3.909831362967388,
      "tokens_seen": 506003456
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042761283851554667,
      "loss": 3.3623,
      "theoretical_loss": 3.9097769364921486,
      "tokens_seen": 506068992
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042760280842527585,
      "loss": 3.1648,
      "theoretical_loss": 3.909722519037903,
      "tokens_seen": 506134528
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042759277833500503,
      "loss": 3.1646,
      "theoretical_loss": 3.909668110601988,
      "tokens_seen": 506200064
    },
    {
      "epoch": 0.15,
      "objective/train/advantage_avg": 0.49482613801956177,
      "objective/train/docs_used": 297118,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.149805784225464,
      "objective/train/original_loss": 3.149805784225464,
      "objective/train/theoretical_loss": 3.9096137111817413,
      "objective/train/tokens_used": 526725600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2449306845664978,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507278442382812,
      "objective/train/weighted_lm_loss": 3.3093981742858887,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0281672477722168,
      "theoretical_loss": 3.9096137111817413,
      "tokens_seen": 506265600
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004275827482447342,
      "loss": 3.1921,
      "theoretical_loss": 3.9096137111817413,
      "tokens_seen": 506265600
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004275727181544634,
      "loss": 3.1828,
      "theoretical_loss": 3.9095593207745045,
      "tokens_seen": 506331136
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004275626880641926,
      "loss": 3.0754,
      "theoretical_loss": 3.9095049393776153,
      "tokens_seen": 506396672
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004275526579739218,
      "loss": 3.1343,
      "theoretical_loss": 3.9094505669884168,
      "tokens_seen": 506462208
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042754262788365094,
      "loss": 3.0498,
      "theoretical_loss": 3.9093962036042518,
      "tokens_seen": 506527744
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042753259779338017,
      "loss": 3.0138,
      "theoretical_loss": 3.909341849222463,
      "tokens_seen": 506593280
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004275225677031093,
      "loss": 3.2737,
      "theoretical_loss": 3.909287503840396,
      "tokens_seen": 506658816
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042751253761283853,
      "loss": 3.2094,
      "theoretical_loss": 3.9092331674553975,
      "tokens_seen": 506724352
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004275025075225677,
      "loss": 3.3101,
      "theoretical_loss": 3.909178840064813,
      "tokens_seen": 506789888
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004274924774322969,
      "loss": 2.9875,
      "theoretical_loss": 3.9091245216659924,
      "tokens_seen": 506855424
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004274824473420261,
      "loss": 3.2263,
      "theoretical_loss": 3.9090702122562844,
      "tokens_seen": 506920960
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004274724172517553,
      "loss": 2.8084,
      "theoretical_loss": 3.9090159118330394,
      "tokens_seen": 506986496
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042746238716148444,
      "loss": 3.1168,
      "theoretical_loss": 3.90896162039361,
      "tokens_seen": 507052032
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004274523570712137,
      "loss": 3.0737,
      "theoretical_loss": 3.9089073379353474,
      "tokens_seen": 507117568
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004274423269809428,
      "loss": 3.2805,
      "theoretical_loss": 3.9088530644556068,
      "tokens_seen": 507183104
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042743229689067204,
      "loss": 2.9489,
      "theoretical_loss": 3.908798799951743,
      "tokens_seen": 507248640
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004274222668004012,
      "loss": 3.2329,
      "theoretical_loss": 3.908744544421112,
      "tokens_seen": 507314176
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004274122367101304,
      "loss": 3.0453,
      "theoretical_loss": 3.908690297861071,
      "tokens_seen": 507379712
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004274022066198596,
      "loss": 3.0495,
      "theoretical_loss": 3.9086360602689787,
      "tokens_seen": 507445248
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042739217652958876,
      "loss": 3.042,
      "theoretical_loss": 3.9085818316421945,
      "tokens_seen": 507510784
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042738214643931794,
      "loss": 3.0925,
      "theoretical_loss": 3.9085276119780787,
      "tokens_seen": 507576320
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004273721163490472,
      "loss": 2.9074,
      "theoretical_loss": 3.9084734012739935,
      "tokens_seen": 507641856
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004273620862587763,
      "loss": 3.0669,
      "theoretical_loss": 3.9084191995273017,
      "tokens_seen": 507707392
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042735205616850554,
      "loss": 3.1392,
      "theoretical_loss": 3.9083650067353677,
      "tokens_seen": 507772928
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042734202607823467,
      "loss": 3.1016,
      "theoretical_loss": 3.9083108228955554,
      "tokens_seen": 507838464
    },
    {
      "epoch": 0.15,
      "objective/train/advantage_avg": 0.4955235421657562,
      "objective/train/docs_used": 298304,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.394134044647217,
      "objective/train/original_loss": 3.394134044647217,
      "objective/train/theoretical_loss": 3.9082566480052314,
      "objective/train/tokens_used": 528364000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24560192227363586,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0508008003234863,
      "objective/train/weighted_lm_loss": 3.566347599029541,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0432155132293701,
      "theoretical_loss": 3.9082566480052314,
      "tokens_seen": 507904000
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004273319959879639,
      "loss": 3.3861,
      "theoretical_loss": 3.9082566480052314,
      "tokens_seen": 507904000
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004273219658976931,
      "loss": 3.3142,
      "theoretical_loss": 3.908202482061764,
      "tokens_seen": 507969536
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042731193580742226,
      "loss": 3.0448,
      "theoretical_loss": 3.9081483250625206,
      "tokens_seen": 508035072
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042730190571715145,
      "loss": 2.6815,
      "theoretical_loss": 3.908094177004872,
      "tokens_seen": 508100608
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004272918756268807,
      "loss": 2.9583,
      "theoretical_loss": 3.9080400378861873,
      "tokens_seen": 508166144
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004272818455366098,
      "loss": 3.1719,
      "theoretical_loss": 3.9079859077038392,
      "tokens_seen": 508231680
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042727181544633904,
      "loss": 3.0724,
      "theoretical_loss": 3.907931786455201,
      "tokens_seen": 508297216
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042726178535606817,
      "loss": 3.4514,
      "theoretical_loss": 3.9078776741376453,
      "tokens_seen": 508362752
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004272517552657974,
      "loss": 2.8428,
      "theoretical_loss": 3.9078235707485485,
      "tokens_seen": 508428288
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004272417251755266,
      "loss": 3.0545,
      "theoretical_loss": 3.907769476285287,
      "tokens_seen": 508493824
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042723169508525577,
      "loss": 2.9006,
      "theoretical_loss": 3.9077153907452367,
      "tokens_seen": 508559360
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042722166499498495,
      "loss": 3.2094,
      "theoretical_loss": 3.907661314125777,
      "tokens_seen": 508624896
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042721163490471413,
      "loss": 3.1307,
      "theoretical_loss": 3.9076072464242877,
      "tokens_seen": 508690432
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004272016048144433,
      "loss": 3.1283,
      "theoretical_loss": 3.9075531876381486,
      "tokens_seen": 508755968
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042719157472417255,
      "loss": 3.1782,
      "theoretical_loss": 3.9074991377647423,
      "tokens_seen": 508821504
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004271815446339017,
      "loss": 3.015,
      "theoretical_loss": 3.9074450968014514,
      "tokens_seen": 508887040
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004271715145436309,
      "loss": 3.0112,
      "theoretical_loss": 3.90739106474566,
      "tokens_seen": 508952576
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004271614844533601,
      "loss": 3.4413,
      "theoretical_loss": 3.9073370415947526,
      "tokens_seen": 509018112
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042715145436308927,
      "loss": 3.2005,
      "theoretical_loss": 3.907283027346116,
      "tokens_seen": 509083648
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004271414242728185,
      "loss": 3.1848,
      "theoretical_loss": 3.9072290219971375,
      "tokens_seen": 509149184
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042713139418254763,
      "loss": 3.1521,
      "theoretical_loss": 3.907175025545205,
      "tokens_seen": 509214720
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042712136409227687,
      "loss": 3.1307,
      "theoretical_loss": 3.907121037987708,
      "tokens_seen": 509280256
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042711133400200605,
      "loss": 3.4145,
      "theoretical_loss": 3.907067059322038,
      "tokens_seen": 509345792
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042710130391173523,
      "loss": 2.9944,
      "theoretical_loss": 3.907013089545586,
      "tokens_seen": 509411328
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004270912738214644,
      "loss": 3.0202,
      "theoretical_loss": 3.906959128655745,
      "tokens_seen": 509476864
    },
    {
      "epoch": 0.15,
      "objective/train/advantage_avg": 0.49561506509780884,
      "objective/train/docs_used": 298986,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.1949210166931152,
      "objective/train/original_loss": 3.1949210166931152,
      "objective/train/theoretical_loss": 3.906905176649909,
      "objective/train/tokens_used": 530002400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2456878274679184,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.05081045627594,
      "objective/train/weighted_lm_loss": 3.357154130935669,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0417574644088745,
      "theoretical_loss": 3.906905176649909,
      "tokens_seen": 509542400
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004270812437311936,
      "loss": 3.3646,
      "theoretical_loss": 3.906905176649909,
      "tokens_seen": 509542400
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004270712136409228,
      "loss": 3.2514,
      "theoretical_loss": 3.9068512335254724,
      "tokens_seen": 509607936
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.000427061183550652,
      "loss": 3.2771,
      "theoretical_loss": 3.906797299279831,
      "tokens_seen": 509673472
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042705115346038114,
      "loss": 3.1244,
      "theoretical_loss": 3.906743373910384,
      "tokens_seen": 509739008
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042704112337011037,
      "loss": 3.3407,
      "theoretical_loss": 3.9066894574145277,
      "tokens_seen": 509804544
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004270310932798395,
      "loss": 3.3402,
      "theoretical_loss": 3.9066355497896623,
      "tokens_seen": 509870080
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042702106318956873,
      "loss": 3.2257,
      "theoretical_loss": 3.9065816510331883,
      "tokens_seen": 509935616
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004270110330992979,
      "loss": 2.9514,
      "theoretical_loss": 3.9065277611425073,
      "tokens_seen": 510001152
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004270010030090271,
      "loss": 3.1483,
      "theoretical_loss": 3.906473880115022,
      "tokens_seen": 510066688
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004269909729187563,
      "loss": 3.035,
      "theoretical_loss": 3.9064200079481357,
      "tokens_seen": 510132224
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004269809428284855,
      "loss": 3.0266,
      "theoretical_loss": 3.9063661446392537,
      "tokens_seen": 510197760
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042697091273821464,
      "loss": 3.0292,
      "theoretical_loss": 3.9063122901857814,
      "tokens_seen": 510263296
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004269608826479439,
      "loss": 3.0724,
      "theoretical_loss": 3.9062584445851267,
      "tokens_seen": 510328832
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.000426950852557673,
      "loss": 3.3058,
      "theoretical_loss": 3.9062046078346975,
      "tokens_seen": 510394368
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042694082246740224,
      "loss": 2.9942,
      "theoretical_loss": 3.906150779931903,
      "tokens_seen": 510459904
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004269307923771314,
      "loss": 2.933,
      "theoretical_loss": 3.906096960874153,
      "tokens_seen": 510525440
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004269207622868606,
      "loss": 3.0001,
      "theoretical_loss": 3.9060431506588595,
      "tokens_seen": 510590976
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004269107321965898,
      "loss": 3.3545,
      "theoretical_loss": 3.905989349283435,
      "tokens_seen": 510656512
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042690070210631896,
      "loss": 3.0382,
      "theoretical_loss": 3.905935556745293,
      "tokens_seen": 510722048
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042689067201604814,
      "loss": 3.0201,
      "theoretical_loss": 3.9058817730418474,
      "tokens_seen": 510787584
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004268806419257774,
      "loss": 2.9918,
      "theoretical_loss": 3.9058279981705155,
      "tokens_seen": 510853120
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004268706118355065,
      "loss": 3.114,
      "theoretical_loss": 3.9057742321287128,
      "tokens_seen": 510918656
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042686058174523574,
      "loss": 2.9465,
      "theoretical_loss": 3.905720474913858,
      "tokens_seen": 510984192
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042685055165496487,
      "loss": 2.8613,
      "theoretical_loss": 3.9056667265233704,
      "tokens_seen": 511049728
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004268405215646941,
      "loss": 2.9775,
      "theoretical_loss": 3.9056129869546696,
      "tokens_seen": 511115264
    },
    {
      "epoch": 0.15,
      "objective/train/advantage_avg": 0.49429699778556824,
      "objective/train/docs_used": 299996,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.676845073699951,
      "objective/train/original_loss": 2.676844835281372,
      "objective/train/theoretical_loss": 3.9055592562051764,
      "objective/train/tokens_used": 531640800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2444119155406952,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506720542907715,
      "objective/train/weighted_lm_loss": 2.8120322227478027,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0363596677780151,
      "theoretical_loss": 3.9055592562051764,
      "tokens_seen": 511180800
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004268304914744233,
      "loss": 3.0807,
      "theoretical_loss": 3.9055592562051764,
      "tokens_seen": 511180800
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042682046138415246,
      "loss": 3.0959,
      "theoretical_loss": 3.9055055342723137,
      "tokens_seen": 511246336
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042681043129388165,
      "loss": 3.0101,
      "theoretical_loss": 3.905451821153505,
      "tokens_seen": 511311872
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.0004268004012036109,
      "loss": 3.2446,
      "theoretical_loss": 3.9053981168461744,
      "tokens_seen": 511377408
    },
    {
      "epoch": 0.15,
      "learning_rate": 0.00042679037111334,
      "loss": 2.963,
      "theoretical_loss": 3.9053444213477473,
      "tokens_seen": 511442944
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042678034102306924,
      "loss": 2.9497,
      "theoretical_loss": 3.905290734655651,
      "tokens_seen": 511508480
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042677031093279837,
      "loss": 3.2249,
      "theoretical_loss": 3.9052370567673123,
      "tokens_seen": 511574016
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004267602808425276,
      "loss": 3.0387,
      "theoretical_loss": 3.905183387680161,
      "tokens_seen": 511639552
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004267502507522568,
      "loss": 3.1983,
      "theoretical_loss": 3.9051297273916257,
      "tokens_seen": 511705088
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042674022066198597,
      "loss": 3.0233,
      "theoretical_loss": 3.9050760758991387,
      "tokens_seen": 511770624
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042673019057171515,
      "loss": 2.9726,
      "theoretical_loss": 3.9050224332001306,
      "tokens_seen": 511836160
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042672016048144433,
      "loss": 3.2299,
      "theoretical_loss": 3.904968799292036,
      "tokens_seen": 511901696
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004267101303911735,
      "loss": 3.1332,
      "theoretical_loss": 3.904915174172288,
      "tokens_seen": 511967232
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042670010030090275,
      "loss": 3.0295,
      "theoretical_loss": 3.9048615578383226,
      "tokens_seen": 512032768
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004266900702106319,
      "loss": 3.0274,
      "theoretical_loss": 3.9048079502875748,
      "tokens_seen": 512098304
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004266800401203611,
      "loss": 3.0305,
      "theoretical_loss": 3.9047543515174836,
      "tokens_seen": 512163840
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042667001003009024,
      "loss": 3.064,
      "theoretical_loss": 3.904700761525487,
      "tokens_seen": 512229376
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042665997993981947,
      "loss": 3.1744,
      "theoretical_loss": 3.904647180309024,
      "tokens_seen": 512294912
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042664994984954865,
      "loss": 3.2147,
      "theoretical_loss": 3.904593607865536,
      "tokens_seen": 512360448
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042663991975927783,
      "loss": 2.958,
      "theoretical_loss": 3.904540044192464,
      "tokens_seen": 512425984
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.000426629889669007,
      "loss": 3.0682,
      "theoretical_loss": 3.904486489287251,
      "tokens_seen": 512491520
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042661985957873625,
      "loss": 3.1429,
      "theoretical_loss": 3.9044329431473406,
      "tokens_seen": 512557056
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004266098294884654,
      "loss": 2.791,
      "theoretical_loss": 3.904379405770179,
      "tokens_seen": 512622592
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004265997993981946,
      "loss": 3.0259,
      "theoretical_loss": 3.9043258771532106,
      "tokens_seen": 512688128
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042658976930792374,
      "loss": 2.8903,
      "theoretical_loss": 3.9042723572938836,
      "tokens_seen": 512753664
    },
    {
      "epoch": 0.16,
      "objective/train/advantage_avg": 0.49403390288352966,
      "objective/train/docs_used": 300643,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.692748785018921,
      "objective/train/original_loss": 2.692748546600342,
      "objective/train/theoretical_loss": 3.904218846189645,
      "objective/train/tokens_used": 533279200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2441299855709076,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506443977355957,
      "objective/train/weighted_lm_loss": 2.828925132751465,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0396549701690674,
      "theoretical_loss": 3.904218846189645,
      "tokens_seen": 512819200
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.000426579739217653,
      "loss": 2.9403,
      "theoretical_loss": 3.904218846189645,
      "tokens_seen": 512819200
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042656970912738216,
      "loss": 3.1092,
      "theoretical_loss": 3.9041653438379456,
      "tokens_seen": 512884736
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042655967903711134,
      "loss": 3.0468,
      "theoretical_loss": 3.9041118502362337,
      "tokens_seen": 512950272
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004265496489468405,
      "loss": 3.245,
      "theoretical_loss": 3.904058365381962,
      "tokens_seen": 513015808
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004265396188565697,
      "loss": 3.0515,
      "theoretical_loss": 3.904004889272583,
      "tokens_seen": 513081344
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004265295887662989,
      "loss": 3.0708,
      "theoretical_loss": 3.90395142190555,
      "tokens_seen": 513146880
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004265195586760281,
      "loss": 3.4846,
      "theoretical_loss": 3.9038979632783164,
      "tokens_seen": 513212416
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042650952858575724,
      "loss": 3.1641,
      "theoretical_loss": 3.9038445133883393,
      "tokens_seen": 513277952
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004264994984954865,
      "loss": 3.1364,
      "theoretical_loss": 3.9037910722330746,
      "tokens_seen": 513343488
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004264894684052156,
      "loss": 3.1198,
      "theoretical_loss": 3.9037376398099806,
      "tokens_seen": 513409024
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042647943831494484,
      "loss": 3.1831,
      "theoretical_loss": 3.903684216116516,
      "tokens_seen": 513474560
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.000426469408224674,
      "loss": 2.8906,
      "theoretical_loss": 3.9036308011501397,
      "tokens_seen": 513540096
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004264593781344032,
      "loss": 2.907,
      "theoretical_loss": 3.9035773949083135,
      "tokens_seen": 513605632
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004264493480441324,
      "loss": 3.2247,
      "theoretical_loss": 3.9035239973884988,
      "tokens_seen": 513671168
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004264393179538616,
      "loss": 3.0135,
      "theoretical_loss": 3.90347060858816,
      "tokens_seen": 513736704
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042642928786359075,
      "loss": 3.1523,
      "theoretical_loss": 3.9034172285047597,
      "tokens_seen": 513802240
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042641925777332,
      "loss": 3.1504,
      "theoretical_loss": 3.903363857135764,
      "tokens_seen": 513867776
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042640922768304916,
      "loss": 2.8387,
      "theoretical_loss": 3.9033104944786388,
      "tokens_seen": 513933312
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042639919759277834,
      "loss": 3.1284,
      "theoretical_loss": 3.903257140530852,
      "tokens_seen": 513998848
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004263891675025076,
      "loss": 3.048,
      "theoretical_loss": 3.90320379528987,
      "tokens_seen": 514064384
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004263791374122367,
      "loss": 3.3321,
      "theoretical_loss": 3.9031504587531645,
      "tokens_seen": 514129920
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042636910732196594,
      "loss": 3.0718,
      "theoretical_loss": 3.903097130918205,
      "tokens_seen": 514195456
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042635907723169507,
      "loss": 2.7574,
      "theoretical_loss": 3.9030438117824633,
      "tokens_seen": 514260992
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004263490471414243,
      "loss": 2.814,
      "theoretical_loss": 3.9029905013434116,
      "tokens_seen": 514326528
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004263390170511535,
      "loss": 3.3213,
      "theoretical_loss": 3.9029371995985236,
      "tokens_seen": 514392064
    },
    {
      "epoch": 0.16,
      "objective/train/advantage_avg": 0.4936845600605011,
      "objective/train/docs_used": 301818,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.521618604660034,
      "objective/train/original_loss": 3.521618366241455,
      "objective/train/theoretical_loss": 3.9028839065452745,
      "objective/train/tokens_used": 534917600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2438356876373291,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506080389022827,
      "objective/train/weighted_lm_loss": 3.699831962585449,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0338950157165527,
      "theoretical_loss": 3.9028839065452745,
      "tokens_seen": 514457600
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042632898696088266,
      "loss": 3.2687,
      "theoretical_loss": 3.9028839065452745,
      "tokens_seen": 514457600
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042631895687061185,
      "loss": 3.1153,
      "theoretical_loss": 3.9028306221811393,
      "tokens_seen": 514523136
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004263089267803411,
      "loss": 3.1912,
      "theoretical_loss": 3.9027773465035955,
      "tokens_seen": 514588672
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004262988966900702,
      "loss": 3.1893,
      "theoretical_loss": 3.9027240795101203,
      "tokens_seen": 514654208
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042628886659979944,
      "loss": 3.0952,
      "theoretical_loss": 3.9026708211981935,
      "tokens_seen": 514719744
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042627883650952857,
      "loss": 3.2016,
      "theoretical_loss": 3.9026175715652944,
      "tokens_seen": 514785280
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004262688064192578,
      "loss": 3.2821,
      "theoretical_loss": 3.902564330608904,
      "tokens_seen": 514850816
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.000426258776328987,
      "loss": 3.1155,
      "theoretical_loss": 3.902511098326505,
      "tokens_seen": 514916352
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042624874623871617,
      "loss": 3.3244,
      "theoretical_loss": 3.90245787471558,
      "tokens_seen": 514981888
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042623871614844535,
      "loss": 2.864,
      "theoretical_loss": 3.9024046597736124,
      "tokens_seen": 515047424
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042622868605817453,
      "loss": 3.1944,
      "theoretical_loss": 3.902351453498089,
      "tokens_seen": 515112960
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004262186559679037,
      "loss": 2.9303,
      "theoretical_loss": 3.902298255886495,
      "tokens_seen": 515178496
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042620862587763295,
      "loss": 3.0655,
      "theoretical_loss": 3.902245066936319,
      "tokens_seen": 515244032
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004261985957873621,
      "loss": 2.9816,
      "theoretical_loss": 3.9021918866450473,
      "tokens_seen": 515309568
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004261885656970913,
      "loss": 2.9343,
      "theoretical_loss": 3.902138715010171,
      "tokens_seen": 515375104
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042617853560682044,
      "loss": 3.0883,
      "theoretical_loss": 3.90208555202918,
      "tokens_seen": 515440640
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042616850551654967,
      "loss": 2.9099,
      "theoretical_loss": 3.9020323976995663,
      "tokens_seen": 515506176
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042615847542627885,
      "loss": 3.1053,
      "theoretical_loss": 3.9019792520188217,
      "tokens_seen": 515571712
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042614844533600803,
      "loss": 3.0399,
      "theoretical_loss": 3.9019261149844393,
      "tokens_seen": 515637248
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004261384152457372,
      "loss": 3.0879,
      "theoretical_loss": 3.901872986593916,
      "tokens_seen": 515702784
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042612838515546645,
      "loss": 3.1928,
      "theoretical_loss": 3.901819866844745,
      "tokens_seen": 515768320
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004261183550651956,
      "loss": 3.0419,
      "theoretical_loss": 3.9017667557344247,
      "tokens_seen": 515833856
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004261083249749248,
      "loss": 3.1179,
      "theoretical_loss": 3.901713653260452,
      "tokens_seen": 515899392
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042609829488465394,
      "loss": 2.9838,
      "theoretical_loss": 3.9016605594203257,
      "tokens_seen": 515964928
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004260882647943832,
      "loss": 3.2383,
      "theoretical_loss": 3.9016074742115463,
      "tokens_seen": 516030464
    },
    {
      "epoch": 0.16,
      "objective/train/advantage_avg": 0.49412497878074646,
      "objective/train/docs_used": 302480,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.3089895248413086,
      "objective/train/original_loss": 3.3089892864227295,
      "objective/train/theoretical_loss": 3.901554397631614,
      "objective/train/tokens_used": 536556000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2442339062690735,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506540536880493,
      "objective/train/weighted_lm_loss": 3.4765360355377197,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 3.901554397631614,
      "tokens_seen": 516096000
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042607823470411236,
      "loss": 3.3801,
      "theoretical_loss": 3.901554397631614,
      "tokens_seen": 516096000
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042606820461384154,
      "loss": 2.8523,
      "theoretical_loss": 3.9015013296780316,
      "tokens_seen": 516161536
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004260581745235707,
      "loss": 3.1233,
      "theoretical_loss": 3.9014482703483013,
      "tokens_seen": 516227072
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004260481444332999,
      "loss": 3.056,
      "theoretical_loss": 3.9013952196399275,
      "tokens_seen": 516292608
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004260381143430291,
      "loss": 2.9707,
      "theoretical_loss": 3.9013421775504153,
      "tokens_seen": 516358144
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004260280842527583,
      "loss": 3.0211,
      "theoretical_loss": 3.901289144077271,
      "tokens_seen": 516423680
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042601805416248744,
      "loss": 2.7048,
      "theoretical_loss": 3.9012361192180007,
      "tokens_seen": 516489216
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004260080240722167,
      "loss": 3.0313,
      "theoretical_loss": 3.901183102970114,
      "tokens_seen": 516554752
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004259979939819458,
      "loss": 2.8848,
      "theoretical_loss": 3.9011300953311188,
      "tokens_seen": 516620288
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042598796389167504,
      "loss": 2.9186,
      "theoretical_loss": 3.9010770962985264,
      "tokens_seen": 516685824
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004259779338014042,
      "loss": 3.139,
      "theoretical_loss": 3.9010241058698476,
      "tokens_seen": 516751360
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004259679037111334,
      "loss": 3.1103,
      "theoretical_loss": 3.9009711240425946,
      "tokens_seen": 516816896
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004259578736208626,
      "loss": 2.9961,
      "theoretical_loss": 3.900918150814281,
      "tokens_seen": 516882432
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004259478435305918,
      "loss": 2.8974,
      "theoretical_loss": 3.900865186182421,
      "tokens_seen": 516947968
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042593781344032095,
      "loss": 3.1514,
      "theoretical_loss": 3.9008122301445307,
      "tokens_seen": 517013504
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004259277833500502,
      "loss": 2.8687,
      "theoretical_loss": 3.9007592826981257,
      "tokens_seen": 517079040
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004259177532597793,
      "loss": 3.1472,
      "theoretical_loss": 3.9007063438407235,
      "tokens_seen": 517144576
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042590772316950854,
      "loss": 2.8773,
      "theoretical_loss": 3.9006534135698434,
      "tokens_seen": 517210112
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004258976930792377,
      "loss": 3.0121,
      "theoretical_loss": 3.9006004918830044,
      "tokens_seen": 517275648
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004258876629889669,
      "loss": 2.9647,
      "theoretical_loss": 3.900547578777727,
      "tokens_seen": 517341184
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004258776328986961,
      "loss": 3.0807,
      "theoretical_loss": 3.900494674251533,
      "tokens_seen": 517406720
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042586760280842527,
      "loss": 3.0721,
      "theoretical_loss": 3.9004417783019454,
      "tokens_seen": 517472256
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042585757271815445,
      "loss": 3.2291,
      "theoretical_loss": 3.900388890926487,
      "tokens_seen": 517537792
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004258475426278837,
      "loss": 3.0822,
      "theoretical_loss": 3.9003360121226835,
      "tokens_seen": 517603328
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004258375125376128,
      "loss": 3.053,
      "theoretical_loss": 3.90028314188806,
      "tokens_seen": 517668864
    },
    {
      "epoch": 0.16,
      "objective/train/advantage_avg": 0.4931814968585968,
      "objective/train/docs_used": 303675,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.0157575607299805,
      "objective/train/original_loss": 3.0157573223114014,
      "objective/train/theoretical_loss": 3.9002302802201427,
      "objective/train/tokens_used": 538194400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.243355855345726,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505551099777222,
      "objective/train/weighted_lm_loss": 3.168142795562744,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 3.9002302802201427,
      "tokens_seen": 517734400
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042582748244734205,
      "loss": 3.0511,
      "theoretical_loss": 3.9002302802201427,
      "tokens_seen": 517734400
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004258174523570712,
      "loss": 3.1636,
      "theoretical_loss": 3.900177427116461,
      "tokens_seen": 517799936
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004258074222668004,
      "loss": 3.2354,
      "theoretical_loss": 3.9001245825745423,
      "tokens_seen": 517865472
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004257973921765296,
      "loss": 2.7945,
      "theoretical_loss": 3.9000717465919177,
      "tokens_seen": 517931008
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042578736208625877,
      "loss": 2.9537,
      "theoretical_loss": 3.9000189191661163,
      "tokens_seen": 517996544
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042577733199598795,
      "loss": 3.1296,
      "theoretical_loss": 3.8999661002946713,
      "tokens_seen": 518062080
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004257673019057172,
      "loss": 3.007,
      "theoretical_loss": 3.8999132899751157,
      "tokens_seen": 518127616
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004257572718154463,
      "loss": 3.0525,
      "theoretical_loss": 3.8998604882049825,
      "tokens_seen": 518193152
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042574724172517555,
      "loss": 3.2638,
      "theoretical_loss": 3.8998076949818072,
      "tokens_seen": 518258688
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004257372116349047,
      "loss": 2.9727,
      "theoretical_loss": 3.899754910303126,
      "tokens_seen": 518324224
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004257271815446339,
      "loss": 3.0974,
      "theoretical_loss": 3.8997021341664757,
      "tokens_seen": 518389760
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004257171514543631,
      "loss": 3.2029,
      "theoretical_loss": 3.8996493665693936,
      "tokens_seen": 518455296
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004257071213640923,
      "loss": 2.8704,
      "theoretical_loss": 3.8995966075094204,
      "tokens_seen": 518520832
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042569709127382145,
      "loss": 3.0681,
      "theoretical_loss": 3.899543856984095,
      "tokens_seen": 518586368
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042568706118355064,
      "loss": 2.8915,
      "theoretical_loss": 3.8994911149909575,
      "tokens_seen": 518651904
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004256770310932798,
      "loss": 2.8462,
      "theoretical_loss": 3.8994383815275517,
      "tokens_seen": 518717440
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042566700100300905,
      "loss": 2.9058,
      "theoretical_loss": 3.899385656591421,
      "tokens_seen": 518782976
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042565697091273823,
      "loss": 3.0286,
      "theoretical_loss": 3.8993329401801082,
      "tokens_seen": 518848512
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004256469408224674,
      "loss": 3.0447,
      "theoretical_loss": 3.899280232291159,
      "tokens_seen": 518914048
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042563691073219665,
      "loss": 2.958,
      "theoretical_loss": 3.899227532922119,
      "tokens_seen": 518979584
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004256268806419258,
      "loss": 2.7642,
      "theoretical_loss": 3.8991748420705363,
      "tokens_seen": 519045120
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.000425616850551655,
      "loss": 3.1247,
      "theoretical_loss": 3.8991221597339587,
      "tokens_seen": 519110656
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042560682046138414,
      "loss": 2.966,
      "theoretical_loss": 3.8990694859099357,
      "tokens_seen": 519176192
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004255967903711134,
      "loss": 3.0964,
      "theoretical_loss": 3.8990168205960174,
      "tokens_seen": 519241728
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042558676028084256,
      "loss": 3.3644,
      "theoretical_loss": 3.8989641637897545,
      "tokens_seen": 519307264
    },
    {
      "epoch": 0.16,
      "objective/train/advantage_avg": 0.49435409903526306,
      "objective/train/docs_used": 304298,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.995551347732544,
      "objective/train/original_loss": 2.995551109313965,
      "objective/train/theoretical_loss": 3.8989115154886997,
      "objective/train/tokens_used": 539832800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2444542944431305,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050678014755249,
      "objective/train/weighted_lm_loss": 3.1473960876464844,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0435127019882202,
      "theoretical_loss": 3.8989115154886997,
      "tokens_seen": 519372800
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042557673019057174,
      "loss": 3.3336,
      "theoretical_loss": 3.8989115154886997,
      "tokens_seen": 519372800
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004255667001003009,
      "loss": 3.1428,
      "theoretical_loss": 3.898858875690406,
      "tokens_seen": 519438336
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004255566700100301,
      "loss": 3.0337,
      "theoretical_loss": 3.8988062443924285,
      "tokens_seen": 519503872
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004255466399197593,
      "loss": 3.3044,
      "theoretical_loss": 3.8987536215923217,
      "tokens_seen": 519569408
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004255366098294885,
      "loss": 3.2087,
      "theoretical_loss": 3.8987010072876425,
      "tokens_seen": 519634944
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042552657973921764,
      "loss": 3.2045,
      "theoretical_loss": 3.8986484014759473,
      "tokens_seen": 519700480
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004255165496489469,
      "loss": 3.0809,
      "theoretical_loss": 3.898595804154795,
      "tokens_seen": 519766016
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.000425506519558676,
      "loss": 2.9709,
      "theoretical_loss": 3.8985432153217454,
      "tokens_seen": 519831552
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042549648946840524,
      "loss": 3.0386,
      "theoretical_loss": 3.898490634974358,
      "tokens_seen": 519897088
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004254864593781344,
      "loss": 3.3589,
      "theoretical_loss": 3.898438063110195,
      "tokens_seen": 519962624
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004254764292878636,
      "loss": 3.029,
      "theoretical_loss": 3.8983854997268175,
      "tokens_seen": 520028160
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004254663991975928,
      "loss": 2.7227,
      "theoretical_loss": 3.8983329448217905,
      "tokens_seen": 520093696
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.000425456369107322,
      "loss": 2.9348,
      "theoretical_loss": 3.898280398392677,
      "tokens_seen": 520159232
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042544633901705115,
      "loss": 2.8176,
      "theoretical_loss": 3.8982278604370437,
      "tokens_seen": 520224768
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004254363089267804,
      "loss": 3.2669,
      "theoretical_loss": 3.898175330952456,
      "tokens_seen": 520290304
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004254262788365095,
      "loss": 3.0439,
      "theoretical_loss": 3.8981228099364804,
      "tokens_seen": 520355840
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042541624874623874,
      "loss": 3.1032,
      "theoretical_loss": 3.8980702973866874,
      "tokens_seen": 520421376
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004254062186559679,
      "loss": 2.8731,
      "theoretical_loss": 3.8980177933006455,
      "tokens_seen": 520486912
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004253961885656971,
      "loss": 3.2608,
      "theoretical_loss": 3.897965297675925,
      "tokens_seen": 520552448
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004253861584754263,
      "loss": 3.0696,
      "theoretical_loss": 3.897912810510097,
      "tokens_seen": 520617984
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042537612838515547,
      "loss": 3.0578,
      "theoretical_loss": 3.897860331800735,
      "tokens_seen": 520683520
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042536609829488465,
      "loss": 2.8436,
      "theoretical_loss": 3.8978078615454113,
      "tokens_seen": 520749056
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004253560682046139,
      "loss": 3.0807,
      "theoretical_loss": 3.8977553997417003,
      "tokens_seen": 520814592
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.000425346038114343,
      "loss": 3.0319,
      "theoretical_loss": 3.897702946387178,
      "tokens_seen": 520880128
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042533600802407225,
      "loss": 2.7688,
      "theoretical_loss": 3.897650501479421,
      "tokens_seen": 520945664
    },
    {
      "epoch": 0.16,
      "objective/train/advantage_avg": 0.49459052085876465,
      "objective/train/docs_used": 305033,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.937790632247925,
      "objective/train/original_loss": 2.9377903938293457,
      "objective/train/theoretical_loss": 3.8975980650160067,
      "objective/train/tokens_used": 541471200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.244686096906662,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507028102874756,
      "objective/train/weighted_lm_loss": 3.086714029312134,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0304546356201172,
      "theoretical_loss": 3.8975980650160067,
      "tokens_seen": 521011200
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004253259779338014,
      "loss": 2.8331,
      "theoretical_loss": 3.8975980650160067,
      "tokens_seen": 521011200
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004253159478435306,
      "loss": 2.9348,
      "theoretical_loss": 3.897545636994513,
      "tokens_seen": 521076736
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004253059177532598,
      "loss": 3.2274,
      "theoretical_loss": 3.8974932174125194,
      "tokens_seen": 521142272
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042529588766298897,
      "loss": 2.8943,
      "theoretical_loss": 3.8974408062676065,
      "tokens_seen": 521207808
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042528585757271815,
      "loss": 3.0407,
      "theoretical_loss": 3.897388403557356,
      "tokens_seen": 521273344
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004252758274824474,
      "loss": 3.1112,
      "theoretical_loss": 3.897336009279349,
      "tokens_seen": 521338880
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004252657973921765,
      "loss": 3.1349,
      "theoretical_loss": 3.8972836234311705,
      "tokens_seen": 521404416
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042525576730190575,
      "loss": 3.0535,
      "theoretical_loss": 3.8972312460104046,
      "tokens_seen": 521469952
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004252457372116349,
      "loss": 3.07,
      "theoretical_loss": 3.8971788770146363,
      "tokens_seen": 521535488
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004252357071213641,
      "loss": 3.2448,
      "theoretical_loss": 3.8971265164414515,
      "tokens_seen": 521601024
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004252256770310933,
      "loss": 3.2685,
      "theoretical_loss": 3.897074164288439,
      "tokens_seen": 521666560
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004252156469408225,
      "loss": 3.0068,
      "theoretical_loss": 3.8970218205531855,
      "tokens_seen": 521732096
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042520561685055165,
      "loss": 2.9935,
      "theoretical_loss": 3.8969694852332815,
      "tokens_seen": 521797632
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042519558676028084,
      "loss": 3.2718,
      "theoretical_loss": 3.8969171583263176,
      "tokens_seen": 521863168
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042518555667001,
      "loss": 2.7916,
      "theoretical_loss": 3.8968648398298846,
      "tokens_seen": 521928704
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042517552657973925,
      "loss": 2.9721,
      "theoretical_loss": 3.8968125297415748,
      "tokens_seen": 521994240
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004251654964894684,
      "loss": 3.0621,
      "theoretical_loss": 3.896760228058982,
      "tokens_seen": 522059776
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004251554663991976,
      "loss": 2.9568,
      "theoretical_loss": 3.8967079347797,
      "tokens_seen": 522125312
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004251454363089268,
      "loss": 2.9497,
      "theoretical_loss": 3.896655649901324,
      "tokens_seen": 522190848
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.000425135406218656,
      "loss": 3.1641,
      "theoretical_loss": 3.896603373421451,
      "tokens_seen": 522256384
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042512537612838516,
      "loss": 3.1313,
      "theoretical_loss": 3.896551105337678,
      "tokens_seen": 522321920
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042511534603811434,
      "loss": 3.2826,
      "theoretical_loss": 3.8964988456476033,
      "tokens_seen": 522387456
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004251053159478435,
      "loss": 2.796,
      "theoretical_loss": 3.8964465943488262,
      "tokens_seen": 522452992
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042509528585757276,
      "loss": 2.9879,
      "theoretical_loss": 3.896394351438947,
      "tokens_seen": 522518528
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004250852557673019,
      "loss": 2.9154,
      "theoretical_loss": 3.8963421169155668,
      "tokens_seen": 522584064
    },
    {
      "epoch": 0.16,
      "objective/train/advantage_avg": 0.4936038851737976,
      "objective/train/docs_used": 306243,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.279616355895996,
      "objective/train/original_loss": 3.279616117477417,
      "objective/train/theoretical_loss": 3.8962898907762877,
      "objective/train/tokens_used": 543109600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24376316368579865,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505995750427246,
      "objective/train/weighted_lm_loss": 3.445645332336426,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 3.8962898907762877,
      "tokens_seen": 522649600
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004250752256770311,
      "loss": 2.9818,
      "theoretical_loss": 3.8962898907762877,
      "tokens_seen": 522649600
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042506519558676024,
      "loss": 3.3666,
      "theoretical_loss": 3.896237673018714,
      "tokens_seen": 522715136
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004250551654964895,
      "loss": 3.1154,
      "theoretical_loss": 3.8961854636404487,
      "tokens_seen": 522780672
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042504513540621866,
      "loss": 3.0973,
      "theoretical_loss": 3.896133262639097,
      "tokens_seen": 522846208
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042503510531594784,
      "loss": 2.9501,
      "theoretical_loss": 3.896081070012266,
      "tokens_seen": 522911744
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.000425025075225677,
      "loss": 3.097,
      "theoretical_loss": 3.8960288857575622,
      "tokens_seen": 522977280
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004250150451354062,
      "loss": 2.9311,
      "theoretical_loss": 3.8959767098725937,
      "tokens_seen": 523042816
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004250050150451354,
      "loss": 3.0671,
      "theoretical_loss": 3.89592454235497,
      "tokens_seen": 523108352
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004249949849548646,
      "loss": 3.2292,
      "theoretical_loss": 3.8958723832023017,
      "tokens_seen": 523173888
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042498495486459375,
      "loss": 2.958,
      "theoretical_loss": 3.8958202324121984,
      "tokens_seen": 523239424
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.000424974924774323,
      "loss": 3.0829,
      "theoretical_loss": 3.8957680899822735,
      "tokens_seen": 523304960
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042496489468405216,
      "loss": 3.2108,
      "theoretical_loss": 3.8957159559101395,
      "tokens_seen": 523370496
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042495486459378135,
      "loss": 2.9417,
      "theoretical_loss": 3.895663830193411,
      "tokens_seen": 523436032
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004249448345035105,
      "loss": 3.1747,
      "theoretical_loss": 3.8956117128297025,
      "tokens_seen": 523501568
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004249348044132397,
      "loss": 2.908,
      "theoretical_loss": 3.89555960381663,
      "tokens_seen": 523567104
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042492477432296894,
      "loss": 3.0118,
      "theoretical_loss": 3.8955075031518103,
      "tokens_seen": 523632640
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004249147442326981,
      "loss": 3.1256,
      "theoretical_loss": 3.8954554108328616,
      "tokens_seen": 523698176
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004249047141424273,
      "loss": 2.911,
      "theoretical_loss": 3.8954033268574033,
      "tokens_seen": 523763712
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004248946840521565,
      "loss": 3.2939,
      "theoretical_loss": 3.8953512512230546,
      "tokens_seen": 523829248
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042488465396188567,
      "loss": 3.0961,
      "theoretical_loss": 3.8952991839274373,
      "tokens_seen": 523894784
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042487462387161485,
      "loss": 2.9043,
      "theoretical_loss": 3.895247124968172,
      "tokens_seen": 523960320
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004248645937813441,
      "loss": 2.9232,
      "theoretical_loss": 3.8951950743428827,
      "tokens_seen": 524025856
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004248545636910732,
      "loss": 3.1005,
      "theoretical_loss": 3.895143032049192,
      "tokens_seen": 524091392
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042484453360080245,
      "loss": 2.9506,
      "theoretical_loss": 3.8950909980847266,
      "tokens_seen": 524156928
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004248345035105316,
      "loss": 3.2974,
      "theoretical_loss": 3.89503897244711,
      "tokens_seen": 524222464
    },
    {
      "epoch": 0.16,
      "objective/train/advantage_avg": 0.4956108629703522,
      "objective/train/docs_used": 306972,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.665851354598999,
      "objective/train/original_loss": 2.665851593017578,
      "objective/train/theoretical_loss": 3.8949869551339704,
      "objective/train/tokens_used": 544748000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24568070471286774,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0508100986480713,
      "objective/train/weighted_lm_loss": 2.8012027740478516,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 3.8949869551339704,
      "tokens_seen": 524288000
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004248244734202608,
      "loss": 2.9652,
      "theoretical_loss": 3.8949869551339704,
      "tokens_seen": 524288000
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042481444332999,
      "loss": 3.2608,
      "theoretical_loss": 3.8949349461429357,
      "tokens_seen": 524353536
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042480441323971917,
      "loss": 3.1115,
      "theoretical_loss": 3.8948829454716334,
      "tokens_seen": 524419072
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042479438314944835,
      "loss": 2.8814,
      "theoretical_loss": 3.894830953117694,
      "tokens_seen": 524484608
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004247843530591776,
      "loss": 3.1129,
      "theoretical_loss": 3.8947789690787475,
      "tokens_seen": 524550144
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004247743229689067,
      "loss": 3.1963,
      "theoretical_loss": 3.8947269933524264,
      "tokens_seen": 524615680
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042476429287863595,
      "loss": 3.2116,
      "theoretical_loss": 3.8946750259363627,
      "tokens_seen": 524681216
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004247542627883651,
      "loss": 2.839,
      "theoretical_loss": 3.8946230668281903,
      "tokens_seen": 524746752
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004247442326980943,
      "loss": 3.0008,
      "theoretical_loss": 3.894571116025543,
      "tokens_seen": 524812288
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004247342026078235,
      "loss": 3.2121,
      "theoretical_loss": 3.894519173526057,
      "tokens_seen": 524877824
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004247241725175527,
      "loss": 3.0874,
      "theoretical_loss": 3.8944672393273683,
      "tokens_seen": 524943360
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042471414242728186,
      "loss": 2.807,
      "theoretical_loss": 3.894415313427115,
      "tokens_seen": 525008896
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042470411233701104,
      "loss": 3.3174,
      "theoretical_loss": 3.8943633958229347,
      "tokens_seen": 525074432
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004246940822467402,
      "loss": 3.1191,
      "theoretical_loss": 3.894311486512467,
      "tokens_seen": 525139968
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042468405215646945,
      "loss": 2.8126,
      "theoretical_loss": 3.8942595854933515,
      "tokens_seen": 525205504
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004246740220661986,
      "loss": 2.8117,
      "theoretical_loss": 3.894207692763231,
      "tokens_seen": 525271040
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004246639919759278,
      "loss": 3.0238,
      "theoretical_loss": 3.8941558083197467,
      "tokens_seen": 525336576
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.000424653961885657,
      "loss": 3.1955,
      "theoretical_loss": 3.8941039321605424,
      "tokens_seen": 525402112
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004246439317953862,
      "loss": 3.1733,
      "theoretical_loss": 3.8940520642832617,
      "tokens_seen": 525467648
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042463390170511536,
      "loss": 3.1302,
      "theoretical_loss": 3.8940002046855495,
      "tokens_seen": 525533184
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042462387161484454,
      "loss": 3.0255,
      "theoretical_loss": 3.8939483533650536,
      "tokens_seen": 525598720
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004246138415245737,
      "loss": 3.0049,
      "theoretical_loss": 3.8938965103194185,
      "tokens_seen": 525664256
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042460381143430296,
      "loss": 2.8827,
      "theoretical_loss": 3.8938446755462945,
      "tokens_seen": 525729792
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004245937813440321,
      "loss": 2.9031,
      "theoretical_loss": 3.8937928490433293,
      "tokens_seen": 525795328
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004245837512537613,
      "loss": 3.043,
      "theoretical_loss": 3.8937410308081737,
      "tokens_seen": 525860864
    },
    {
      "epoch": 0.16,
      "objective/train/advantage_avg": 0.4947790205478668,
      "objective/train/docs_used": 308263,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.184535026550293,
      "objective/train/original_loss": 3.184535026550293,
      "objective/train/theoretical_loss": 3.8936892208384775,
      "objective/train/tokens_used": 546386400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24486343562602997,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050722599029541,
      "objective/train/weighted_lm_loss": 3.3460376262664795,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0425469875335693,
      "theoretical_loss": 3.8936892208384775,
      "tokens_seen": 525926400
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042457372116349044,
      "loss": 3.0454,
      "theoretical_loss": 3.8936892208384775,
      "tokens_seen": 525926400
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004245636910732197,
      "loss": 3.0994,
      "theoretical_loss": 3.8936374191318937,
      "tokens_seen": 525991936
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042455366098294886,
      "loss": 2.9637,
      "theoretical_loss": 3.893585625686074,
      "tokens_seen": 526057472
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042454363089267804,
      "loss": 3.0834,
      "theoretical_loss": 3.8935338404986735,
      "tokens_seen": 526123008
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004245336008024072,
      "loss": 3.2529,
      "theoretical_loss": 3.8934820635673457,
      "tokens_seen": 526188544
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004245235707121364,
      "loss": 3.2036,
      "theoretical_loss": 3.8934302948897472,
      "tokens_seen": 526254080
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004245135406218656,
      "loss": 3.0722,
      "theoretical_loss": 3.893378534463534,
      "tokens_seen": 526319616
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004245035105315948,
      "loss": 3.1422,
      "theoretical_loss": 3.8933267822863646,
      "tokens_seen": 526385152
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042449348044132395,
      "loss": 3.0504,
      "theoretical_loss": 3.8932750383558963,
      "tokens_seen": 526450688
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004244834503510532,
      "loss": 3.1253,
      "theoretical_loss": 3.8932233026697896,
      "tokens_seen": 526516224
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042447342026078236,
      "loss": 3.0733,
      "theoretical_loss": 3.8931715752257054,
      "tokens_seen": 526581760
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042446339017051155,
      "loss": 2.8555,
      "theoretical_loss": 3.893119856021304,
      "tokens_seen": 526647296
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004244533600802407,
      "loss": 3.1176,
      "theoretical_loss": 3.893068145054248,
      "tokens_seen": 526712832
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004244433299899699,
      "loss": 2.8781,
      "theoretical_loss": 3.8930164423222013,
      "tokens_seen": 526778368
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004244332998996991,
      "loss": 3.1074,
      "theoretical_loss": 3.8929647478228278,
      "tokens_seen": 526843904
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004244232698094283,
      "loss": 3.0932,
      "theoretical_loss": 3.8929130615537932,
      "tokens_seen": 526909440
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042441323971915745,
      "loss": 2.8913,
      "theoretical_loss": 3.8928613835127632,
      "tokens_seen": 526974976
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004244032096288867,
      "loss": 2.8894,
      "theoretical_loss": 3.892809713697405,
      "tokens_seen": 527040512
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004243931795386158,
      "loss": 2.9954,
      "theoretical_loss": 3.8927580521053873,
      "tokens_seen": 527106048
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042438314944834505,
      "loss": 2.98,
      "theoretical_loss": 3.8927063987343784,
      "tokens_seen": 527171584
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042437311935807423,
      "loss": 2.994,
      "theoretical_loss": 3.8926547535820486,
      "tokens_seen": 527237120
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004243630892678034,
      "loss": 3.0832,
      "theoretical_loss": 3.892603116646069,
      "tokens_seen": 527302656
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004243530591775326,
      "loss": 3.069,
      "theoretical_loss": 3.8925514879241114,
      "tokens_seen": 527368192
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004243430290872618,
      "loss": 2.9525,
      "theoretical_loss": 3.8924998674138487,
      "tokens_seen": 527433728
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042433299899699095,
      "loss": 3.1241,
      "theoretical_loss": 3.8924482551129547,
      "tokens_seen": 527499264
    },
    {
      "epoch": 0.16,
      "objective/train/advantage_avg": 0.49543362855911255,
      "objective/train/docs_used": 309026,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.9031453132629395,
      "objective/train/original_loss": 2.9031450748443604,
      "objective/train/theoretical_loss": 3.892396651019104,
      "objective/train/tokens_used": 548024800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24549831449985504,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507912635803223,
      "objective/train/weighted_lm_loss": 3.0504672527313232,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.042169213294983,
      "theoretical_loss": 3.892396651019104,
      "tokens_seen": 527564800
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004243229689067202,
      "loss": 3.05,
      "theoretical_loss": 3.892396651019104,
      "tokens_seen": 527564800
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004243129388164493,
      "loss": 2.9147,
      "theoretical_loss": 3.8923450551299728,
      "tokens_seen": 527630336
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042430290872617855,
      "loss": 3.0156,
      "theoretical_loss": 3.8922934674432375,
      "tokens_seen": 527695872
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042429287863590773,
      "loss": 3.372,
      "theoretical_loss": 3.892241887956575,
      "tokens_seen": 527761408
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004242828485456369,
      "loss": 3.196,
      "theoretical_loss": 3.892190316667665,
      "tokens_seen": 527826944
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004242728184553661,
      "loss": 3.2227,
      "theoretical_loss": 3.8921387535741863,
      "tokens_seen": 527892480
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004242627883650953,
      "loss": 3.1558,
      "theoretical_loss": 3.8920871986738197,
      "tokens_seen": 527958016
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042425275827482446,
      "loss": 2.8435,
      "theoretical_loss": 3.892035651964246,
      "tokens_seen": 528023552
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004242427281845537,
      "loss": 2.8183,
      "theoretical_loss": 3.891984113443148,
      "tokens_seen": 528089088
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004242326980942828,
      "loss": 3.0955,
      "theoretical_loss": 3.8919325831082086,
      "tokens_seen": 528154624
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042422266800401206,
      "loss": 3.0614,
      "theoretical_loss": 3.891881060957113,
      "tokens_seen": 528220160
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004242126379137412,
      "loss": 3.1953,
      "theoretical_loss": 3.8918295469875446,
      "tokens_seen": 528285696
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004242026078234704,
      "loss": 3.3451,
      "theoretical_loss": 3.891778041197191,
      "tokens_seen": 528351232
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004241925777331996,
      "loss": 2.7851,
      "theoretical_loss": 3.891726543583739,
      "tokens_seen": 528416768
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004241825476429288,
      "loss": 2.9592,
      "theoretical_loss": 3.8916750541448764,
      "tokens_seen": 528482304
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.000424172517552658,
      "loss": 3.0062,
      "theoretical_loss": 3.8916235728782915,
      "tokens_seen": 528547840
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004241624874623872,
      "loss": 3.0511,
      "theoretical_loss": 3.8915720997816745,
      "tokens_seen": 528613376
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004241524573721164,
      "loss": 3.0032,
      "theoretical_loss": 3.891520634852717,
      "tokens_seen": 528678912
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042414242728184556,
      "loss": 3.0803,
      "theoretical_loss": 3.89146917808911,
      "tokens_seen": 528744448
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042413239719157474,
      "loss": 3.1276,
      "theoretical_loss": 3.8914177294885457,
      "tokens_seen": 528809984
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004241223671013039,
      "loss": 3.3372,
      "theoretical_loss": 3.891366289048719,
      "tokens_seen": 528875520
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042411233701103316,
      "loss": 3.1712,
      "theoretical_loss": 3.8913148567673232,
      "tokens_seen": 528941056
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004241023069207623,
      "loss": 3.0705,
      "theoretical_loss": 3.891263432642055,
      "tokens_seen": 529006592
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004240922768304915,
      "loss": 3.0729,
      "theoretical_loss": 3.8912120166706092,
      "tokens_seen": 529072128
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042408224674022065,
      "loss": 2.9057,
      "theoretical_loss": 3.8911606088506847,
      "tokens_seen": 529137664
    },
    {
      "epoch": 0.16,
      "objective/train/advantage_avg": 0.493380069732666,
      "objective/train/docs_used": 310459,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.8043107986450195,
      "objective/train/original_loss": 2.8043107986450195,
      "objective/train/theoretical_loss": 3.8911092091799793,
      "objective/train/tokens_used": 549663200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24349981546401978,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505757331848145,
      "objective/train/weighted_lm_loss": 2.946136951446533,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0425469875335693,
      "theoretical_loss": 3.8911092091799793,
      "tokens_seen": 529203200
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004240722166499499,
      "loss": 2.9389,
      "theoretical_loss": 3.8911092091799793,
      "tokens_seen": 529203200
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042406218655967906,
      "loss": 3.0186,
      "theoretical_loss": 3.891057817656192,
      "tokens_seen": 529268736
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042405215646940824,
      "loss": 3.1148,
      "theoretical_loss": 3.891006434277023,
      "tokens_seen": 529334272
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004240421263791374,
      "loss": 3.0123,
      "theoretical_loss": 3.8909550590401736,
      "tokens_seen": 529399808
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004240320962888666,
      "loss": 3.0875,
      "theoretical_loss": 3.890903691943346,
      "tokens_seen": 529465344
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004240220661985958,
      "loss": 3.0379,
      "theoretical_loss": 3.890852332984242,
      "tokens_seen": 529530880
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.000424012036108325,
      "loss": 2.8695,
      "theoretical_loss": 3.890800982160567,
      "tokens_seen": 529596416
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042400200601805415,
      "loss": 3.1367,
      "theoretical_loss": 3.890749639470025,
      "tokens_seen": 529661952
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004239919759277834,
      "loss": 2.9686,
      "theoretical_loss": 3.8906983049103223,
      "tokens_seen": 529727488
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042398194583751256,
      "loss": 2.9756,
      "theoretical_loss": 3.8906469784791646,
      "tokens_seen": 529793024
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042397191574724175,
      "loss": 3.3234,
      "theoretical_loss": 3.8905956601742604,
      "tokens_seen": 529858560
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004239618856569709,
      "loss": 2.9123,
      "theoretical_loss": 3.890544349993318,
      "tokens_seen": 529924096
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004239518555667001,
      "loss": 2.9209,
      "theoretical_loss": 3.890493047934047,
      "tokens_seen": 529989632
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004239418254764293,
      "loss": 2.9722,
      "theoretical_loss": 3.890441753994158,
      "tokens_seen": 530055168
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004239317953861585,
      "loss": 2.9414,
      "theoretical_loss": 3.8903904681713612,
      "tokens_seen": 530120704
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042392176529588765,
      "loss": 3.0508,
      "theoretical_loss": 3.89033919046337,
      "tokens_seen": 530186240
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004239117352056169,
      "loss": 3.0374,
      "theoretical_loss": 3.8902879208678973,
      "tokens_seen": 530251776
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.000423901705115346,
      "loss": 3.2712,
      "theoretical_loss": 3.8902366593826567,
      "tokens_seen": 530317312
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042389167502507525,
      "loss": 2.907,
      "theoretical_loss": 3.890185406005364,
      "tokens_seen": 530382848
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042388164493480443,
      "loss": 3.1275,
      "theoretical_loss": 3.8901341607337354,
      "tokens_seen": 530448384
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004238716148445336,
      "loss": 3.2734,
      "theoretical_loss": 3.8900829235654872,
      "tokens_seen": 530513920
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004238615847542628,
      "loss": 2.9035,
      "theoretical_loss": 3.890031694498337,
      "tokens_seen": 530579456
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.000423851554663992,
      "loss": 2.7728,
      "theoretical_loss": 3.889980473530004,
      "tokens_seen": 530644992
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042384152457372115,
      "loss": 3.1136,
      "theoretical_loss": 3.8899292606582074,
      "tokens_seen": 530710528
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004238314944834504,
      "loss": 2.994,
      "theoretical_loss": 3.8898780558806685,
      "tokens_seen": 530776064
    },
    {
      "epoch": 0.16,
      "objective/train/advantage_avg": 0.49468815326690674,
      "objective/train/docs_used": 311126,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.9695303440093994,
      "objective/train/original_loss": 2.9695305824279785,
      "objective/train/theoretical_loss": 3.889826859195108,
      "objective/train/tokens_used": 551301600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24480140209197998,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507131814956665,
      "objective/train/weighted_lm_loss": 3.1200921535491943,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0304546356201172,
      "theoretical_loss": 3.889826859195108,
      "tokens_seen": 530841600
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004238214643931795,
      "loss": 3.0934,
      "theoretical_loss": 3.889826859195108,
      "tokens_seen": 530841600
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042381143430290875,
      "loss": 3.1411,
      "theoretical_loss": 3.8897756705992492,
      "tokens_seen": 530907136
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042380140421263793,
      "loss": 3.0346,
      "theoretical_loss": 3.8897244900908152,
      "tokens_seen": 530972672
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004237913741223671,
      "loss": 3.2248,
      "theoretical_loss": 3.8896733176675298,
      "tokens_seen": 531038208
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004237813440320963,
      "loss": 3.1227,
      "theoretical_loss": 3.8896221533271182,
      "tokens_seen": 531103744
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004237713139418255,
      "loss": 2.6808,
      "theoretical_loss": 3.8895709970673074,
      "tokens_seen": 531169280
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042376128385155466,
      "loss": 2.9806,
      "theoretical_loss": 3.889519848885824,
      "tokens_seen": 531234816
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004237512537612839,
      "loss": 3.0917,
      "theoretical_loss": 3.889468708780395,
      "tokens_seen": 531300352
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.000423741223671013,
      "loss": 2.9398,
      "theoretical_loss": 3.8894175767487504,
      "tokens_seen": 531365888
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042373119358074226,
      "loss": 3.111,
      "theoretical_loss": 3.88936645278862,
      "tokens_seen": 531431424
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004237211634904714,
      "loss": 3.1072,
      "theoretical_loss": 3.889315336897734,
      "tokens_seen": 531496960
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004237111334002006,
      "loss": 2.9604,
      "theoretical_loss": 3.8892642290738246,
      "tokens_seen": 531562496
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004237011033099298,
      "loss": 3.1272,
      "theoretical_loss": 3.8892131293146237,
      "tokens_seen": 531628032
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.000423691073219659,
      "loss": 3.0513,
      "theoretical_loss": 3.889162037617865,
      "tokens_seen": 531693568
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042368104312938816,
      "loss": 2.8656,
      "theoretical_loss": 3.8891109539812834,
      "tokens_seen": 531759104
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004236710130391174,
      "loss": 3.0358,
      "theoretical_loss": 3.889059878402614,
      "tokens_seen": 531824640
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004236609829488465,
      "loss": 3.1711,
      "theoretical_loss": 3.8890088108795924,
      "tokens_seen": 531890176
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042365095285857576,
      "loss": 2.9462,
      "theoretical_loss": 3.888957751409956,
      "tokens_seen": 531955712
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004236409227683049,
      "loss": 3.0827,
      "theoretical_loss": 3.888906699991443,
      "tokens_seen": 532021248
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004236308926780341,
      "loss": 3.3931,
      "theoretical_loss": 3.8888556566217924,
      "tokens_seen": 532086784
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004236208625877633,
      "loss": 2.9661,
      "theoretical_loss": 3.8888046212987444,
      "tokens_seen": 532152320
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004236108324974925,
      "loss": 3.1784,
      "theoretical_loss": 3.888753594020039,
      "tokens_seen": 532217856
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042360080240722166,
      "loss": 2.8912,
      "theoretical_loss": 3.8887025747834185,
      "tokens_seen": 532283392
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042359077231695085,
      "loss": 2.9848,
      "theoretical_loss": 3.8886515635866252,
      "tokens_seen": 532348928
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042358074222668,
      "loss": 3.2543,
      "theoretical_loss": 3.8886005604274025,
      "tokens_seen": 532414464
    },
    {
      "epoch": 0.16,
      "objective/train/advantage_avg": 0.4937783479690552,
      "objective/train/docs_used": 312276,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.268366813659668,
      "objective/train/original_loss": 3.268366813659668,
      "objective/train/theoretical_loss": 3.888549565303496,
      "objective/train/tokens_used": 552940000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24388016760349274,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506175756454468,
      "objective/train/weighted_lm_loss": 3.433738946914673,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0425469875335693,
      "theoretical_loss": 3.888549565303496,
      "tokens_seen": 532480000
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042357071213640926,
      "loss": 3.2151,
      "theoretical_loss": 3.888549565303496,
      "tokens_seen": 532480000
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004235606820461384,
      "loss": 2.9475,
      "theoretical_loss": 3.8884985782126495,
      "tokens_seen": 532545536
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004235506519558676,
      "loss": 2.7783,
      "theoretical_loss": 3.8884475991526095,
      "tokens_seen": 532611072
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042354062186559675,
      "loss": 3.1036,
      "theoretical_loss": 3.888396628121124,
      "tokens_seen": 532676608
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.000423530591775326,
      "loss": 2.8275,
      "theoretical_loss": 3.8883456651159407,
      "tokens_seen": 532742144
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042352056168505517,
      "loss": 2.8417,
      "theoretical_loss": 3.888294710134808,
      "tokens_seen": 532807680
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042351053159478435,
      "loss": 2.8942,
      "theoretical_loss": 3.8882437631754767,
      "tokens_seen": 532873216
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042350050150451353,
      "loss": 3.2227,
      "theoretical_loss": 3.8881928242356976,
      "tokens_seen": 532938752
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042349047141424276,
      "loss": 3.1617,
      "theoretical_loss": 3.8881418933132212,
      "tokens_seen": 533004288
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004234804413239719,
      "loss": 3.2681,
      "theoretical_loss": 3.888090970405801,
      "tokens_seen": 533069824
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042347041123370113,
      "loss": 2.9329,
      "theoretical_loss": 3.8880400555111914,
      "tokens_seen": 533135360
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042346038114343025,
      "loss": 3.0502,
      "theoretical_loss": 3.8879891486271445,
      "tokens_seen": 533200896
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004234503510531595,
      "loss": 3.131,
      "theoretical_loss": 3.887938249751418,
      "tokens_seen": 533266432
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042344032096288867,
      "loss": 3.0446,
      "theoretical_loss": 3.887887358881766,
      "tokens_seen": 533331968
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042343029087261785,
      "loss": 3.086,
      "theoretical_loss": 3.8878364760159476,
      "tokens_seen": 533397504
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004234202607823471,
      "loss": 2.9397,
      "theoretical_loss": 3.88778560115172,
      "tokens_seen": 533463040
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004234102306920762,
      "loss": 3.0279,
      "theoretical_loss": 3.887734734286842,
      "tokens_seen": 533528576
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042340020060180545,
      "loss": 3.0379,
      "theoretical_loss": 3.8876838754190732,
      "tokens_seen": 533594112
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042339017051153463,
      "loss": 3.1687,
      "theoretical_loss": 3.887633024546175,
      "tokens_seen": 533659648
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004233801404212638,
      "loss": 2.8732,
      "theoretical_loss": 3.887582181665909,
      "tokens_seen": 533725184
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.000423370110330993,
      "loss": 2.9568,
      "theoretical_loss": 3.8875313467760373,
      "tokens_seen": 533790720
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004233600802407222,
      "loss": 3.185,
      "theoretical_loss": 3.887480519874323,
      "tokens_seen": 533856256
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042335005015045135,
      "loss": 3.087,
      "theoretical_loss": 3.887429700958532,
      "tokens_seen": 533921792
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004233400200601806,
      "loss": 3.0775,
      "theoretical_loss": 3.8873788900264286,
      "tokens_seen": 533987328
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004233299899699097,
      "loss": 2.9142,
      "theoretical_loss": 3.887328087075778,
      "tokens_seen": 534052864
    },
    {
      "epoch": 0.16,
      "objective/train/advantage_avg": 0.49368903040885925,
      "objective/train/docs_used": 312792,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.230130910873413,
      "objective/train/original_loss": 3.230130672454834,
      "objective/train/theoretical_loss": 3.887277292104349,
      "objective/train/tokens_used": 554578400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24379783868789673,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506082773208618,
      "objective/train/weighted_lm_loss": 3.393442153930664,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0437887907028198,
      "theoretical_loss": 3.887277292104349,
      "tokens_seen": 534118400
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042331995987963895,
      "loss": 3.3005,
      "theoretical_loss": 3.887277292104349,
      "tokens_seen": 534118400
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042330992978936813,
      "loss": 3.0515,
      "theoretical_loss": 3.8872265051099086,
      "tokens_seen": 534183936
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004232998996990973,
      "loss": 2.9348,
      "theoretical_loss": 3.8871757260902253,
      "tokens_seen": 534249472
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004232898696088265,
      "loss": 3.1087,
      "theoretical_loss": 3.88712495504307,
      "tokens_seen": 534315008
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004232798395185557,
      "loss": 3.2178,
      "theoretical_loss": 3.887074191966213,
      "tokens_seen": 534380544
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042326980942828486,
      "loss": 3.051,
      "theoretical_loss": 3.887023436857424,
      "tokens_seen": 534446080
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004232597793380141,
      "loss": 3.0097,
      "theoretical_loss": 3.8869726897144785,
      "tokens_seen": 534511616
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004232497492477432,
      "loss": 3.2939,
      "theoretical_loss": 3.8869219505351476,
      "tokens_seen": 534577152
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042323971915747246,
      "loss": 3.1419,
      "theoretical_loss": 3.8868712193172064,
      "tokens_seen": 534642688
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004232296890672016,
      "loss": 2.9521,
      "theoretical_loss": 3.8868204960584296,
      "tokens_seen": 534708224
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004232196589769308,
      "loss": 2.974,
      "theoretical_loss": 3.8867697807565937,
      "tokens_seen": 534773760
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042320962888666,
      "loss": 3.1353,
      "theoretical_loss": 3.886719073409475,
      "tokens_seen": 534839296
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004231995987963892,
      "loss": 2.9639,
      "theoretical_loss": 3.886668374014852,
      "tokens_seen": 534904832
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042318956870611836,
      "loss": 2.861,
      "theoretical_loss": 3.8866176825705034,
      "tokens_seen": 534970368
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004231795386158476,
      "loss": 2.7527,
      "theoretical_loss": 3.8865669990742084,
      "tokens_seen": 535035904
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004231695085255767,
      "loss": 3.0405,
      "theoretical_loss": 3.8865163235237468,
      "tokens_seen": 535101440
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042315947843530596,
      "loss": 2.8367,
      "theoretical_loss": 3.8864656559169015,
      "tokens_seen": 535166976
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004231494483450351,
      "loss": 2.7405,
      "theoretical_loss": 3.886414996251454,
      "tokens_seen": 535232512
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004231394182547643,
      "loss": 2.895,
      "theoretical_loss": 3.8863643445251874,
      "tokens_seen": 535298048
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004231293881644935,
      "loss": 2.9765,
      "theoretical_loss": 3.8863137007358857,
      "tokens_seen": 535363584
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004231193580742227,
      "loss": 2.8674,
      "theoretical_loss": 3.8862630648813337,
      "tokens_seen": 535429120
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042310932798395186,
      "loss": 2.9129,
      "theoretical_loss": 3.886212436959317,
      "tokens_seen": 535494656
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042309929789368105,
      "loss": 3.1221,
      "theoretical_loss": 3.8861618169676237,
      "tokens_seen": 535560192
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004230892678034102,
      "loss": 3.0467,
      "theoretical_loss": 3.88611120490404,
      "tokens_seen": 535625728
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042307923771313946,
      "loss": 3.0985,
      "theoretical_loss": 3.8860606007663554,
      "tokens_seen": 535691264
    },
    {
      "epoch": 0.16,
      "objective/train/advantage_avg": 0.4930369257926941,
      "objective/train/docs_used": 314102,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.0567221641540527,
      "objective/train/original_loss": 3.0567221641540527,
      "objective/train/theoretical_loss": 3.8860100045523582,
      "objective/train/tokens_used": 556216800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24317625164985657,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505397319793701,
      "objective/train/weighted_lm_loss": 3.211085081100464,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 3.8860100045523582,
      "tokens_seen": 535756800
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004230692076228686,
      "loss": 2.8346,
      "theoretical_loss": 3.8860100045523582,
      "tokens_seen": 535756800
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004230591775325978,
      "loss": 3.0336,
      "theoretical_loss": 3.8859594162598396,
      "tokens_seen": 535822336
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042304914744232695,
      "loss": 2.7539,
      "theoretical_loss": 3.88590883588659,
      "tokens_seen": 535887872
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004230391173520562,
      "loss": 3.1273,
      "theoretical_loss": 3.8858582634304017,
      "tokens_seen": 535953408
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042302908726178537,
      "loss": 3.1073,
      "theoretical_loss": 3.885807698889068,
      "tokens_seen": 536018944
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042301905717151455,
      "loss": 3.2068,
      "theoretical_loss": 3.885757142260382,
      "tokens_seen": 536084480
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042300902708124373,
      "loss": 3.0527,
      "theoretical_loss": 3.8857065935421398,
      "tokens_seen": 536150016
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042299899699097297,
      "loss": 2.9718,
      "theoretical_loss": 3.885656052732135,
      "tokens_seen": 536215552
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004229889669007021,
      "loss": 3.1153,
      "theoretical_loss": 3.8856055198281654,
      "tokens_seen": 536281088
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042297893681043133,
      "loss": 2.9838,
      "theoretical_loss": 3.885554994828028,
      "tokens_seen": 536346624
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042296890672016045,
      "loss": 3.1638,
      "theoretical_loss": 3.885504477729521,
      "tokens_seen": 536412160
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004229588766298897,
      "loss": 3.096,
      "theoretical_loss": 3.8854539685304434,
      "tokens_seen": 536477696
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042294884653961887,
      "loss": 3.2823,
      "theoretical_loss": 3.885403467228595,
      "tokens_seen": 536543232
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042293881644934805,
      "loss": 2.9183,
      "theoretical_loss": 3.885352973821778,
      "tokens_seen": 536608768
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042292878635907723,
      "loss": 3.0064,
      "theoretical_loss": 3.885302488307792,
      "tokens_seen": 536674304
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004229187562688064,
      "loss": 2.9888,
      "theoretical_loss": 3.8852520106844413,
      "tokens_seen": 536739840
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004229087261785356,
      "loss": 2.9234,
      "theoretical_loss": 3.885201540949529,
      "tokens_seen": 536805376
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042289869608826483,
      "loss": 3.1328,
      "theoretical_loss": 3.8851510791008588,
      "tokens_seen": 536870912
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042288866599799396,
      "loss": 3.0803,
      "theoretical_loss": 3.885100625136237,
      "tokens_seen": 536936448
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004228786359077232,
      "loss": 3.0029,
      "theoretical_loss": 3.8850501790534695,
      "tokens_seen": 537001984
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004228686058174523,
      "loss": 2.8566,
      "theoretical_loss": 3.884999740850363,
      "tokens_seen": 537067520
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042285857572718155,
      "loss": 3.1216,
      "theoretical_loss": 3.884949310524725,
      "tokens_seen": 537133056
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042284854563691074,
      "loss": 3.1241,
      "theoretical_loss": 3.884898888074365,
      "tokens_seen": 537198592
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004228385155466399,
      "loss": 2.8513,
      "theoretical_loss": 3.8848484734970925,
      "tokens_seen": 537264128
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004228284854563691,
      "loss": 3.2806,
      "theoretical_loss": 3.884798066790718,
      "tokens_seen": 537329664
    },
    {
      "epoch": 0.16,
      "objective/train/advantage_avg": 0.49554502964019775,
      "objective/train/docs_used": 314596,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.305720567703247,
      "objective/train/original_loss": 2.305720329284668,
      "objective/train/theoretical_loss": 3.884747667953053,
      "objective/train/tokens_used": 557855200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24562275409698486,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0508031845092773,
      "objective/train/weighted_lm_loss": 2.422844886779785,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0442860126495361,
      "theoretical_loss": 3.884747667953053,
      "tokens_seen": 537395200
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042281845536609833,
      "loss": 3.012,
      "theoretical_loss": 3.884747667953053,
      "tokens_seen": 537395200
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042280842527582746,
      "loss": 3.0151,
      "theoretical_loss": 3.8846972769819095,
      "tokens_seen": 537460736
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004227983951855567,
      "loss": 3.2922,
      "theoretical_loss": 3.884646893875101,
      "tokens_seen": 537526272
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004227883650952858,
      "loss": 3.1246,
      "theoretical_loss": 3.884596518630441,
      "tokens_seen": 537591808
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042277833500501506,
      "loss": 3.1174,
      "theoretical_loss": 3.884546151245745,
      "tokens_seen": 537657344
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042276830491474424,
      "loss": 2.8841,
      "theoretical_loss": 3.8844957917188285,
      "tokens_seen": 537722880
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004227582748244734,
      "loss": 3.1274,
      "theoretical_loss": 3.8844454400475086,
      "tokens_seen": 537788416
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004227482447342026,
      "loss": 2.9112,
      "theoretical_loss": 3.8843950962296026,
      "tokens_seen": 537853952
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004227382146439318,
      "loss": 3.2371,
      "theoretical_loss": 3.8843447602629277,
      "tokens_seen": 537919488
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042272818455366096,
      "loss": 3.056,
      "theoretical_loss": 3.884294432145305,
      "tokens_seen": 537985024
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004227181544633902,
      "loss": 3.0624,
      "theoretical_loss": 3.884244111874554,
      "tokens_seen": 538050560
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004227081243731193,
      "loss": 3.1068,
      "theoretical_loss": 3.8841937994484956,
      "tokens_seen": 538116096
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042269809428284856,
      "loss": 3.0069,
      "theoretical_loss": 3.8841434948649507,
      "tokens_seen": 538181632
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004226880641925777,
      "loss": 2.8255,
      "theoretical_loss": 3.884093198121744,
      "tokens_seen": 538247168
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004226780341023069,
      "loss": 2.9604,
      "theoretical_loss": 3.884042909216698,
      "tokens_seen": 538312704
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042266800401203616,
      "loss": 2.8563,
      "theoretical_loss": 3.883992628147637,
      "tokens_seen": 538378240
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004226579739217653,
      "loss": 3.1363,
      "theoretical_loss": 3.883942354912387,
      "tokens_seen": 538443776
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004226479438314945,
      "loss": 3.068,
      "theoretical_loss": 3.883892089508774,
      "tokens_seen": 538509312
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004226379137412237,
      "loss": 2.858,
      "theoretical_loss": 3.8838418319346246,
      "tokens_seen": 538574848
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004226278836509529,
      "loss": 3.0113,
      "theoretical_loss": 3.8837915821877678,
      "tokens_seen": 538640384
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042261785356068206,
      "loss": 3.2063,
      "theoretical_loss": 3.883741340266031,
      "tokens_seen": 538705920
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042260782347041125,
      "loss": 2.8703,
      "theoretical_loss": 3.8836911061672454,
      "tokens_seen": 538771456
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004225977933801404,
      "loss": 3.1841,
      "theoretical_loss": 3.883640879889241,
      "tokens_seen": 538836992
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042258776328986966,
      "loss": 3.1395,
      "theoretical_loss": 3.8835906614298485,
      "tokens_seen": 538902528
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004225777331995988,
      "loss": 3.0711,
      "theoretical_loss": 3.883540450786901,
      "tokens_seen": 538968064
    },
    {
      "epoch": 0.16,
      "objective/train/advantage_avg": 0.49470221996307373,
      "objective/train/docs_used": 315427,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.8916468620300293,
      "objective/train/original_loss": 2.8916471004486084,
      "objective/train/theoretical_loss": 3.8834902479582314,
      "objective/train/tokens_used": 559493600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24479177594184875,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507146120071411,
      "objective/train/weighted_lm_loss": 3.0382230281829834,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0396549701690674,
      "theoretical_loss": 3.8834902479582314,
      "tokens_seen": 539033600
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.000422567703109328,
      "loss": 3.0068,
      "theoretical_loss": 3.8834902479582314,
      "tokens_seen": 539033600
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042255767301905715,
      "loss": 2.9858,
      "theoretical_loss": 3.883440052941674,
      "tokens_seen": 539099136
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004225476429287864,
      "loss": 2.7578,
      "theoretical_loss": 3.8833898657350634,
      "tokens_seen": 539164672
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042253761283851557,
      "loss": 2.9477,
      "theoretical_loss": 3.8833396863362353,
      "tokens_seen": 539230208
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042252758274824475,
      "loss": 3.0208,
      "theoretical_loss": 3.8832895147430264,
      "tokens_seen": 539295744
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042251755265797393,
      "loss": 2.9925,
      "theoretical_loss": 3.8832393509532745,
      "tokens_seen": 539361280
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042250752256770317,
      "loss": 3.0231,
      "theoretical_loss": 3.8831891949648174,
      "tokens_seen": 539426816
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004224974924774323,
      "loss": 2.9934,
      "theoretical_loss": 3.883139046775495,
      "tokens_seen": 539492352
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042248746238716153,
      "loss": 2.6888,
      "theoretical_loss": 3.883088906383146,
      "tokens_seen": 539557888
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042247743229689065,
      "loss": 3.018,
      "theoretical_loss": 3.8830387737856134,
      "tokens_seen": 539623424
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004224674022066199,
      "loss": 2.8639,
      "theoretical_loss": 3.882988648980737,
      "tokens_seen": 539688960
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042245737211634907,
      "loss": 2.9603,
      "theoretical_loss": 3.88293853196636,
      "tokens_seen": 539754496
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042244734202607825,
      "loss": 2.8667,
      "theoretical_loss": 3.8828884227403266,
      "tokens_seen": 539820032
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042243731193580743,
      "loss": 3.083,
      "theoretical_loss": 3.882838321300481,
      "tokens_seen": 539885568
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004224272818455366,
      "loss": 3.1502,
      "theoretical_loss": 3.8827882276446672,
      "tokens_seen": 539951104
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004224172517552658,
      "loss": 3.0521,
      "theoretical_loss": 3.8827381417707327,
      "tokens_seen": 540016640
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042240722166499503,
      "loss": 3.0174,
      "theoretical_loss": 3.8826880636765235,
      "tokens_seen": 540082176
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042239719157472416,
      "loss": 3.1246,
      "theoretical_loss": 3.8826379933598885,
      "tokens_seen": 540147712
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004223871614844534,
      "loss": 3.0275,
      "theoretical_loss": 3.8825879308186755,
      "tokens_seen": 540213248
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004223771313941825,
      "loss": 2.8765,
      "theoretical_loss": 3.8825378760507334,
      "tokens_seen": 540278784
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042236710130391176,
      "loss": 3.095,
      "theoretical_loss": 3.8824878290539138,
      "tokens_seen": 540344320
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042235707121364094,
      "loss": 2.8666,
      "theoretical_loss": 3.8824377898260676,
      "tokens_seen": 540409856
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004223470411233701,
      "loss": 2.9998,
      "theoretical_loss": 3.882387758365046,
      "tokens_seen": 540475392
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004223370110330993,
      "loss": 2.8622,
      "theoretical_loss": 3.8823377346687034,
      "tokens_seen": 540540928
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042232698094282853,
      "loss": 3.0211,
      "theoretical_loss": 3.882287718734892,
      "tokens_seen": 540606464
    },
    {
      "epoch": 0.16,
      "objective/train/advantage_avg": 0.4927524924278259,
      "objective/train/docs_used": 316530,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.347719669342041,
      "objective/train/original_loss": 3.34771990776062,
      "objective/train/theoretical_loss": 3.8822377105614674,
      "objective/train/tokens_used": 561132000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24296723306179047,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050510287284851,
      "objective/train/weighted_lm_loss": 3.516854763031006,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0253151655197144,
      "theoretical_loss": 3.8822377105614674,
      "tokens_seen": 540672000
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042231695085255766,
      "loss": 3.0074,
      "theoretical_loss": 3.8822377105614674,
      "tokens_seen": 540672000
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004223069207622869,
      "loss": 2.9016,
      "theoretical_loss": 3.882187710146285,
      "tokens_seen": 540737536
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.000422296890672016,
      "loss": 3.0161,
      "theoretical_loss": 3.882137717487201,
      "tokens_seen": 540803072
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042228686058174526,
      "loss": 3.1353,
      "theoretical_loss": 3.8820877325820726,
      "tokens_seen": 540868608
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042227683049147444,
      "loss": 2.9111,
      "theoretical_loss": 3.8820377554287573,
      "tokens_seen": 540934144
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004222668004012036,
      "loss": 3.0374,
      "theoretical_loss": 3.881987786025115,
      "tokens_seen": 540999680
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004222567703109328,
      "loss": 3.0635,
      "theoretical_loss": 3.8819378243690044,
      "tokens_seen": 541065216
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.000422246740220662,
      "loss": 2.8764,
      "theoretical_loss": 3.8818878704582866,
      "tokens_seen": 541130752
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042223671013039116,
      "loss": 3.0058,
      "theoretical_loss": 3.8818379242908234,
      "tokens_seen": 541196288
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004222266800401204,
      "loss": 2.7828,
      "theoretical_loss": 3.8817879858644764,
      "tokens_seen": 541261824
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004222166499498495,
      "loss": 2.9576,
      "theoretical_loss": 3.8817380551771095,
      "tokens_seen": 541327360
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042220661985957876,
      "loss": 3.1589,
      "theoretical_loss": 3.881688132226586,
      "tokens_seen": 541392896
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004221965897693079,
      "loss": 2.9911,
      "theoretical_loss": 3.881638217010771,
      "tokens_seen": 541458432
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004221865596790371,
      "loss": 3.0252,
      "theoretical_loss": 3.88158830952753,
      "tokens_seen": 541523968
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004221765295887663,
      "loss": 3.0799,
      "theoretical_loss": 3.88153840977473,
      "tokens_seen": 541589504
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004221664994984955,
      "loss": 2.9249,
      "theoretical_loss": 3.8814885177502374,
      "tokens_seen": 541655040
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042215646940822467,
      "loss": 3.0891,
      "theoretical_loss": 3.8814386334519213,
      "tokens_seen": 541720576
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004221464393179539,
      "loss": 3.0281,
      "theoretical_loss": 3.881388756877651,
      "tokens_seen": 541786112
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042213640922768303,
      "loss": 3.0414,
      "theoretical_loss": 3.8813388880252955,
      "tokens_seen": 541851648
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042212637913741226,
      "loss": 2.9536,
      "theoretical_loss": 3.881289026892726,
      "tokens_seen": 541917184
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004221163490471414,
      "loss": 2.9806,
      "theoretical_loss": 3.881239173477814,
      "tokens_seen": 541982720
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004221063189568706,
      "loss": 3.0446,
      "theoretical_loss": 3.8811893277784324,
      "tokens_seen": 542048256
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004220962888665998,
      "loss": 3.0398,
      "theoretical_loss": 3.881139489792454,
      "tokens_seen": 542113792
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.000422086258776329,
      "loss": 2.7248,
      "theoretical_loss": 3.8810896595177526,
      "tokens_seen": 542179328
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042207622868605817,
      "loss": 2.9005,
      "theoretical_loss": 3.881039836952204,
      "tokens_seen": 542244864
    },
    {
      "epoch": 0.16,
      "objective/train/advantage_avg": 0.49187248945236206,
      "objective/train/docs_used": 317270,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.1898584365844727,
      "objective/train/original_loss": 3.1898584365844727,
      "objective/train/theoretical_loss": 3.880990022093684,
      "objective/train/tokens_used": 562770400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24204303324222565,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0504175424575806,
      "objective/train/weighted_lm_loss": 3.350656032562256,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.044721245765686,
      "theoretical_loss": 3.880990022093684,
      "tokens_seen": 542310400
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042206619859578735,
      "loss": 3.1017,
      "theoretical_loss": 3.880990022093684,
      "tokens_seen": 542310400
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042205616850551653,
      "loss": 2.8656,
      "theoretical_loss": 3.8809402149400682,
      "tokens_seen": 542375936
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042204613841524577,
      "loss": 3.1352,
      "theoretical_loss": 3.880890415489235,
      "tokens_seen": 542441472
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004220361083249749,
      "loss": 3.1243,
      "theoretical_loss": 3.8808406237390622,
      "tokens_seen": 542507008
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042202607823470413,
      "loss": 2.8138,
      "theoretical_loss": 3.8807908396874295,
      "tokens_seen": 542572544
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004220160481444333,
      "loss": 3.0547,
      "theoretical_loss": 3.880741063332217,
      "tokens_seen": 542638080
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004220060180541625,
      "loss": 3.0936,
      "theoretical_loss": 3.880691294671305,
      "tokens_seen": 542703616
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004219959879638917,
      "loss": 2.8482,
      "theoretical_loss": 3.8806415337025753,
      "tokens_seen": 542769152
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042198595787362085,
      "loss": 3.1989,
      "theoretical_loss": 3.880591780423911,
      "tokens_seen": 542834688
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042197592778335004,
      "loss": 3.1956,
      "theoretical_loss": 3.880542034833195,
      "tokens_seen": 542900224
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042196589769307927,
      "loss": 3.1829,
      "theoretical_loss": 3.880492296928312,
      "tokens_seen": 542965760
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004219558676028084,
      "loss": 3.0803,
      "theoretical_loss": 3.880442566707146,
      "tokens_seen": 543031296
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042194583751253763,
      "loss": 3.0402,
      "theoretical_loss": 3.880392844167584,
      "tokens_seen": 543096832
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042193580742226676,
      "loss": 3.1254,
      "theoretical_loss": 3.880343129307512,
      "tokens_seen": 543162368
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.000421925777331996,
      "loss": 3.1958,
      "theoretical_loss": 3.8802934221248178,
      "tokens_seen": 543227904
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042191574724172523,
      "loss": 2.909,
      "theoretical_loss": 3.88024372261739,
      "tokens_seen": 543293440
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042190571715145436,
      "loss": 3.0071,
      "theoretical_loss": 3.8801940307831178,
      "tokens_seen": 543358976
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004218956870611836,
      "loss": 3.2354,
      "theoretical_loss": 3.8801443466198906,
      "tokens_seen": 543424512
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004218856569709127,
      "loss": 3.0898,
      "theoretical_loss": 3.8800946701256005,
      "tokens_seen": 543490048
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042187562688064196,
      "loss": 2.8707,
      "theoretical_loss": 3.880045001298138,
      "tokens_seen": 543555584
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042186559679037114,
      "loss": 2.9823,
      "theoretical_loss": 3.8799953401353964,
      "tokens_seen": 543621120
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004218555667001003,
      "loss": 2.7546,
      "theoretical_loss": 3.8799456866352693,
      "tokens_seen": 543686656
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004218455366098295,
      "loss": 3.0751,
      "theoretical_loss": 3.8798960407956504,
      "tokens_seen": 543752192
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042183550651955873,
      "loss": 2.9303,
      "theoretical_loss": 3.879846402614435,
      "tokens_seen": 543817728
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042182547642928786,
      "loss": 3.1209,
      "theoretical_loss": 3.8797967720895183,
      "tokens_seen": 543883264
    },
    {
      "epoch": 0.16,
      "objective/train/advantage_avg": 0.4950852692127228,
      "objective/train/docs_used": 318285,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.0618860721588135,
      "objective/train/original_loss": 3.0618855953216553,
      "objective/train/theoretical_loss": 3.8797471492187987,
      "objective/train/tokens_used": 564408800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24517139792442322,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507547855377197,
      "objective/train/weighted_lm_loss": 3.217257022857666,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0408108234405518,
      "theoretical_loss": 3.8797471492187987,
      "tokens_seen": 543948800
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004218154463390171,
      "loss": 3.1877,
      "theoretical_loss": 3.8797471492187987,
      "tokens_seen": 543948800
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004218054162487462,
      "loss": 2.9847,
      "theoretical_loss": 3.8796975340001723,
      "tokens_seen": 544014336
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042179538615847546,
      "loss": 3.094,
      "theoretical_loss": 3.879647926431538,
      "tokens_seen": 544079872
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042178535606820464,
      "loss": 3.3256,
      "theoretical_loss": 3.879598326510795,
      "tokens_seen": 544145408
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004217753259779338,
      "loss": 3.2454,
      "theoretical_loss": 3.879548734235843,
      "tokens_seen": 544210944
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.000421765295887663,
      "loss": 3.1626,
      "theoretical_loss": 3.879499149604583,
      "tokens_seen": 544276480
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004217552657973922,
      "loss": 2.996,
      "theoretical_loss": 3.879449572614918,
      "tokens_seen": 544342016
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.00042174523570712136,
      "loss": 3.1873,
      "theoretical_loss": 3.8794000032647484,
      "tokens_seen": 544407552
    },
    {
      "epoch": 0.16,
      "learning_rate": 0.0004217352056168506,
      "loss": 2.8612,
      "theoretical_loss": 3.8793504415519795,
      "tokens_seen": 544473088
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004217251755265797,
      "loss": 3.0889,
      "theoretical_loss": 3.879300887474514,
      "tokens_seen": 544538624
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00042171514543630896,
      "loss": 3.2576,
      "theoretical_loss": 3.8792513410302574,
      "tokens_seen": 544604160
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004217051153460381,
      "loss": 2.8575,
      "theoretical_loss": 3.8792018022171164,
      "tokens_seen": 544669696
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004216950852557673,
      "loss": 3.0926,
      "theoretical_loss": 3.8791522710329964,
      "tokens_seen": 544735232
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004216850551654965,
      "loss": 3.0253,
      "theoretical_loss": 3.8791027474758057,
      "tokens_seen": 544800768
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004216750250752257,
      "loss": 3.2174,
      "theoretical_loss": 3.8790532315434523,
      "tokens_seen": 544866304
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00042166499498495487,
      "loss": 2.8033,
      "theoretical_loss": 3.879003723233846,
      "tokens_seen": 544931840
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004216549648946841,
      "loss": 3.1719,
      "theoretical_loss": 3.8789542225448956,
      "tokens_seen": 544997376
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00042164493480441323,
      "loss": 2.9205,
      "theoretical_loss": 3.878904729474513,
      "tokens_seen": 545062912
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00042163490471414246,
      "loss": 3.1511,
      "theoretical_loss": 3.8788552440206088,
      "tokens_seen": 545128448
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004216248746238716,
      "loss": 3.0188,
      "theoretical_loss": 3.8788057661810966,
      "tokens_seen": 545193984
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004216148445336008,
      "loss": 2.8366,
      "theoretical_loss": 3.878756295953889,
      "tokens_seen": 545259520
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00042160481444333,
      "loss": 3.3281,
      "theoretical_loss": 3.8787068333369,
      "tokens_seen": 545325056
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004215947843530592,
      "loss": 3.233,
      "theoretical_loss": 3.878657378328045,
      "tokens_seen": 545390592
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00042158475426278837,
      "loss": 2.9976,
      "theoretical_loss": 3.878607930925239,
      "tokens_seen": 545456128
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00042157472417251755,
      "loss": 2.986,
      "theoretical_loss": 3.878558491126399,
      "tokens_seen": 545521664
    },
    {
      "epoch": 0.17,
      "objective/train/advantage_avg": 0.49273401498794556,
      "objective/train/docs_used": 318961,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.5888352394104004,
      "objective/train/original_loss": 2.5888350009918213,
      "objective/train/theoretical_loss": 3.878509058929443,
      "objective/train/tokens_used": 566047200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24294516444206238,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505084991455078,
      "objective/train/weighted_lm_loss": 2.7194180488586426,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0363596677780151,
      "theoretical_loss": 3.878509058929443,
      "tokens_seen": 545587200
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00042156469408224673,
      "loss": 3.1423,
      "theoretical_loss": 3.878509058929443,
      "tokens_seen": 545587200
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00042155466399197597,
      "loss": 3.1539,
      "theoretical_loss": 3.8784596343322884,
      "tokens_seen": 545652736
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004215446339017051,
      "loss": 2.9489,
      "theoretical_loss": 3.8784102173328545,
      "tokens_seen": 545718272
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00042153460381143433,
      "loss": 2.958,
      "theoretical_loss": 3.8783608079290603,
      "tokens_seen": 545783808
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004215245737211635,
      "loss": 2.8326,
      "theoretical_loss": 3.8783114061188284,
      "tokens_seen": 545849344
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004215145436308927,
      "loss": 3.1252,
      "theoretical_loss": 3.8782620119000786,
      "tokens_seen": 545914880
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004215045135406219,
      "loss": 2.8556,
      "theoretical_loss": 3.8782126252707334,
      "tokens_seen": 545980416
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00042149448345035105,
      "loss": 2.9516,
      "theoretical_loss": 3.8781632462287163,
      "tokens_seen": 546045952
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00042148445336008024,
      "loss": 2.8861,
      "theoretical_loss": 3.8781138747719517,
      "tokens_seen": 546111488
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00042147442326980947,
      "loss": 3.1199,
      "theoretical_loss": 3.8780645108983633,
      "tokens_seen": 546177024
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004214643931795386,
      "loss": 2.8654,
      "theoretical_loss": 3.8780151546058774,
      "tokens_seen": 546242560
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00042145436308926783,
      "loss": 3.1534,
      "theoretical_loss": 3.87796580589242,
      "tokens_seen": 546308096
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00042144433299899696,
      "loss": 2.9191,
      "theoretical_loss": 3.8779164647559186,
      "tokens_seen": 546373632
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004214343029087262,
      "loss": 2.904,
      "theoretical_loss": 3.877867131194301,
      "tokens_seen": 546439168
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004214242728184554,
      "loss": 3.1395,
      "theoretical_loss": 3.8778178052054963,
      "tokens_seen": 546504704
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00042141424272818456,
      "loss": 2.9645,
      "theoretical_loss": 3.877768486787434,
      "tokens_seen": 546570240
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00042140421263791374,
      "loss": 2.7652,
      "theoretical_loss": 3.877719175938044,
      "tokens_seen": 546635776
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004213941825476429,
      "loss": 3.0399,
      "theoretical_loss": 3.8776698726552583,
      "tokens_seen": 546701312
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004213841524573721,
      "loss": 2.9003,
      "theoretical_loss": 3.877620576937009,
      "tokens_seen": 546766848
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00042137412236710134,
      "loss": 3.1857,
      "theoretical_loss": 3.877571288781229,
      "tokens_seen": 546832384
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00042136409227683046,
      "loss": 3.0993,
      "theoretical_loss": 3.877522008185851,
      "tokens_seen": 546897920
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004213540621865597,
      "loss": 2.9677,
      "theoretical_loss": 3.877472735148811,
      "tokens_seen": 546963456
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004213440320962889,
      "loss": 2.9697,
      "theoretical_loss": 3.8774234696680434,
      "tokens_seen": 547028992
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00042133400200601806,
      "loss": 2.9388,
      "theoretical_loss": 3.8773742117414844,
      "tokens_seen": 547094528
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00042132397191574724,
      "loss": 3.0826,
      "theoretical_loss": 3.877324961367072,
      "tokens_seen": 547160064
    },
    {
      "epoch": 0.17,
      "objective/train/advantage_avg": 0.4951331317424774,
      "objective/train/docs_used": 320353,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.270343065261841,
      "objective/train/original_loss": 3.2703428268432617,
      "objective/train/theoretical_loss": 3.877275718542742,
      "objective/train/tokens_used": 567685600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24522043764591217,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507597923278809,
      "objective/train/weighted_lm_loss": 3.436295509338379,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0448545217514038,
      "theoretical_loss": 3.877275718542742,
      "tokens_seen": 547225600
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004213139418254764,
      "loss": 3.0714,
      "theoretical_loss": 3.877275718542742,
      "tokens_seen": 547225600
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004213039117352056,
      "loss": 2.9092,
      "theoretical_loss": 3.8772264832664347,
      "tokens_seen": 547291136
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00042129388164493484,
      "loss": 2.8121,
      "theoretical_loss": 3.877177255536089,
      "tokens_seen": 547356672
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00042128385155466397,
      "loss": 2.989,
      "theoretical_loss": 3.8771280353496445,
      "tokens_seen": 547422208
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004212738214643932,
      "loss": 3.0472,
      "theoretical_loss": 3.877078822705043,
      "tokens_seen": 547487744
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00042126379137412233,
      "loss": 3.0609,
      "theoretical_loss": 3.877029617600226,
      "tokens_seen": 547553280
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00042125376128385156,
      "loss": 3.1234,
      "theoretical_loss": 3.876980420033136,
      "tokens_seen": 547618816
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00042124373119358075,
      "loss": 3.0062,
      "theoretical_loss": 3.8769312300017167,
      "tokens_seen": 547684352
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004212337011033099,
      "loss": 3.1039,
      "theoretical_loss": 3.876882047503912,
      "tokens_seen": 547749888
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004212236710130391,
      "loss": 3.0565,
      "theoretical_loss": 3.8768328725376673,
      "tokens_seen": 547815424
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004212136409227683,
      "loss": 2.8228,
      "theoretical_loss": 3.8767837051009284,
      "tokens_seen": 547880960
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00042120361083249747,
      "loss": 2.8261,
      "theoretical_loss": 3.876734545191642,
      "tokens_seen": 547946496
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004211935807422267,
      "loss": 3.0382,
      "theoretical_loss": 3.876685392807755,
      "tokens_seen": 548012032
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00042118355065195583,
      "loss": 2.8272,
      "theoretical_loss": 3.876636247947216,
      "tokens_seen": 548077568
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00042117352056168507,
      "loss": 3.2659,
      "theoretical_loss": 3.8765871106079746,
      "tokens_seen": 548143104
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004211634904714143,
      "loss": 3.0238,
      "theoretical_loss": 3.8765379807879796,
      "tokens_seen": 548208640
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00042115346038114343,
      "loss": 2.898,
      "theoretical_loss": 3.8764888584851827,
      "tokens_seen": 548274176
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00042114343029087266,
      "loss": 2.9598,
      "theoretical_loss": 3.876439743697535,
      "tokens_seen": 548339712
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004211334002006018,
      "loss": 3.3988,
      "theoretical_loss": 3.8763906364229888,
      "tokens_seen": 548405248
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00042112337011033103,
      "loss": 3.1093,
      "theoretical_loss": 3.8763415366594964,
      "tokens_seen": 548470784
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004211133400200602,
      "loss": 3.0325,
      "theoretical_loss": 3.8762924444050135,
      "tokens_seen": 548536320
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004211033099297894,
      "loss": 3.0359,
      "theoretical_loss": 3.8762433596574932,
      "tokens_seen": 548601856
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00042109327983951857,
      "loss": 2.8658,
      "theoretical_loss": 3.8761942824148914,
      "tokens_seen": 548667392
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00042108324974924775,
      "loss": 2.8641,
      "theoretical_loss": 3.8761452126751648,
      "tokens_seen": 548732928
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00042107321965897693,
      "loss": 2.8782,
      "theoretical_loss": 3.87609615043627,
      "tokens_seen": 548798464
    },
    {
      "epoch": 0.17,
      "objective/train/advantage_avg": 0.49360597133636475,
      "objective/train/docs_used": 320989,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.8246920108795166,
      "objective/train/original_loss": 2.8246920108795166,
      "objective/train/theoretical_loss": 3.8760470956961655,
      "objective/train/tokens_used": 569324000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24379122257232666,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505999326705933,
      "objective/train/weighted_lm_loss": 2.967517375946045,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0363596677780151,
      "theoretical_loss": 3.8760470956961655,
      "tokens_seen": 548864000
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00042106318956870617,
      "loss": 3.0632,
      "theoretical_loss": 3.8760470956961655,
      "tokens_seen": 548864000
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004210531594784353,
      "loss": 2.9114,
      "theoretical_loss": 3.8759980484528094,
      "tokens_seen": 548929536
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00042104312938816453,
      "loss": 2.8434,
      "theoretical_loss": 3.8759490087041613,
      "tokens_seen": 548995072
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004210330992978937,
      "loss": 3.0187,
      "theoretical_loss": 3.875899976448182,
      "tokens_seen": 549060608
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004210230692076229,
      "loss": 3.0261,
      "theoretical_loss": 3.8758509516828314,
      "tokens_seen": 549126144
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004210130391173521,
      "loss": 3.0781,
      "theoretical_loss": 3.8758019344060726,
      "tokens_seen": 549191680
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00042100300902708125,
      "loss": 2.9312,
      "theoretical_loss": 3.8757529246158677,
      "tokens_seen": 549257216
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00042099297893681044,
      "loss": 3.2115,
      "theoretical_loss": 3.875703922310181,
      "tokens_seen": 549322752
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00042098294884653967,
      "loss": 3.0203,
      "theoretical_loss": 3.875654927486975,
      "tokens_seen": 549388288
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004209729187562688,
      "loss": 2.8534,
      "theoretical_loss": 3.875605940144217,
      "tokens_seen": 549453824
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00042096288866599803,
      "loss": 2.9274,
      "theoretical_loss": 3.875556960279871,
      "tokens_seen": 549519360
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00042095285857572716,
      "loss": 2.6555,
      "theoretical_loss": 3.875507987891905,
      "tokens_seen": 549584896
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004209428284854564,
      "loss": 3.0045,
      "theoretical_loss": 3.8754590229782853,
      "tokens_seen": 549650432
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004209327983951856,
      "loss": 2.7546,
      "theoretical_loss": 3.8754100655369808,
      "tokens_seen": 549715968
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00042092276830491476,
      "loss": 2.995,
      "theoretical_loss": 3.875361115565961,
      "tokens_seen": 549781504
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00042091273821464394,
      "loss": 2.672,
      "theoretical_loss": 3.875312173063195,
      "tokens_seen": 549847040
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004209027081243731,
      "loss": 3.0209,
      "theoretical_loss": 3.875263238026654,
      "tokens_seen": 549912576
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004208926780341023,
      "loss": 2.9307,
      "theoretical_loss": 3.8752143104543086,
      "tokens_seen": 549978112
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00042088264794383154,
      "loss": 3.0271,
      "theoretical_loss": 3.875165390344132,
      "tokens_seen": 550043648
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00042087261785356066,
      "loss": 3.0943,
      "theoretical_loss": 3.8751164776940965,
      "tokens_seen": 550109184
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004208625877632899,
      "loss": 2.8787,
      "theoretical_loss": 3.875067572502177,
      "tokens_seen": 550174720
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004208525576730191,
      "loss": 2.9846,
      "theoretical_loss": 3.875018674766346,
      "tokens_seen": 550240256
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00042084252758274826,
      "loss": 2.9607,
      "theoretical_loss": 3.874969784484581,
      "tokens_seen": 550305792
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00042083249749247744,
      "loss": 3.0394,
      "theoretical_loss": 3.874920901654857,
      "tokens_seen": 550371328
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004208224674022066,
      "loss": 2.9653,
      "theoretical_loss": 3.8748720262751517,
      "tokens_seen": 550436864
    },
    {
      "epoch": 0.17,
      "objective/train/advantage_avg": 0.4933524429798126,
      "objective/train/docs_used": 322374,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.6347970962524414,
      "objective/train/original_loss": 2.6347970962524414,
      "objective/train/theoretical_loss": 3.8748231583434425,
      "objective/train/tokens_used": 570962400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24349209666252136,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505728721618652,
      "objective/train/weighted_lm_loss": 2.7680788040161133,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 3.8748231583434425,
      "tokens_seen": 550502400
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004208124373119358,
      "loss": 3.0599,
      "theoretical_loss": 3.8748231583434425,
      "tokens_seen": 550502400
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00042080240722166504,
      "loss": 3.1271,
      "theoretical_loss": 3.8747742978577078,
      "tokens_seen": 550567936
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00042079237713139417,
      "loss": 3.114,
      "theoretical_loss": 3.874725444815927,
      "tokens_seen": 550633472
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004207823470411234,
      "loss": 2.7127,
      "theoretical_loss": 3.8746765992160808,
      "tokens_seen": 550699008
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00042077231695085253,
      "loss": 2.8221,
      "theoretical_loss": 3.874627761056149,
      "tokens_seen": 550764544
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00042076228686058176,
      "loss": 3.0236,
      "theoretical_loss": 3.874578930334115,
      "tokens_seen": 550830080
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00042075225677031095,
      "loss": 3.2342,
      "theoretical_loss": 3.8745301070479594,
      "tokens_seen": 550895616
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004207422266800401,
      "loss": 3.2699,
      "theoretical_loss": 3.874481291195667,
      "tokens_seen": 550961152
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004207321965897693,
      "loss": 3.1316,
      "theoretical_loss": 3.8744324827752203,
      "tokens_seen": 551026688
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004207221664994985,
      "loss": 3.1727,
      "theoretical_loss": 3.8743836817846065,
      "tokens_seen": 551092224
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00042071213640922767,
      "loss": 2.7155,
      "theoretical_loss": 3.8743348882218083,
      "tokens_seen": 551157760
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004207021063189569,
      "loss": 3.0245,
      "theoretical_loss": 3.874286102084815,
      "tokens_seen": 551223296
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00042069207622868603,
      "loss": 3.0052,
      "theoretical_loss": 3.8742373233716116,
      "tokens_seen": 551288832
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00042068204613841527,
      "loss": 2.8262,
      "theoretical_loss": 3.874188552080187,
      "tokens_seen": 551354368
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00042067201604814445,
      "loss": 3.1855,
      "theoretical_loss": 3.8741397882085304,
      "tokens_seen": 551419904
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00042066198595787363,
      "loss": 3.0055,
      "theoretical_loss": 3.874091031754631,
      "tokens_seen": 551485440
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004206519558676028,
      "loss": 2.9302,
      "theoretical_loss": 3.8740422827164784,
      "tokens_seen": 551550976
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.000420641925777332,
      "loss": 3.1258,
      "theoretical_loss": 3.873993541092065,
      "tokens_seen": 551616512
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004206318956870612,
      "loss": 2.7946,
      "theoretical_loss": 3.873944806879382,
      "tokens_seen": 551682048
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004206218655967904,
      "loss": 2.9359,
      "theoretical_loss": 3.873896080076422,
      "tokens_seen": 551747584
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00042061183550651954,
      "loss": 3.0963,
      "theoretical_loss": 3.8738473606811787,
      "tokens_seen": 551813120
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00042060180541624877,
      "loss": 2.8685,
      "theoretical_loss": 3.873798648691647,
      "tokens_seen": 551878656
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004205917753259779,
      "loss": 2.9771,
      "theoretical_loss": 3.8737499441058207,
      "tokens_seen": 551944192
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00042058174523570713,
      "loss": 3.0732,
      "theoretical_loss": 3.8737012469216965,
      "tokens_seen": 552009728
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004205717151454363,
      "loss": 2.9289,
      "theoretical_loss": 3.8736525571372704,
      "tokens_seen": 552075264
    },
    {
      "epoch": 0.17,
      "objective/train/advantage_avg": 0.4946116507053375,
      "objective/train/docs_used": 322945,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.0244951248168945,
      "objective/train/original_loss": 3.0244951248168945,
      "objective/train/theoretical_loss": 3.8736038747505406,
      "objective/train/tokens_used": 572600800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24470281600952148,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050705075263977,
      "objective/train/weighted_lm_loss": 3.1777842044830322,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0441278219223022,
      "theoretical_loss": 3.8736038747505406,
      "tokens_seen": 552140800
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004205616850551655,
      "loss": 2.9912,
      "theoretical_loss": 3.8736038747505406,
      "tokens_seen": 552140800
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004205516549648947,
      "loss": 2.8941,
      "theoretical_loss": 3.8735551997595046,
      "tokens_seen": 552206336
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004205416248746239,
      "loss": 2.96,
      "theoretical_loss": 3.8735065321621613,
      "tokens_seen": 552271872
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00042053159478435304,
      "loss": 2.9056,
      "theoretical_loss": 3.8734578719565107,
      "tokens_seen": 552337408
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004205215646940823,
      "loss": 2.8413,
      "theoretical_loss": 3.873409219140554,
      "tokens_seen": 552402944
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004205115346038114,
      "loss": 3.0182,
      "theoretical_loss": 3.873360573712291,
      "tokens_seen": 552468480
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00042050150451354064,
      "loss": 2.9274,
      "theoretical_loss": 3.873311935669725,
      "tokens_seen": 552534016
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004204914744232698,
      "loss": 3.0835,
      "theoretical_loss": 3.873263305010858,
      "tokens_seen": 552599552
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.000420481444332999,
      "loss": 2.8877,
      "theoretical_loss": 3.8732146817336934,
      "tokens_seen": 552665088
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004204714142427282,
      "loss": 2.5959,
      "theoretical_loss": 3.873166065836237,
      "tokens_seen": 552730624
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00042046138415245736,
      "loss": 2.96,
      "theoretical_loss": 3.873117457316493,
      "tokens_seen": 552796160
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00042045135406218654,
      "loss": 3.0053,
      "theoretical_loss": 3.8730688561724667,
      "tokens_seen": 552861696
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004204413239719158,
      "loss": 3.229,
      "theoretical_loss": 3.873020262402166,
      "tokens_seen": 552927232
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004204312938816449,
      "loss": 3.2684,
      "theoretical_loss": 3.872971676003598,
      "tokens_seen": 552992768
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00042042126379137414,
      "loss": 2.8587,
      "theoretical_loss": 3.872923096974771,
      "tokens_seen": 553058304
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004204112337011033,
      "loss": 3.1205,
      "theoretical_loss": 3.872874525313694,
      "tokens_seen": 553123840
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004204012036108325,
      "loss": 2.986,
      "theoretical_loss": 3.872825961018376,
      "tokens_seen": 553189376
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00042039117352056174,
      "loss": 2.8866,
      "theoretical_loss": 3.872777404086829,
      "tokens_seen": 553254912
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00042038114343029086,
      "loss": 2.9296,
      "theoretical_loss": 3.8727288545170637,
      "tokens_seen": 553320448
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004203711133400201,
      "loss": 3.0009,
      "theoretical_loss": 3.872680312307092,
      "tokens_seen": 553385984
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004203610832497493,
      "loss": 2.9351,
      "theoretical_loss": 3.8726317774549273,
      "tokens_seen": 553451520
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00042035105315947846,
      "loss": 3.1568,
      "theoretical_loss": 3.8725832499585824,
      "tokens_seen": 553517056
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00042034102306920764,
      "loss": 2.896,
      "theoretical_loss": 3.872534729816073,
      "tokens_seen": 553582592
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004203309929789368,
      "loss": 3.3245,
      "theoretical_loss": 3.872486217025413,
      "tokens_seen": 553648128
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.000420320962888666,
      "loss": 3.0246,
      "theoretical_loss": 3.8724377115846194,
      "tokens_seen": 553713664
    },
    {
      "epoch": 0.17,
      "objective/train/advantage_avg": 0.4953135848045349,
      "objective/train/docs_used": 324208,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.9541215896606445,
      "objective/train/original_loss": 2.9541211128234863,
      "objective/train/theoretical_loss": 3.872389213491709,
      "objective/train/tokens_used": 574239200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24537904560565948,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050778865814209,
      "objective/train/weighted_lm_loss": 3.104098320007324,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0408108234405518,
      "theoretical_loss": 3.872389213491709,
      "tokens_seen": 553779200
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00042031093279839524,
      "loss": 2.8099,
      "theoretical_loss": 3.872389213491709,
      "tokens_seen": 553779200
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00042030090270812437,
      "loss": 2.9416,
      "theoretical_loss": 3.872340722744698,
      "tokens_seen": 553844736
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004202908726178536,
      "loss": 3.2105,
      "theoretical_loss": 3.872292239341606,
      "tokens_seen": 553910272
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00042028084252758273,
      "loss": 3.0991,
      "theoretical_loss": 3.872243763280452,
      "tokens_seen": 553975808
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00042027081243731196,
      "loss": 2.8544,
      "theoretical_loss": 3.872195294559255,
      "tokens_seen": 554041344
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00042026078234704115,
      "loss": 3.1658,
      "theoretical_loss": 3.8721468331760365,
      "tokens_seen": 554106880
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004202507522567703,
      "loss": 2.9818,
      "theoretical_loss": 3.872098379128817,
      "tokens_seen": 554172416
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004202407221664995,
      "loss": 2.704,
      "theoretical_loss": 3.87204993241562,
      "tokens_seen": 554237952
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004202306920762287,
      "loss": 2.9952,
      "theoretical_loss": 3.8720014930344666,
      "tokens_seen": 554303488
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00042022066198595787,
      "loss": 2.8992,
      "theoretical_loss": 3.8719530609833814,
      "tokens_seen": 554369024
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004202106318956871,
      "loss": 2.7896,
      "theoretical_loss": 3.8719046362603895,
      "tokens_seen": 554434560
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00042020060180541623,
      "loss": 3.0478,
      "theoretical_loss": 3.871856218863515,
      "tokens_seen": 554500096
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00042019057171514547,
      "loss": 3.1611,
      "theoretical_loss": 3.8718078087907846,
      "tokens_seen": 554565632
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00042018054162487465,
      "loss": 3.037,
      "theoretical_loss": 3.8717594060402245,
      "tokens_seen": 554631168
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00042017051153460383,
      "loss": 3.0876,
      "theoretical_loss": 3.8717110106098627,
      "tokens_seen": 554696704
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.000420160481444333,
      "loss": 3.0382,
      "theoretical_loss": 3.871662622497727,
      "tokens_seen": 554762240
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004201504513540622,
      "loss": 3.2373,
      "theoretical_loss": 3.871614241701846,
      "tokens_seen": 554827776
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004201404212637914,
      "loss": 2.975,
      "theoretical_loss": 3.8715658682202507,
      "tokens_seen": 554893312
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004201303911735206,
      "loss": 2.7827,
      "theoretical_loss": 3.8715175020509713,
      "tokens_seen": 554958848
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00042012036108324974,
      "loss": 3.0798,
      "theoretical_loss": 3.8714691431920385,
      "tokens_seen": 555024384
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00042011033099297897,
      "loss": 3.1617,
      "theoretical_loss": 3.871420791641485,
      "tokens_seen": 555089920
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004201003009027081,
      "loss": 3.0535,
      "theoretical_loss": 3.871372447397343,
      "tokens_seen": 555155456
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00042009027081243733,
      "loss": 2.9323,
      "theoretical_loss": 3.871324110457647,
      "tokens_seen": 555220992
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004200802407221665,
      "loss": 2.78,
      "theoretical_loss": 3.8712757808204303,
      "tokens_seen": 555286528
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004200702106318957,
      "loss": 2.6274,
      "theoretical_loss": 3.8712274584837294,
      "tokens_seen": 555352064
    },
    {
      "epoch": 0.17,
      "objective/train/advantage_avg": 0.49460819363594055,
      "objective/train/docs_used": 324930,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.498317003250122,
      "objective/train/original_loss": 2.498316526412964,
      "objective/train/theoretical_loss": 3.8711791434455787,
      "objective/train/tokens_used": 575877600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24470669031143188,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507047176361084,
      "objective/train/weighted_lm_loss": 2.624704599380493,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0363596677780151,
      "theoretical_loss": 3.8711791434455787,
      "tokens_seen": 555417600
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004200601805416249,
      "loss": 2.8241,
      "theoretical_loss": 3.8711791434455787,
      "tokens_seen": 555417600
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004200501504513541,
      "loss": 2.8434,
      "theoretical_loss": 3.8711308357040157,
      "tokens_seen": 555483136
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00042004012036108324,
      "loss": 2.7509,
      "theoretical_loss": 3.8710825352570772,
      "tokens_seen": 555548672
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004200300902708125,
      "loss": 2.9641,
      "theoretical_loss": 3.8710342421028026,
      "tokens_seen": 555614208
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004200200601805416,
      "loss": 2.9379,
      "theoretical_loss": 3.8709859562392293,
      "tokens_seen": 555679744
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00042001003009027084,
      "loss": 3.1364,
      "theoretical_loss": 3.870937677664398,
      "tokens_seen": 555745280
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00042,
      "loss": 2.9057,
      "theoretical_loss": 3.8708894063763486,
      "tokens_seen": 555810816
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004199899699097292,
      "loss": 3.1932,
      "theoretical_loss": 3.8708411423731226,
      "tokens_seen": 555876352
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004199799398194584,
      "loss": 2.7955,
      "theoretical_loss": 3.870792885652762,
      "tokens_seen": 555941888
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041996990972918756,
      "loss": 2.8051,
      "theoretical_loss": 3.8707446362133098,
      "tokens_seen": 556007424
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041995987963891674,
      "loss": 3.114,
      "theoretical_loss": 3.870696394052808,
      "tokens_seen": 556072960
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.000419949849548646,
      "loss": 3.1572,
      "theoretical_loss": 3.8706481591693027,
      "tokens_seen": 556138496
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004199398194583751,
      "loss": 2.7362,
      "theoretical_loss": 3.8705999315608377,
      "tokens_seen": 556204032
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041992978936810434,
      "loss": 2.5866,
      "theoretical_loss": 3.8705517112254593,
      "tokens_seen": 556269568
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041991975927783347,
      "loss": 3.0657,
      "theoretical_loss": 3.870503498161214,
      "tokens_seen": 556335104
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004199097291875627,
      "loss": 3.0938,
      "theoretical_loss": 3.870455292366148,
      "tokens_seen": 556400640
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004198996990972919,
      "loss": 3.2731,
      "theoretical_loss": 3.8704070938383106,
      "tokens_seen": 556466176
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041988966900702106,
      "loss": 3.2666,
      "theoretical_loss": 3.87035890257575,
      "tokens_seen": 556531712
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041987963891675025,
      "loss": 2.7896,
      "theoretical_loss": 3.870310718576516,
      "tokens_seen": 556597248
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004198696088264795,
      "loss": 3.2962,
      "theoretical_loss": 3.8702625418386587,
      "tokens_seen": 556662784
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004198595787362086,
      "loss": 3.0454,
      "theoretical_loss": 3.8702143723602287,
      "tokens_seen": 556728320
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041984954864593784,
      "loss": 2.7526,
      "theoretical_loss": 3.870166210139278,
      "tokens_seen": 556793856
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041983951855566697,
      "loss": 2.8453,
      "theoretical_loss": 3.87011805517386,
      "tokens_seen": 556859392
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004198294884653962,
      "loss": 2.8182,
      "theoretical_loss": 3.8700699074620264,
      "tokens_seen": 556924928
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004198194583751254,
      "loss": 3.1013,
      "theoretical_loss": 3.870021767001832,
      "tokens_seen": 556990464
    },
    {
      "epoch": 0.17,
      "objective/train/advantage_avg": 0.4950423240661621,
      "objective/train/docs_used": 326252,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.8855807781219482,
      "objective/train/original_loss": 2.885580539703369,
      "objective/train/theoretical_loss": 3.869973633791332,
      "objective/train/tokens_used": 577516000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24511033296585083,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507502555847168,
      "objective/train/weighted_lm_loss": 3.032127618789673,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0338950157165527,
      "theoretical_loss": 3.869973633791332,
      "tokens_seen": 557056000
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041980942828485457,
      "loss": 2.9375,
      "theoretical_loss": 3.869973633791332,
      "tokens_seen": 557056000
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041979939819458375,
      "loss": 2.9705,
      "theoretical_loss": 3.869925507828582,
      "tokens_seen": 557121536
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041978936810431293,
      "loss": 3.1797,
      "theoretical_loss": 3.8698773891116374,
      "tokens_seen": 557187072
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004197793380140421,
      "loss": 3.1607,
      "theoretical_loss": 3.8698292776385554,
      "tokens_seen": 557252608
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041976930792377135,
      "loss": 2.9489,
      "theoretical_loss": 3.869781173407394,
      "tokens_seen": 557318144
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041975927783350047,
      "loss": 3.0428,
      "theoretical_loss": 3.869733076416212,
      "tokens_seen": 557383680
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004197492477432297,
      "loss": 3.0038,
      "theoretical_loss": 3.869684986663068,
      "tokens_seen": 557449216
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041973921765295883,
      "loss": 2.8043,
      "theoretical_loss": 3.869636904146023,
      "tokens_seen": 557514752
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041972918756268807,
      "loss": 2.6081,
      "theoretical_loss": 3.8695888288631366,
      "tokens_seen": 557580288
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041971915747241725,
      "loss": 2.9039,
      "theoretical_loss": 3.8695407608124714,
      "tokens_seen": 557645824
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041970912738214643,
      "loss": 2.9866,
      "theoretical_loss": 3.869492699992089,
      "tokens_seen": 557711360
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004196990972918756,
      "loss": 3.0275,
      "theoretical_loss": 3.8694446464000523,
      "tokens_seen": 557776896
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041968906720160485,
      "loss": 3.0258,
      "theoretical_loss": 3.8693966000344253,
      "tokens_seen": 557842432
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.000419679037111334,
      "loss": 3.022,
      "theoretical_loss": 3.8693485608932727,
      "tokens_seen": 557907968
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004196690070210632,
      "loss": 2.6696,
      "theoretical_loss": 3.8693005289746596,
      "tokens_seen": 557973504
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004196589769307924,
      "loss": 2.6698,
      "theoretical_loss": 3.869252504276652,
      "tokens_seen": 558039040
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004196489468405216,
      "loss": 3.0815,
      "theoretical_loss": 3.869204486797316,
      "tokens_seen": 558104576
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004196389167502508,
      "loss": 2.848,
      "theoretical_loss": 3.8691564765347204,
      "tokens_seen": 558170112
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041962888665997994,
      "loss": 3.0782,
      "theoretical_loss": 3.869108473486932,
      "tokens_seen": 558235648
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041961885656970917,
      "loss": 2.6938,
      "theoretical_loss": 3.869060477652021,
      "tokens_seen": 558301184
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004196088264794383,
      "loss": 3.0977,
      "theoretical_loss": 3.8690124890280564,
      "tokens_seen": 558366720
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041959879638916753,
      "loss": 2.9158,
      "theoretical_loss": 3.8689645076131085,
      "tokens_seen": 558432256
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004195887662988967,
      "loss": 2.9624,
      "theoretical_loss": 3.8689165334052493,
      "tokens_seen": 558497792
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004195787362086259,
      "loss": 2.8283,
      "theoretical_loss": 3.8688685664025497,
      "tokens_seen": 558563328
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004195687061183551,
      "loss": 2.718,
      "theoretical_loss": 3.8688206066030837,
      "tokens_seen": 558628864
    },
    {
      "epoch": 0.17,
      "objective/train/advantage_avg": 0.4935312569141388,
      "objective/train/docs_used": 326886,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.5030667781829834,
      "objective/train/original_loss": 2.5030665397644043,
      "objective/train/theoretical_loss": 3.8687726540049234,
      "objective/train/tokens_used": 579154400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24369119107723236,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505918264389038,
      "objective/train/weighted_lm_loss": 2.629786968231201,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0304546356201172,
      "theoretical_loss": 3.8687726540049234,
      "tokens_seen": 558694400
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004195586760280843,
      "loss": 2.9153,
      "theoretical_loss": 3.8687726540049234,
      "tokens_seen": 558694400
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041954864593781344,
      "loss": 2.8178,
      "theoretical_loss": 3.868724708606144,
      "tokens_seen": 558759936
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004195386158475427,
      "loss": 3.0126,
      "theoretical_loss": 3.8686767704048197,
      "tokens_seen": 558825472
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004195285857572718,
      "loss": 3.0559,
      "theoretical_loss": 3.868628839399026,
      "tokens_seen": 558891008
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041951855566700104,
      "loss": 2.8787,
      "theoretical_loss": 3.86858091558684,
      "tokens_seen": 558956544
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004195085255767302,
      "loss": 2.9876,
      "theoretical_loss": 3.8685329989663386,
      "tokens_seen": 559022080
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004194984954864594,
      "loss": 3.1,
      "theoretical_loss": 3.868485089535599,
      "tokens_seen": 559087616
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004194884653961886,
      "loss": 3.1539,
      "theoretical_loss": 3.8684371872927015,
      "tokens_seen": 559153152
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041947843530591776,
      "loss": 3.1217,
      "theoretical_loss": 3.8683892922357233,
      "tokens_seen": 559218688
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041946840521564694,
      "loss": 2.7833,
      "theoretical_loss": 3.868341404362746,
      "tokens_seen": 559284224
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004194583751253762,
      "loss": 3.0528,
      "theoretical_loss": 3.8682935236718494,
      "tokens_seen": 559349760
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004194483450351053,
      "loss": 2.9677,
      "theoretical_loss": 3.868245650161116,
      "tokens_seen": 559415296
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041943831494483454,
      "loss": 2.9657,
      "theoretical_loss": 3.8681977838286272,
      "tokens_seen": 559480832
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041942828485456367,
      "loss": 2.9046,
      "theoretical_loss": 3.868149924672467,
      "tokens_seen": 559546368
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004194182547642929,
      "loss": 3.0638,
      "theoretical_loss": 3.8681020726907187,
      "tokens_seen": 559611904
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004194082246740221,
      "loss": 2.9064,
      "theoretical_loss": 3.868054227881466,
      "tokens_seen": 559677440
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041939819458375126,
      "loss": 3.0134,
      "theoretical_loss": 3.8680063902427957,
      "tokens_seen": 559742976
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041938816449348045,
      "loss": 3.066,
      "theoretical_loss": 3.8679585597727923,
      "tokens_seen": 559808512
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004193781344032097,
      "loss": 2.7273,
      "theoretical_loss": 3.8679107364695433,
      "tokens_seen": 559874048
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004193681043129388,
      "loss": 3.0031,
      "theoretical_loss": 3.8678629203311368,
      "tokens_seen": 559939584
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041935807422266804,
      "loss": 3.0087,
      "theoretical_loss": 3.867815111355659,
      "tokens_seen": 560005120
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041934804413239717,
      "loss": 2.9559,
      "theoretical_loss": 3.8677673095412013,
      "tokens_seen": 560070656
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004193380140421264,
      "loss": 2.9591,
      "theoretical_loss": 3.867719514885851,
      "tokens_seen": 560136192
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004193279839518556,
      "loss": 2.859,
      "theoretical_loss": 3.8676717273877,
      "tokens_seen": 560201728
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041931795386158477,
      "loss": 3.0974,
      "theoretical_loss": 3.867623947044839,
      "tokens_seen": 560267264
    },
    {
      "epoch": 0.17,
      "objective/train/advantage_avg": 0.4947275221347809,
      "objective/train/docs_used": 327930,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.5330898761749268,
      "objective/train/original_loss": 2.5330893993377686,
      "objective/train/theoretical_loss": 3.8675761738553596,
      "objective/train/tokens_used": 580792800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24480664730072021,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507171154022217,
      "objective/train/weighted_lm_loss": 2.661428689956665,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.045280933380127,
      "theoretical_loss": 3.8675761738553596,
      "tokens_seen": 560332800
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041930792377131395,
      "loss": 2.8652,
      "theoretical_loss": 3.8675761738553596,
      "tokens_seen": 560332800
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041929789368104313,
      "loss": 2.9805,
      "theoretical_loss": 3.8675284078173546,
      "tokens_seen": 560398336
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004192878635907723,
      "loss": 2.9783,
      "theoretical_loss": 3.8674806489289173,
      "tokens_seen": 560463872
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041927783350050155,
      "loss": 2.9044,
      "theoretical_loss": 3.8674328971881415,
      "tokens_seen": 560529408
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041926780341023067,
      "loss": 3.0653,
      "theoretical_loss": 3.867385152593122,
      "tokens_seen": 560594944
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004192577733199599,
      "loss": 2.9685,
      "theoretical_loss": 3.8673374151419546,
      "tokens_seen": 560660480
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041924774322968904,
      "loss": 3.0656,
      "theoretical_loss": 3.867289684832736,
      "tokens_seen": 560726016
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041923771313941827,
      "loss": 2.7676,
      "theoretical_loss": 3.867241961663562,
      "tokens_seen": 560791552
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041922768304914745,
      "loss": 2.9643,
      "theoretical_loss": 3.867194245632531,
      "tokens_seen": 560857088
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041921765295887663,
      "loss": 3.0973,
      "theoretical_loss": 3.867146536737741,
      "tokens_seen": 560922624
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004192076228686058,
      "loss": 3.1132,
      "theoretical_loss": 3.8670988349772912,
      "tokens_seen": 560988160
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041919759277833505,
      "loss": 3.3139,
      "theoretical_loss": 3.8670511403492824,
      "tokens_seen": 561053696
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004191875626880642,
      "loss": 2.8955,
      "theoretical_loss": 3.8670034528518142,
      "tokens_seen": 561119232
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004191775325977934,
      "loss": 2.7157,
      "theoretical_loss": 3.866955772482988,
      "tokens_seen": 561184768
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041916750250752254,
      "loss": 2.9423,
      "theoretical_loss": 3.866908099240906,
      "tokens_seen": 561250304
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004191574724172518,
      "loss": 3.0217,
      "theoretical_loss": 3.866860433123671,
      "tokens_seen": 561315840
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041914744232698095,
      "loss": 2.8417,
      "theoretical_loss": 3.8668127741293867,
      "tokens_seen": 561381376
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041913741223671014,
      "loss": 2.7651,
      "theoretical_loss": 3.866765122256158,
      "tokens_seen": 561446912
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004191273821464393,
      "loss": 3.1227,
      "theoretical_loss": 3.866717477502088,
      "tokens_seen": 561512448
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004191173520561685,
      "loss": 3.1288,
      "theoretical_loss": 3.866669839865284,
      "tokens_seen": 561577984
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004191073219658977,
      "loss": 3.1238,
      "theoretical_loss": 3.866622209343851,
      "tokens_seen": 561643520
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004190972918756269,
      "loss": 2.994,
      "theoretical_loss": 3.866574585935898,
      "tokens_seen": 561709056
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041908726178535604,
      "loss": 3.009,
      "theoretical_loss": 3.8665269696395312,
      "tokens_seen": 561774592
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004190772316950853,
      "loss": 3.1299,
      "theoretical_loss": 3.8664793604528604,
      "tokens_seen": 561840128
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004190672016048144,
      "loss": 3.1659,
      "theoretical_loss": 3.8664317583739938,
      "tokens_seen": 561905664
    },
    {
      "epoch": 0.17,
      "objective/train/advantage_avg": 0.49459919333457947,
      "objective/train/docs_used": 328746,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.746138572692871,
      "objective/train/original_loss": 2.746138572692871,
      "objective/train/theoretical_loss": 3.866384163401042,
      "objective/train/tokens_used": 582431200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24468249082565308,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507036447525024,
      "objective/train/weighted_lm_loss": 2.8851466178894043,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 3.866384163401042,
      "tokens_seen": 561971200
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041905717151454364,
      "loss": 2.996,
      "theoretical_loss": 3.866384163401042,
      "tokens_seen": 561971200
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004190471414242728,
      "loss": 2.9076,
      "theoretical_loss": 3.8663365755321157,
      "tokens_seen": 562036736
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.000419037111334002,
      "loss": 3.0782,
      "theoretical_loss": 3.866288994765327,
      "tokens_seen": 562102272
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004190270812437312,
      "loss": 2.6901,
      "theoretical_loss": 3.866241421098787,
      "tokens_seen": 562167808
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004190170511534604,
      "loss": 2.8932,
      "theoretical_loss": 3.8661938545306094,
      "tokens_seen": 562233344
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041900702106318954,
      "loss": 3.0452,
      "theoretical_loss": 3.8661462950589076,
      "tokens_seen": 562298880
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004189969909729188,
      "loss": 3.034,
      "theoretical_loss": 3.8660987426817957,
      "tokens_seen": 562364416
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004189869608826479,
      "loss": 3.2198,
      "theoretical_loss": 3.866051197397389,
      "tokens_seen": 562429952
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041897693079237714,
      "loss": 2.8731,
      "theoretical_loss": 3.8660036592038036,
      "tokens_seen": 562495488
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004189669007021063,
      "loss": 2.9207,
      "theoretical_loss": 3.865956128099155,
      "tokens_seen": 562561024
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004189568706118355,
      "loss": 2.9114,
      "theoretical_loss": 3.865908604081562,
      "tokens_seen": 562626560
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004189468405215647,
      "loss": 2.7835,
      "theoretical_loss": 3.8658610871491415,
      "tokens_seen": 562692096
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041893681043129387,
      "loss": 3.4689,
      "theoretical_loss": 3.8658135773000124,
      "tokens_seen": 562757632
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041892678034102305,
      "loss": 2.9525,
      "theoretical_loss": 3.8657660745322944,
      "tokens_seen": 562823168
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004189167502507523,
      "loss": 3.009,
      "theoretical_loss": 3.865718578844107,
      "tokens_seen": 562888704
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041890672016048146,
      "loss": 3.2584,
      "theoretical_loss": 3.8656710902335716,
      "tokens_seen": 562954240
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041889669007021065,
      "loss": 2.9709,
      "theoretical_loss": 3.8656236086988094,
      "tokens_seen": 563019776
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004188866599799399,
      "loss": 3.0953,
      "theoretical_loss": 3.865576134237943,
      "tokens_seen": 563085312
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.000418876629889669,
      "loss": 2.9152,
      "theoretical_loss": 3.865528666849095,
      "tokens_seen": 563150848
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041886659979939824,
      "loss": 2.7827,
      "theoretical_loss": 3.8654812065303896,
      "tokens_seen": 563216384
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041885656970912737,
      "loss": 2.893,
      "theoretical_loss": 3.8654337532799508,
      "tokens_seen": 563281920
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004188465396188566,
      "loss": 2.8858,
      "theoretical_loss": 3.865386307095904,
      "tokens_seen": 563347456
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004188365095285858,
      "loss": 3.1107,
      "theoretical_loss": 3.8653388679763747,
      "tokens_seen": 563412992
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041882647943831497,
      "loss": 2.7649,
      "theoretical_loss": 3.86529143591949,
      "tokens_seen": 563478528
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041881644934804415,
      "loss": 3.0673,
      "theoretical_loss": 3.8652440109233765,
      "tokens_seen": 563544064
    },
    {
      "epoch": 0.17,
      "objective/train/advantage_avg": 0.49563688039779663,
      "objective/train/docs_used": 329425,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.7309389114379883,
      "objective/train/original_loss": 2.730938673019409,
      "objective/train/theoretical_loss": 3.8651965929861625,
      "objective/train/tokens_used": 584069600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24570471048355103,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0508127212524414,
      "objective/train/weighted_lm_loss": 2.869701385498047,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0425469875335693,
      "theoretical_loss": 3.8651965929861625,
      "tokens_seen": 563609600
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041880641925777333,
      "loss": 3.035,
      "theoretical_loss": 3.8651965929861625,
      "tokens_seen": 563609600
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004187963891675025,
      "loss": 2.8463,
      "theoretical_loss": 3.865149182105977,
      "tokens_seen": 563675136
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041878635907723175,
      "loss": 2.842,
      "theoretical_loss": 3.865101778280949,
      "tokens_seen": 563740672
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004187763289869609,
      "loss": 3.0313,
      "theoretical_loss": 3.8650543815092098,
      "tokens_seen": 563806208
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004187662988966901,
      "loss": 3.0726,
      "theoretical_loss": 3.865006991788888,
      "tokens_seen": 563871744
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041875626880641924,
      "loss": 3.1243,
      "theoretical_loss": 3.864959609118117,
      "tokens_seen": 563937280
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041874623871614847,
      "loss": 3.2367,
      "theoretical_loss": 3.864912233495029,
      "tokens_seen": 564002816
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041873620862587765,
      "loss": 2.8232,
      "theoretical_loss": 3.864864864917756,
      "tokens_seen": 564068352
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041872617853560683,
      "loss": 3.2024,
      "theoretical_loss": 3.8648175033844323,
      "tokens_seen": 564133888
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.000418716148445336,
      "loss": 3.0316,
      "theoretical_loss": 3.864770148893192,
      "tokens_seen": 564199424
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041870611835506525,
      "loss": 3.0233,
      "theoretical_loss": 3.864722801442171,
      "tokens_seen": 564264960
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004186960882647944,
      "loss": 3.2656,
      "theoretical_loss": 3.864675461029504,
      "tokens_seen": 564330496
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004186860581745236,
      "loss": 2.9737,
      "theoretical_loss": 3.8646281276533285,
      "tokens_seen": 564396032
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041867602808425274,
      "loss": 3.1754,
      "theoretical_loss": 3.864580801311781,
      "tokens_seen": 564461568
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.000418665997993982,
      "loss": 2.9653,
      "theoretical_loss": 3.864533482003,
      "tokens_seen": 564527104
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041865596790371115,
      "loss": 3.0605,
      "theoretical_loss": 3.864486169725124,
      "tokens_seen": 564592640
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041864593781344034,
      "loss": 3.052,
      "theoretical_loss": 3.8644388644762926,
      "tokens_seen": 564658176
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004186359077231695,
      "loss": 2.7645,
      "theoretical_loss": 3.8643915662546453,
      "tokens_seen": 564723712
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004186258776328987,
      "loss": 2.9174,
      "theoretical_loss": 3.8643442750583237,
      "tokens_seen": 564789248
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004186158475426279,
      "loss": 3.185,
      "theoretical_loss": 3.8642969908854683,
      "tokens_seen": 564854784
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004186058174523571,
      "loss": 3.2687,
      "theoretical_loss": 3.864249713734222,
      "tokens_seen": 564920320
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041859578736208624,
      "loss": 3.1636,
      "theoretical_loss": 3.8642024436027276,
      "tokens_seen": 564985856
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004185857572718155,
      "loss": 2.9412,
      "theoretical_loss": 3.8641551804891288,
      "tokens_seen": 565051392
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004185757271815446,
      "loss": 3.0002,
      "theoretical_loss": 3.86410792439157,
      "tokens_seen": 565116928
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041856569709127384,
      "loss": 3.0128,
      "theoretical_loss": 3.8640606753081954,
      "tokens_seen": 565182464
    },
    {
      "epoch": 0.17,
      "objective/train/advantage_avg": 0.49237358570098877,
      "objective/train/docs_used": 329612,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.180604934692383,
      "objective/train/original_loss": 3.1806042194366455,
      "objective/train/theoretical_loss": 3.864013433237152,
      "objective/train/tokens_used": 585708000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24249307811260223,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0504698753356934,
      "objective/train/weighted_lm_loss": 3.3411705493927,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.044721245765686,
      "theoretical_loss": 3.864013433237152,
      "tokens_seen": 565248000
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.000418555667001003,
      "loss": 3.1742,
      "theoretical_loss": 3.864013433237152,
      "tokens_seen": 565248000
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004185456369107322,
      "loss": 3.1057,
      "theoretical_loss": 3.8639661981765854,
      "tokens_seen": 565313536
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004185356068204614,
      "loss": 3.2529,
      "theoretical_loss": 3.863918970124643,
      "tokens_seen": 565379072
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004185255767301906,
      "loss": 3.1976,
      "theoretical_loss": 3.8638717490794723,
      "tokens_seen": 565444608
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041851554663991974,
      "loss": 3.1703,
      "theoretical_loss": 3.863824535039223,
      "tokens_seen": 565510144
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.000418505516549649,
      "loss": 2.9187,
      "theoretical_loss": 3.863777328002042,
      "tokens_seen": 565575680
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004184954864593781,
      "loss": 3.345,
      "theoretical_loss": 3.8637301279660825,
      "tokens_seen": 565641216
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041848545636910734,
      "loss": 3.2015,
      "theoretical_loss": 3.8636829349294928,
      "tokens_seen": 565706752
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004184754262788365,
      "loss": 3.1214,
      "theoretical_loss": 3.8636357488904247,
      "tokens_seen": 565772288
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004184653961885657,
      "loss": 2.9651,
      "theoretical_loss": 3.863588569847031,
      "tokens_seen": 565837824
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004184553660982949,
      "loss": 3.1611,
      "theoretical_loss": 3.8635413977974635,
      "tokens_seen": 565903360
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041844533600802407,
      "loss": 3.1239,
      "theoretical_loss": 3.8634942327398765,
      "tokens_seen": 565968896
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041843530591775325,
      "loss": 2.8893,
      "theoretical_loss": 3.8634470746724237,
      "tokens_seen": 566034432
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004184252758274825,
      "loss": 3.2244,
      "theoretical_loss": 3.86339992359326,
      "tokens_seen": 566099968
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004184152457372116,
      "loss": 3.1344,
      "theoretical_loss": 3.8633527795005413,
      "tokens_seen": 566165504
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041840521564694085,
      "loss": 2.7183,
      "theoretical_loss": 3.8633056423924232,
      "tokens_seen": 566231040
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041839518555666997,
      "loss": 2.8724,
      "theoretical_loss": 3.8632585122670635,
      "tokens_seen": 566296576
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004183851554663992,
      "loss": 3.5301,
      "theoretical_loss": 3.86321138912262,
      "tokens_seen": 566362112
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004183751253761284,
      "loss": 3.1334,
      "theoretical_loss": 3.86316427295725,
      "tokens_seen": 566427648
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041836509528585757,
      "loss": 3.2296,
      "theoretical_loss": 3.8631171637691137,
      "tokens_seen": 566493184
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041835506519558675,
      "loss": 3.0403,
      "theoretical_loss": 3.86307006155637,
      "tokens_seen": 566558720
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.000418345035105316,
      "loss": 2.9918,
      "theoretical_loss": 3.86302296631718,
      "tokens_seen": 566624256
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004183350050150451,
      "loss": 2.9835,
      "theoretical_loss": 3.8629758780497037,
      "tokens_seen": 566689792
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041832497492477435,
      "loss": 2.9141,
      "theoretical_loss": 3.8629287967521044,
      "tokens_seen": 566755328
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004183149448345035,
      "loss": 3.2351,
      "theoretical_loss": 3.8628817224225447,
      "tokens_seen": 566820864
    },
    {
      "epoch": 0.17,
      "objective/train/advantage_avg": 0.49422430992126465,
      "objective/train/docs_used": 331071,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.1194655895233154,
      "objective/train/original_loss": 3.1194660663604736,
      "objective/train/theoretical_loss": 3.8628346550591868,
      "objective/train/tokens_used": 587346400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24435335397720337,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506645441055298,
      "objective/train/weighted_lm_loss": 3.2774438858032227,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 3.8628346550591868,
      "tokens_seen": 566886400
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004183049147442327,
      "loss": 2.984,
      "theoretical_loss": 3.8628346550591868,
      "tokens_seen": 566886400
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004182948846539619,
      "loss": 3.1822,
      "theoretical_loss": 3.862787594660195,
      "tokens_seen": 566951936
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004182848545636911,
      "loss": 3.2541,
      "theoretical_loss": 3.862740541223734,
      "tokens_seen": 567017472
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041827482447342025,
      "loss": 3.011,
      "theoretical_loss": 3.86269349474797,
      "tokens_seen": 567083008
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041826479438314944,
      "loss": 2.9992,
      "theoretical_loss": 3.8626464552310678,
      "tokens_seen": 567148544
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004182547642928786,
      "loss": 3.0835,
      "theoretical_loss": 3.8625994226711944,
      "tokens_seen": 567214080
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041824473420260785,
      "loss": 2.9121,
      "theoretical_loss": 3.8625523970665174,
      "tokens_seen": 567279616
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.000418234704112337,
      "loss": 3.0678,
      "theoretical_loss": 3.8625053784152055,
      "tokens_seen": 567345152
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004182246740220662,
      "loss": 3.2662,
      "theoretical_loss": 3.8624583667154266,
      "tokens_seen": 567410688
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004182146439317954,
      "loss": 3.0846,
      "theoretical_loss": 3.862411361965351,
      "tokens_seen": 567476224
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004182046138415246,
      "loss": 2.9834,
      "theoretical_loss": 3.862364364163148,
      "tokens_seen": 567541760
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041819458375125376,
      "loss": 3.1952,
      "theoretical_loss": 3.8623173733069898,
      "tokens_seen": 567607296
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041818455366098294,
      "loss": 3.0283,
      "theoretical_loss": 3.8622703893950465,
      "tokens_seen": 567672832
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004181745235707121,
      "loss": 3.1276,
      "theoretical_loss": 3.8622234124254917,
      "tokens_seen": 567738368
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041816449348044136,
      "loss": 3.2169,
      "theoretical_loss": 3.862176442396498,
      "tokens_seen": 567803904
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041815446339017054,
      "loss": 2.9861,
      "theoretical_loss": 3.862129479306238,
      "tokens_seen": 567869440
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004181444332998997,
      "loss": 3.161,
      "theoretical_loss": 3.8620825231528877,
      "tokens_seen": 567934976
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004181344032096289,
      "loss": 3.1645,
      "theoretical_loss": 3.8620355739346213,
      "tokens_seen": 568000512
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004181243731193581,
      "loss": 3.1357,
      "theoretical_loss": 3.861988631649614,
      "tokens_seen": 568066048
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004181143430290873,
      "loss": 3.1816,
      "theoretical_loss": 3.861941696296044,
      "tokens_seen": 568131584
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041810431293881644,
      "loss": 3.1485,
      "theoretical_loss": 3.8618947678720863,
      "tokens_seen": 568197120
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004180942828485457,
      "loss": 2.9767,
      "theoretical_loss": 3.86184784637592,
      "tokens_seen": 568262656
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004180842527582748,
      "loss": 3.1979,
      "theoretical_loss": 3.8618009318057234,
      "tokens_seen": 568328192
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041807422266800404,
      "loss": 3.4419,
      "theoretical_loss": 3.8617540241596755,
      "tokens_seen": 568393728
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004180641925777332,
      "loss": 3.0992,
      "theoretical_loss": 3.8617071234359566,
      "tokens_seen": 568459264
    },
    {
      "epoch": 0.17,
      "objective/train/advantage_avg": 0.495907723903656,
      "objective/train/docs_used": 331416,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.995197296142578,
      "objective/train/original_loss": 2.995196580886841,
      "objective/train/theoretical_loss": 3.8616602296327467,
      "objective/train/tokens_used": 588984800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2459595948457718,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0508410930633545,
      "objective/train/weighted_lm_loss": 3.147336721420288,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0396549701690674,
      "theoretical_loss": 3.8616602296327467,
      "tokens_seen": 568524800
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004180541624874624,
      "loss": 2.7659,
      "theoretical_loss": 3.8616602296327467,
      "tokens_seen": 568524800
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004180441323971916,
      "loss": 2.9115,
      "theoretical_loss": 3.8616133427482273,
      "tokens_seen": 568590336
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004180341023069208,
      "loss": 3.2941,
      "theoretical_loss": 3.861566462780581,
      "tokens_seen": 568655872
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041802407221664994,
      "loss": 3.1894,
      "theoretical_loss": 3.8615195897279886,
      "tokens_seen": 568721408
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004180140421263792,
      "loss": 3.3317,
      "theoretical_loss": 3.8614727235886352,
      "tokens_seen": 568786944
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004180040120361083,
      "loss": 2.7879,
      "theoretical_loss": 3.8614258643607045,
      "tokens_seen": 568852480
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041799398194583754,
      "loss": 3.216,
      "theoretical_loss": 3.8613790120423808,
      "tokens_seen": 568918016
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004179839518555667,
      "loss": 3.0326,
      "theoretical_loss": 3.8613321666318496,
      "tokens_seen": 568983552
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004179739217652959,
      "loss": 3.102,
      "theoretical_loss": 3.8612853281272965,
      "tokens_seen": 569049088
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004179638916750251,
      "loss": 3.1049,
      "theoretical_loss": 3.8612384965269095,
      "tokens_seen": 569114624
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041795386158475427,
      "loss": 2.9899,
      "theoretical_loss": 3.8611916718288746,
      "tokens_seen": 569180160
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041794383149448345,
      "loss": 3.0007,
      "theoretical_loss": 3.86114485403138,
      "tokens_seen": 569245696
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004179338014042127,
      "loss": 2.9708,
      "theoretical_loss": 3.8610980431326163,
      "tokens_seen": 569311232
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004179237713139418,
      "loss": 3.1947,
      "theoretical_loss": 3.861051239130771,
      "tokens_seen": 569376768
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041791374122367105,
      "loss": 3.0033,
      "theoretical_loss": 3.8610044420240355,
      "tokens_seen": 569442304
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041790371113340017,
      "loss": 3.0346,
      "theoretical_loss": 3.8609576518105992,
      "tokens_seen": 569507840
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004178936810431294,
      "loss": 3.1519,
      "theoretical_loss": 3.8609108684886557,
      "tokens_seen": 569573376
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004178836509528586,
      "loss": 2.9017,
      "theoretical_loss": 3.8608640920563957,
      "tokens_seen": 569638912
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041787362086258777,
      "loss": 2.8402,
      "theoretical_loss": 3.8608173225120126,
      "tokens_seen": 569704448
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041786359077231695,
      "loss": 3.2802,
      "theoretical_loss": 3.8607705598536994,
      "tokens_seen": 569769984
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004178535606820462,
      "loss": 2.6938,
      "theoretical_loss": 3.860723804079651,
      "tokens_seen": 569835520
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004178435305917753,
      "loss": 2.9115,
      "theoretical_loss": 3.8606770551880625,
      "tokens_seen": 569901056
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041783350050150455,
      "loss": 3.0172,
      "theoretical_loss": 3.860630313177129,
      "tokens_seen": 569966592
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004178234704112337,
      "loss": 3.3554,
      "theoretical_loss": 3.860583578045047,
      "tokens_seen": 570032128
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004178134403209629,
      "loss": 2.9734,
      "theoretical_loss": 3.860536849790013,
      "tokens_seen": 570097664
    },
    {
      "epoch": 0.17,
      "objective/train/advantage_avg": 0.491466760635376,
      "objective/train/docs_used": 332819,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.4090259075164795,
      "objective/train/original_loss": 3.4090259075164795,
      "objective/train/theoretical_loss": 3.8604901284102264,
      "objective/train/tokens_used": 590623200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24166525900363922,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0503751039505005,
      "objective/train/weighted_lm_loss": 3.5806069374084473,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0396549701690674,
      "theoretical_loss": 3.8604901284102264,
      "tokens_seen": 570163200
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004178034102306921,
      "loss": 3.0824,
      "theoretical_loss": 3.8604901284102264,
      "tokens_seen": 570163200
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004177933801404213,
      "loss": 2.7076,
      "theoretical_loss": 3.8604434139038837,
      "tokens_seen": 570228736
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041778335005015045,
      "loss": 2.9526,
      "theoretical_loss": 3.860396706269184,
      "tokens_seen": 570294272
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041777331995987964,
      "loss": 3.0841,
      "theoretical_loss": 3.860350005504328,
      "tokens_seen": 570359808
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004177632898696088,
      "loss": 2.9026,
      "theoretical_loss": 3.860303311607516,
      "tokens_seen": 570425344
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041775325977933805,
      "loss": 3.3135,
      "theoretical_loss": 3.860256624576948,
      "tokens_seen": 570490880
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004177432296890672,
      "loss": 3.2313,
      "theoretical_loss": 3.860209944410827,
      "tokens_seen": 570556416
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004177331995987964,
      "loss": 2.908,
      "theoretical_loss": 3.8601632711073544,
      "tokens_seen": 570621952
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004177231695085256,
      "loss": 2.9727,
      "theoretical_loss": 3.860116604664734,
      "tokens_seen": 570687488
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004177131394182548,
      "loss": 3.0322,
      "theoretical_loss": 3.860069945081169,
      "tokens_seen": 570753024
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041770310932798396,
      "loss": 3.0745,
      "theoretical_loss": 3.8600232923548643,
      "tokens_seen": 570818560
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041769307923771314,
      "loss": 2.7054,
      "theoretical_loss": 3.8599766464840246,
      "tokens_seen": 570884096
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004176830491474423,
      "loss": 2.9713,
      "theoretical_loss": 3.8599300074668563,
      "tokens_seen": 570949632
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041767301905717156,
      "loss": 2.9357,
      "theoretical_loss": 3.8598833753015653,
      "tokens_seen": 571015168
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004176629889669007,
      "loss": 3.076,
      "theoretical_loss": 3.8598367499863593,
      "tokens_seen": 571080704
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004176529588766299,
      "loss": 2.9418,
      "theoretical_loss": 3.8597901315194454,
      "tokens_seen": 571146240
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041764292878635904,
      "loss": 2.7777,
      "theoretical_loss": 3.859743519899032,
      "tokens_seen": 571211776
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004176328986960883,
      "loss": 3.1197,
      "theoretical_loss": 3.8596969151233296,
      "tokens_seen": 571277312
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041762286860581746,
      "loss": 3.0084,
      "theoretical_loss": 3.859650317190547,
      "tokens_seen": 571342848
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041761283851554664,
      "loss": 3.2791,
      "theoretical_loss": 3.8596037260988947,
      "tokens_seen": 571408384
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004176028084252758,
      "loss": 2.7868,
      "theoretical_loss": 3.859557141846584,
      "tokens_seen": 571473920
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.000417592778335005,
      "loss": 2.8394,
      "theoretical_loss": 3.8595105644318273,
      "tokens_seen": 571539456
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004175827482447342,
      "loss": 3.3346,
      "theoretical_loss": 3.8594639938528363,
      "tokens_seen": 571604992
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004175727181544634,
      "loss": 3.1885,
      "theoretical_loss": 3.8594174301078246,
      "tokens_seen": 571670528
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041756268806419255,
      "loss": 3.1323,
      "theoretical_loss": 3.8593708731950063,
      "tokens_seen": 571736064
    },
    {
      "epoch": 0.17,
      "objective/train/advantage_avg": 0.4952373206615448,
      "objective/train/docs_used": 333556,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.1380772590637207,
      "objective/train/original_loss": 3.1380770206451416,
      "objective/train/theoretical_loss": 3.8593243231125958,
      "objective/train/tokens_used": 592261600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24533917009830475,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507709980010986,
      "objective/train/weighted_lm_loss": 3.2974050045013428,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0338950157165527,
      "theoretical_loss": 3.8593243231125958,
      "tokens_seen": 571801600
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004175526579739218,
      "loss": 3.0039,
      "theoretical_loss": 3.8593243231125958,
      "tokens_seen": 571801600
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041754262788365096,
      "loss": 2.9668,
      "theoretical_loss": 3.8592777798588083,
      "tokens_seen": 571867136
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041753259779338015,
      "loss": 3.052,
      "theoretical_loss": 3.8592312434318594,
      "tokens_seen": 571932672
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004175225677031093,
      "loss": 2.9521,
      "theoretical_loss": 3.8591847138299658,
      "tokens_seen": 571998208
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004175125376128385,
      "loss": 3.0093,
      "theoretical_loss": 3.859138191051345,
      "tokens_seen": 572063744
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004175025075225677,
      "loss": 3.152,
      "theoretical_loss": 3.8590916750942146,
      "tokens_seen": 572129280
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004174924774322969,
      "loss": 3.1157,
      "theoretical_loss": 3.8590451659567937,
      "tokens_seen": 572194816
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041748244734202605,
      "loss": 3.0825,
      "theoretical_loss": 3.8589986636373013,
      "tokens_seen": 572260352
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004174724172517553,
      "loss": 3.0412,
      "theoretical_loss": 3.8589521681339565,
      "tokens_seen": 572325888
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004174623871614844,
      "loss": 2.9202,
      "theoretical_loss": 3.8589056794449808,
      "tokens_seen": 572391424
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041745235707121365,
      "loss": 3.0002,
      "theoretical_loss": 3.858859197568595,
      "tokens_seen": 572456960
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041744232698094283,
      "loss": 3.132,
      "theoretical_loss": 3.858812722503022,
      "tokens_seen": 572522496
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.000417432296890672,
      "loss": 2.8161,
      "theoretical_loss": 3.8587662542464827,
      "tokens_seen": 572588032
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004174222668004012,
      "loss": 2.8677,
      "theoretical_loss": 3.858719792797202,
      "tokens_seen": 572653568
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041741223671013037,
      "loss": 2.9733,
      "theoretical_loss": 3.8586733381534026,
      "tokens_seen": 572719104
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004174022066198596,
      "loss": 2.9585,
      "theoretical_loss": 3.8586268903133094,
      "tokens_seen": 572784640
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004173921765295888,
      "loss": 2.7914,
      "theoretical_loss": 3.8585804492751485,
      "tokens_seen": 572850176
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041738214643931797,
      "loss": 3.1164,
      "theoretical_loss": 3.8585340150371445,
      "tokens_seen": 572915712
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041737211634904715,
      "loss": 3.1585,
      "theoretical_loss": 3.8584875875975246,
      "tokens_seen": 572981248
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004173620862587764,
      "loss": 3.0437,
      "theoretical_loss": 3.858441166954516,
      "tokens_seen": 573046784
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004173520561685055,
      "loss": 3.1131,
      "theoretical_loss": 3.858394753106347,
      "tokens_seen": 573112320
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041734202607823475,
      "loss": 2.974,
      "theoretical_loss": 3.8583483460512458,
      "tokens_seen": 573177856
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004173319959879639,
      "loss": 3.2608,
      "theoretical_loss": 3.8583019457874412,
      "tokens_seen": 573243392
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004173219658976931,
      "loss": 3.0054,
      "theoretical_loss": 3.8582555523131643,
      "tokens_seen": 573308928
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004173119358074223,
      "loss": 3.3472,
      "theoretical_loss": 3.8582091656266444,
      "tokens_seen": 573374464
    },
    {
      "epoch": 0.17,
      "objective/train/advantage_avg": 0.49577268958091736,
      "objective/train/docs_used": 334721,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.9797611236572266,
      "objective/train/original_loss": 2.9797613620758057,
      "objective/train/theoretical_loss": 3.8581627857261136,
      "objective/train/tokens_used": 593900000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24586111307144165,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0508270263671875,
      "objective/train/weighted_lm_loss": 3.13105845451355,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0363596677780151,
      "theoretical_loss": 3.8581627857261136,
      "tokens_seen": 573440000
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004173019057171515,
      "loss": 3.1366,
      "theoretical_loss": 3.8581627857261136,
      "tokens_seen": 573440000
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041729187562688065,
      "loss": 2.8157,
      "theoretical_loss": 3.858116412609803,
      "tokens_seen": 573505536
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041728184553660984,
      "loss": 3.045,
      "theoretical_loss": 3.8580700462759463,
      "tokens_seen": 573571072
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.000417271815446339,
      "loss": 3.0327,
      "theoretical_loss": 3.8580236867227757,
      "tokens_seen": 573636608
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041726178535606825,
      "loss": 3.0981,
      "theoretical_loss": 3.8579773339485257,
      "tokens_seen": 573702144
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004172517552657974,
      "loss": 3.1352,
      "theoretical_loss": 3.85793098795143,
      "tokens_seen": 573767680
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004172417251755266,
      "loss": 3.1281,
      "theoretical_loss": 3.8578846487297245,
      "tokens_seen": 573833216
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004172316950852558,
      "loss": 2.8573,
      "theoretical_loss": 3.8578383162816454,
      "tokens_seen": 573898752
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.000417221664994985,
      "loss": 2.9716,
      "theoretical_loss": 3.8577919906054285,
      "tokens_seen": 573964288
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041721163490471416,
      "loss": 3.1505,
      "theoretical_loss": 3.8577456716993117,
      "tokens_seen": 574029824
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041720160481444334,
      "loss": 2.7611,
      "theoretical_loss": 3.857699359561532,
      "tokens_seen": 574095360
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004171915747241725,
      "loss": 2.8221,
      "theoretical_loss": 3.857653054190328,
      "tokens_seen": 574160896
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041718154463390176,
      "loss": 3.2015,
      "theoretical_loss": 3.8576067555839395,
      "tokens_seen": 574226432
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004171715145436309,
      "loss": 2.7289,
      "theoretical_loss": 3.8575604637406062,
      "tokens_seen": 574291968
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004171614844533601,
      "loss": 2.7514,
      "theoretical_loss": 3.857514178658568,
      "tokens_seen": 574357504
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041715145436308924,
      "loss": 3.0737,
      "theoretical_loss": 3.857467900336067,
      "tokens_seen": 574423040
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004171414242728185,
      "loss": 2.9461,
      "theoretical_loss": 3.8574216287713434,
      "tokens_seen": 574488576
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041713139418254766,
      "loss": 2.971,
      "theoretical_loss": 3.8573753639626416,
      "tokens_seen": 574554112
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041712136409227684,
      "loss": 2.9397,
      "theoretical_loss": 3.857329105908203,
      "tokens_seen": 574619648
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.000417111334002006,
      "loss": 2.7754,
      "theoretical_loss": 3.857282854606273,
      "tokens_seen": 574685184
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004171013039117352,
      "loss": 3.221,
      "theoretical_loss": 3.8572366100550948,
      "tokens_seen": 574750720
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004170912738214644,
      "loss": 3.2962,
      "theoretical_loss": 3.857190372252914,
      "tokens_seen": 574816256
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004170812437311936,
      "loss": 3.1426,
      "theoretical_loss": 3.8571441411979763,
      "tokens_seen": 574881792
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041707121364092275,
      "loss": 2.9933,
      "theoretical_loss": 3.857097916888528,
      "tokens_seen": 574947328
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.000417061183550652,
      "loss": 3.0399,
      "theoretical_loss": 3.857051699322816,
      "tokens_seen": 575012864
    },
    {
      "epoch": 0.17,
      "objective/train/advantage_avg": 0.4925279915332794,
      "objective/train/docs_used": 335233,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.93564510345459,
      "objective/train/original_loss": 2.93564510345459,
      "objective/train/theoretical_loss": 3.8570054884990883,
      "objective/train/tokens_used": 595538400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2426604926586151,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0504863262176514,
      "objective/train/weighted_lm_loss": 3.0837242603302,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0425469875335693,
      "theoretical_loss": 3.8570054884990883,
      "tokens_seen": 575078400
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041705115346038116,
      "loss": 3.1551,
      "theoretical_loss": 3.8570054884990883,
      "tokens_seen": 575078400
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041704112337011035,
      "loss": 2.9114,
      "theoretical_loss": 3.856959284415593,
      "tokens_seen": 575143936
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004170310932798395,
      "loss": 2.982,
      "theoretical_loss": 3.8569130870705797,
      "tokens_seen": 575209472
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004170210631895687,
      "loss": 3.2694,
      "theoretical_loss": 3.8568668964622974,
      "tokens_seen": 575275008
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004170110330992979,
      "loss": 2.9642,
      "theoretical_loss": 3.8568207125889966,
      "tokens_seen": 575340544
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004170010030090271,
      "loss": 3.2649,
      "theoretical_loss": 3.8567745354489285,
      "tokens_seen": 575406080
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041699097291875625,
      "loss": 3.1896,
      "theoretical_loss": 3.856728365040345,
      "tokens_seen": 575471616
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004169809428284855,
      "loss": 3.1624,
      "theoretical_loss": 3.8566822013614974,
      "tokens_seen": 575537152
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004169709127382146,
      "loss": 2.9984,
      "theoretical_loss": 3.8566360444106396,
      "tokens_seen": 575602688
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041696088264794385,
      "loss": 2.86,
      "theoretical_loss": 3.8565898941860244,
      "tokens_seen": 575668224
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041695085255767303,
      "loss": 3.1221,
      "theoretical_loss": 3.8565437506859066,
      "tokens_seen": 575733760
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004169408224674022,
      "loss": 3.2794,
      "theoretical_loss": 3.856497613908541,
      "tokens_seen": 575799296
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004169307923771314,
      "loss": 3.1683,
      "theoretical_loss": 3.856451483852183,
      "tokens_seen": 575864832
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004169207622868606,
      "loss": 2.9151,
      "theoretical_loss": 3.8564053605150894,
      "tokens_seen": 575930368
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041691073219658975,
      "loss": 2.9781,
      "theoretical_loss": 3.8563592438955157,
      "tokens_seen": 575995904
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.000416900702106319,
      "loss": 2.9236,
      "theoretical_loss": 3.856313133991721,
      "tokens_seen": 576061440
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004168906720160481,
      "loss": 3.0674,
      "theoretical_loss": 3.856267030801962,
      "tokens_seen": 576126976
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041688064192577735,
      "loss": 3.1561,
      "theoretical_loss": 3.856220934324499,
      "tokens_seen": 576192512
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041687061183550653,
      "loss": 2.9357,
      "theoretical_loss": 3.8561748445575903,
      "tokens_seen": 576258048
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004168605817452357,
      "loss": 2.7893,
      "theoretical_loss": 3.8561287614994963,
      "tokens_seen": 576323584
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004168505516549649,
      "loss": 2.8486,
      "theoretical_loss": 3.8560826851484777,
      "tokens_seen": 576389120
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004168405215646941,
      "loss": 2.8467,
      "theoretical_loss": 3.8560366155027963,
      "tokens_seen": 576454656
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041683049147442326,
      "loss": 3.0788,
      "theoretical_loss": 3.855990552560714,
      "tokens_seen": 576520192
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004168204613841525,
      "loss": 2.8487,
      "theoretical_loss": 3.8559444963204923,
      "tokens_seen": 576585728
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004168104312938816,
      "loss": 2.7794,
      "theoretical_loss": 3.8558984467803965,
      "tokens_seen": 576651264
    },
    {
      "epoch": 0.17,
      "objective/train/advantage_avg": 0.49501481652259827,
      "objective/train/docs_used": 335838,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.5680785179138184,
      "objective/train/original_loss": 2.5680782794952393,
      "objective/train/theoretical_loss": 3.855852403938689,
      "objective/train/tokens_used": 597176800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24509496986865997,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507473945617676,
      "objective/train/weighted_lm_loss": 2.698227643966675,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0437887907028198,
      "theoretical_loss": 3.855852403938689,
      "tokens_seen": 576716800
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041680040120361085,
      "loss": 3.047,
      "theoretical_loss": 3.855852403938689,
      "tokens_seen": 576716800
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041679037111334,
      "loss": 2.8439,
      "theoretical_loss": 3.855806367793636,
      "tokens_seen": 576782336
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004167803410230692,
      "loss": 2.9016,
      "theoretical_loss": 3.8557603383435013,
      "tokens_seen": 576847872
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004167703109327984,
      "loss": 3.0558,
      "theoretical_loss": 3.8557143155865514,
      "tokens_seen": 576913408
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004167602808425276,
      "loss": 2.8847,
      "theoretical_loss": 3.855668299521053,
      "tokens_seen": 576978944
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041675025075225676,
      "loss": 3.3937,
      "theoretical_loss": 3.8556222901452735,
      "tokens_seen": 577044480
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.000416740220661986,
      "loss": 3.2152,
      "theoretical_loss": 3.85557628745748,
      "tokens_seen": 577110016
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004167301905717151,
      "loss": 3.2116,
      "theoretical_loss": 3.8555302914559415,
      "tokens_seen": 577175552
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.00041672016048144436,
      "loss": 2.8698,
      "theoretical_loss": 3.855484302138927,
      "tokens_seen": 577241088
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004167101303911735,
      "loss": 3.0695,
      "theoretical_loss": 3.8554383195047066,
      "tokens_seen": 577306624
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004167001003009027,
      "loss": 3.3049,
      "theoretical_loss": 3.855392343551551,
      "tokens_seen": 577372160
    },
    {
      "epoch": 0.17,
      "learning_rate": 0.0004166900702106319,
      "loss": 3.0505,
      "theoretical_loss": 3.8553463742777305,
      "tokens_seen": 577437696
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004166800401203611,
      "loss": 2.7861,
      "theoretical_loss": 3.855300411681517,
      "tokens_seen": 577503232
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041667001003009026,
      "loss": 3.0091,
      "theoretical_loss": 3.8552544557611834,
      "tokens_seen": 577568768
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041665997993981944,
      "loss": 3.1295,
      "theoretical_loss": 3.855208506515002,
      "tokens_seen": 577634304
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004166499498495487,
      "loss": 2.8988,
      "theoretical_loss": 3.8551625639412475,
      "tokens_seen": 577699840
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041663991975927786,
      "loss": 3.2009,
      "theoretical_loss": 3.8551166280381928,
      "tokens_seen": 577765376
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041662988966900704,
      "loss": 2.7845,
      "theoretical_loss": 3.8550706988041137,
      "tokens_seen": 577830912
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004166198595787362,
      "loss": 2.9714,
      "theoretical_loss": 3.8550247762372862,
      "tokens_seen": 577896448
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004166098294884654,
      "loss": 3.1317,
      "theoretical_loss": 3.8549788603359856,
      "tokens_seen": 577961984
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004165997993981946,
      "loss": 2.912,
      "theoretical_loss": 3.8549329510984887,
      "tokens_seen": 578027520
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004165897693079238,
      "loss": 2.7629,
      "theoretical_loss": 3.854887048523074,
      "tokens_seen": 578093056
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041657973921765295,
      "loss": 3.226,
      "theoretical_loss": 3.8548411526080195,
      "tokens_seen": 578158592
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004165697091273822,
      "loss": 3.0354,
      "theoretical_loss": 3.854795263351603,
      "tokens_seen": 578224128
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041655967903711136,
      "loss": 3.3154,
      "theoretical_loss": 3.854749380752105,
      "tokens_seen": 578289664
    },
    {
      "epoch": 0.18,
      "objective/train/advantage_avg": 0.49487727880477905,
      "objective/train/docs_used": 337192,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.1660261154174805,
      "objective/train/original_loss": 3.1660265922546387,
      "objective/train/theoretical_loss": 3.854703504807805,
      "objective/train/tokens_used": 598815200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24495400488376617,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507328510284424,
      "objective/train/weighted_lm_loss": 3.3266191482543945,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.044721245765686,
      "theoretical_loss": 3.854703504807805,
      "tokens_seen": 578355200
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041654964894684055,
      "loss": 3.0227,
      "theoretical_loss": 3.854703504807805,
      "tokens_seen": 578355200
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004165396188565697,
      "loss": 2.9139,
      "theoretical_loss": 3.8546576355169835,
      "tokens_seen": 578420736
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004165295887662989,
      "loss": 3.2201,
      "theoretical_loss": 3.8546117728779223,
      "tokens_seen": 578486272
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004165195586760281,
      "loss": 2.9697,
      "theoretical_loss": 3.8545659168889035,
      "tokens_seen": 578551808
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004165095285857573,
      "loss": 3.1397,
      "theoretical_loss": 3.854520067548209,
      "tokens_seen": 578617344
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041649949849548645,
      "loss": 3.3129,
      "theoretical_loss": 3.854474224854123,
      "tokens_seen": 578682880
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004164894684052157,
      "loss": 2.9798,
      "theoretical_loss": 3.8544283888049287,
      "tokens_seen": 578748416
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004164794383149448,
      "loss": 3.0368,
      "theoretical_loss": 3.854382559398911,
      "tokens_seen": 578813952
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041646940822467405,
      "loss": 2.7858,
      "theoretical_loss": 3.854336736634355,
      "tokens_seen": 578879488
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041645937813440323,
      "loss": 3.0964,
      "theoretical_loss": 3.854290920509546,
      "tokens_seen": 578945024
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004164493480441324,
      "loss": 3.2531,
      "theoretical_loss": 3.854245111022771,
      "tokens_seen": 579010560
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004164393179538616,
      "loss": 3.0242,
      "theoretical_loss": 3.8541993081723174,
      "tokens_seen": 579076096
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004164292878635908,
      "loss": 3.0505,
      "theoretical_loss": 3.8541535119564725,
      "tokens_seen": 579141632
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041641925777331995,
      "loss": 3.0947,
      "theoretical_loss": 3.8541077223735245,
      "tokens_seen": 579207168
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004164092276830492,
      "loss": 2.8429,
      "theoretical_loss": 3.8540619394217623,
      "tokens_seen": 579272704
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004163991975927783,
      "loss": 2.9818,
      "theoretical_loss": 3.8540161630994763,
      "tokens_seen": 579338240
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041638916750250755,
      "loss": 3.0826,
      "theoretical_loss": 3.853970393404956,
      "tokens_seen": 579403776
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041637913741223673,
      "loss": 3.254,
      "theoretical_loss": 3.8539246303364925,
      "tokens_seen": 579469312
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004163691073219659,
      "loss": 3.152,
      "theoretical_loss": 3.853878873892377,
      "tokens_seen": 579534848
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004163590772316951,
      "loss": 3.1323,
      "theoretical_loss": 3.8538331240709027,
      "tokens_seen": 579600384
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004163490471414243,
      "loss": 3.0184,
      "theoretical_loss": 3.8537873808703615,
      "tokens_seen": 579665920
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041633901705115346,
      "loss": 2.6076,
      "theoretical_loss": 3.8537416442890464,
      "tokens_seen": 579731456
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004163289869608827,
      "loss": 3.2312,
      "theoretical_loss": 3.853695914325253,
      "tokens_seen": 579796992
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004163189568706118,
      "loss": 2.8662,
      "theoretical_loss": 3.8536501909772745,
      "tokens_seen": 579862528
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041630892678034105,
      "loss": 2.8232,
      "theoretical_loss": 3.853604474243407,
      "tokens_seen": 579928064
    },
    {
      "epoch": 0.18,
      "objective/train/advantage_avg": 0.4956149458885193,
      "objective/train/docs_used": 337878,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.1683638095855713,
      "objective/train/original_loss": 2.168363571166992,
      "objective/train/theoretical_loss": 3.8535587641219466,
      "objective/train/tokens_used": 600453600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2456687092781067,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0508103370666504,
      "objective/train/weighted_lm_loss": 2.2783756256103516,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.047652244567871,
      "theoretical_loss": 3.8535587641219466,
      "tokens_seen": 579993600
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004162988966900702,
      "loss": 2.6013,
      "theoretical_loss": 3.8535587641219466,
      "tokens_seen": 579993600
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004162888665997994,
      "loss": 2.9385,
      "theoretical_loss": 3.853513060611189,
      "tokens_seen": 580059136
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004162788365095286,
      "loss": 3.0729,
      "theoretical_loss": 3.853467363709432,
      "tokens_seen": 580124672
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004162688064192578,
      "loss": 3.0439,
      "theoretical_loss": 3.8534216734149735,
      "tokens_seen": 580190208
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041625877632898696,
      "loss": 2.6294,
      "theoretical_loss": 3.8533759897261124,
      "tokens_seen": 580255744
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004162487462387162,
      "loss": 3.0134,
      "theoretical_loss": 3.853330312641147,
      "tokens_seen": 580321280
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004162387161484453,
      "loss": 3.0172,
      "theoretical_loss": 3.8532846421583775,
      "tokens_seen": 580386816
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041622868605817456,
      "loss": 3.0678,
      "theoretical_loss": 3.853238978276104,
      "tokens_seen": 580452352
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004162186559679037,
      "loss": 3.2313,
      "theoretical_loss": 3.8531933209926272,
      "tokens_seen": 580517888
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004162086258776329,
      "loss": 2.9014,
      "theoretical_loss": 3.8531476703062495,
      "tokens_seen": 580583424
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004161985957873621,
      "loss": 2.8994,
      "theoretical_loss": 3.8531020262152733,
      "tokens_seen": 580648960
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004161885656970913,
      "loss": 3.2697,
      "theoretical_loss": 3.8530563887180005,
      "tokens_seen": 580714496
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041617853560682046,
      "loss": 3.0992,
      "theoretical_loss": 3.853010757812735,
      "tokens_seen": 580780032
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041616850551654964,
      "loss": 3.1091,
      "theoretical_loss": 3.852965133497781,
      "tokens_seen": 580845568
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004161584754262788,
      "loss": 2.9985,
      "theoretical_loss": 3.852919515771444,
      "tokens_seen": 580911104
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041614844533600806,
      "loss": 3.1542,
      "theoretical_loss": 3.852873904632028,
      "tokens_seen": 580976640
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004161384152457372,
      "loss": 2.9035,
      "theoretical_loss": 3.8528283000778405,
      "tokens_seen": 581042176
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004161283851554664,
      "loss": 2.8908,
      "theoretical_loss": 3.852782702107186,
      "tokens_seen": 581107712
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041611835506519555,
      "loss": 2.7984,
      "theoretical_loss": 3.8527371107183748,
      "tokens_seen": 581173248
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004161083249749248,
      "loss": 3.1275,
      "theoretical_loss": 3.8526915259097123,
      "tokens_seen": 581238784
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041609829488465397,
      "loss": 2.9764,
      "theoretical_loss": 3.852645947679508,
      "tokens_seen": 581304320
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041608826479438315,
      "loss": 2.816,
      "theoretical_loss": 3.8526003760260714,
      "tokens_seen": 581369856
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041607823470411233,
      "loss": 3.0337,
      "theoretical_loss": 3.8525548109477112,
      "tokens_seen": 581435392
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041606820461384156,
      "loss": 3.086,
      "theoretical_loss": 3.852509252442739,
      "tokens_seen": 581500928
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004160581745235707,
      "loss": 3.0329,
      "theoretical_loss": 3.852463700509465,
      "tokens_seen": 581566464
    },
    {
      "epoch": 0.18,
      "objective/train/advantage_avg": 0.495735764503479,
      "objective/train/docs_used": 339259,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.5387542247772217,
      "objective/train/original_loss": 2.5387542247772217,
      "objective/train/theoretical_loss": 3.8524181551462013,
      "objective/train/tokens_used": 602092000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24579422175884247,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0508230924606323,
      "objective/train/weighted_lm_loss": 2.6677615642547607,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0467102527618408,
      "theoretical_loss": 3.8524181551462013,
      "tokens_seen": 581632000
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004160481444332999,
      "loss": 2.6219,
      "theoretical_loss": 3.8524181551462013,
      "tokens_seen": 581632000
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041603811434302905,
      "loss": 2.8471,
      "theoretical_loss": 3.8523726163512597,
      "tokens_seen": 581697536
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004160280842527583,
      "loss": 3.0132,
      "theoretical_loss": 3.8523270841229538,
      "tokens_seen": 581763072
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041601805416248747,
      "loss": 3.1341,
      "theoretical_loss": 3.8522815584595964,
      "tokens_seen": 581828608
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041600802407221665,
      "loss": 2.998,
      "theoretical_loss": 3.852236039359502,
      "tokens_seen": 581894144
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041599799398194583,
      "loss": 3.0888,
      "theoretical_loss": 3.8521905268209857,
      "tokens_seen": 581959680
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.000415987963891675,
      "loss": 3.109,
      "theoretical_loss": 3.8521450208423618,
      "tokens_seen": 582025216
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004159779338014042,
      "loss": 2.8926,
      "theoretical_loss": 3.852099521421948,
      "tokens_seen": 582090752
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041596790371113343,
      "loss": 3.0305,
      "theoretical_loss": 3.8520540285580593,
      "tokens_seen": 582156288
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041595787362086256,
      "loss": 3.0623,
      "theoretical_loss": 3.8520085422490142,
      "tokens_seen": 582221824
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004159478435305918,
      "loss": 3.0999,
      "theoretical_loss": 3.8519630624931294,
      "tokens_seen": 582287360
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004159378134403209,
      "loss": 3.0824,
      "theoretical_loss": 3.8519175892887243,
      "tokens_seen": 582352896
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041592778335005015,
      "loss": 2.6836,
      "theoretical_loss": 3.851872122634118,
      "tokens_seen": 582418432
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041591775325977934,
      "loss": 2.7825,
      "theoretical_loss": 3.851826662527629,
      "tokens_seen": 582483968
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004159077231695085,
      "loss": 2.8237,
      "theoretical_loss": 3.85178120896758,
      "tokens_seen": 582549504
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041589769307923775,
      "loss": 2.9955,
      "theoretical_loss": 3.85173576195229,
      "tokens_seen": 582615040
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041588766298896693,
      "loss": 2.9963,
      "theoretical_loss": 3.8516903214800813,
      "tokens_seen": 582680576
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004158776328986961,
      "loss": 2.7534,
      "theoretical_loss": 3.851644887549276,
      "tokens_seen": 582746112
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004158676028084253,
      "loss": 2.9855,
      "theoretical_loss": 3.851599460158197,
      "tokens_seen": 582811648
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004158575727181545,
      "loss": 2.8409,
      "theoretical_loss": 3.8515540393051673,
      "tokens_seen": 582877184
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041584754262788366,
      "loss": 3.0117,
      "theoretical_loss": 3.851508624988512,
      "tokens_seen": 582942720
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004158375125376129,
      "loss": 3.0006,
      "theoretical_loss": 3.851463217206555,
      "tokens_seen": 583008256
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.000415827482447342,
      "loss": 2.883,
      "theoretical_loss": 3.851417815957622,
      "tokens_seen": 583073792
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041581745235707126,
      "loss": 3.107,
      "theoretical_loss": 3.8513724212400384,
      "tokens_seen": 583139328
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004158074222668004,
      "loss": 2.9199,
      "theoretical_loss": 3.851327033052131,
      "tokens_seen": 583204864
    },
    {
      "epoch": 0.18,
      "objective/train/advantage_avg": 0.4949757754802704,
      "objective/train/docs_used": 340083,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.161888599395752,
      "objective/train/original_loss": 3.161888837814331,
      "objective/train/theoretical_loss": 3.8512816513922274,
      "objective/train/tokens_used": 603730400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24506136775016785,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507433414459229,
      "objective/train/weighted_lm_loss": 3.322366952896118,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0417574644088745,
      "theoretical_loss": 3.8512816513922274,
      "tokens_seen": 583270400
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004157973921765296,
      "loss": 3.1798,
      "theoretical_loss": 3.8512816513922274,
      "tokens_seen": 583270400
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004157873620862588,
      "loss": 3.1216,
      "theoretical_loss": 3.851236276258655,
      "tokens_seen": 583335936
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.000415777331995988,
      "loss": 3.1552,
      "theoretical_loss": 3.8511909076497424,
      "tokens_seen": 583401472
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041576730190571716,
      "loss": 2.8931,
      "theoretical_loss": 3.851145545563818,
      "tokens_seen": 583467008
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004157572718154464,
      "loss": 3.0498,
      "theoretical_loss": 3.8511001899992126,
      "tokens_seen": 583532544
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004157472417251755,
      "loss": 2.9013,
      "theoretical_loss": 3.851054840954255,
      "tokens_seen": 583598080
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041573721163490476,
      "loss": 3.0078,
      "theoretical_loss": 3.8510094984272767,
      "tokens_seen": 583663616
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004157271815446339,
      "loss": 2.9759,
      "theoretical_loss": 3.8509641624166098,
      "tokens_seen": 583729152
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004157171514543631,
      "loss": 2.7248,
      "theoretical_loss": 3.850918832920585,
      "tokens_seen": 583794688
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004157071213640923,
      "loss": 3.1564,
      "theoretical_loss": 3.8508735099375366,
      "tokens_seen": 583860224
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004156970912738215,
      "loss": 2.8679,
      "theoretical_loss": 3.850828193465797,
      "tokens_seen": 583925760
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041568706118355066,
      "loss": 3.1978,
      "theoretical_loss": 3.8507828835036997,
      "tokens_seen": 583991296
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041567703109327984,
      "loss": 2.9871,
      "theoretical_loss": 3.85073758004958,
      "tokens_seen": 584056832
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.000415667001003009,
      "loss": 3.2178,
      "theoretical_loss": 3.850692283101773,
      "tokens_seen": 584122368
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041565697091273826,
      "loss": 3.0052,
      "theoretical_loss": 3.850646992658614,
      "tokens_seen": 584187904
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004156469408224674,
      "loss": 2.8235,
      "theoretical_loss": 3.8506017087184405,
      "tokens_seen": 584253440
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004156369107321966,
      "loss": 2.9924,
      "theoretical_loss": 3.850556431279588,
      "tokens_seen": 584318976
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041562688064192575,
      "loss": 2.9046,
      "theoretical_loss": 3.8505111603403948,
      "tokens_seen": 584384512
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.000415616850551655,
      "loss": 3.1815,
      "theoretical_loss": 3.850465895899199,
      "tokens_seen": 584450048
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041560682046138417,
      "loss": 2.9512,
      "theoretical_loss": 3.8504206379543398,
      "tokens_seen": 584515584
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041559679037111335,
      "loss": 3.1071,
      "theoretical_loss": 3.8503753865041563,
      "tokens_seen": 584581120
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041558676028084253,
      "loss": 2.8655,
      "theoretical_loss": 3.8503301415469884,
      "tokens_seen": 584646656
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041557673019057176,
      "loss": 2.8889,
      "theoretical_loss": 3.8502849030811768,
      "tokens_seen": 584712192
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004155667001003009,
      "loss": 3.1053,
      "theoretical_loss": 3.8502396711050633,
      "tokens_seen": 584777728
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004155566700100301,
      "loss": 2.7721,
      "theoretical_loss": 3.850194445616989,
      "tokens_seen": 584843264
    },
    {
      "epoch": 0.18,
      "objective/train/advantage_avg": 0.4950791597366333,
      "objective/train/docs_used": 341398,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.9058704376220703,
      "objective/train/original_loss": 2.905870199203491,
      "objective/train/theoretical_loss": 3.8501492266152972,
      "objective/train/tokens_used": 605368800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24514710903167725,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050754189491272,
      "objective/train/weighted_lm_loss": 3.0532217025756836,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.044721245765686,
      "theoretical_loss": 3.8501492266152972,
      "tokens_seen": 584908800
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041554663991975925,
      "loss": 3.141,
      "theoretical_loss": 3.8501492266152972,
      "tokens_seen": 584908800
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004155366098294885,
      "loss": 2.7508,
      "theoretical_loss": 3.8501040140983305,
      "tokens_seen": 584974336
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041552657973921767,
      "loss": 2.9184,
      "theoretical_loss": 3.850058808064432,
      "tokens_seen": 585039872
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041551654964894685,
      "loss": 3.0906,
      "theoretical_loss": 3.850013608511947,
      "tokens_seen": 585105408
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041550651955867603,
      "loss": 2.9287,
      "theoretical_loss": 3.8499684154392204,
      "tokens_seen": 585170944
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004154964894684052,
      "loss": 2.9994,
      "theoretical_loss": 3.849923228844598,
      "tokens_seen": 585236480
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004154864593781344,
      "loss": 2.9544,
      "theoretical_loss": 3.849878048726424,
      "tokens_seen": 585302016
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041547642928786363,
      "loss": 2.8002,
      "theoretical_loss": 3.849832875083047,
      "tokens_seen": 585367552
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041546639919759276,
      "loss": 3.0438,
      "theoretical_loss": 3.849787707912814,
      "tokens_seen": 585433088
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.000415456369107322,
      "loss": 2.8273,
      "theoretical_loss": 3.849742547214073,
      "tokens_seen": 585498624
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004154463390170511,
      "loss": 2.9441,
      "theoretical_loss": 3.8496973929851723,
      "tokens_seen": 585564160
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041543630892678035,
      "loss": 2.8543,
      "theoretical_loss": 3.8496522452244606,
      "tokens_seen": 585629696
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041542627883650954,
      "loss": 2.7974,
      "theoretical_loss": 3.8496071039302886,
      "tokens_seen": 585695232
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004154162487462387,
      "loss": 2.9801,
      "theoretical_loss": 3.849561969101006,
      "tokens_seen": 585760768
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004154062186559679,
      "loss": 2.9632,
      "theoretical_loss": 3.8495168407349647,
      "tokens_seen": 585826304
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041539618856569713,
      "loss": 2.9879,
      "theoretical_loss": 3.8494717188305154,
      "tokens_seen": 585891840
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041538615847542626,
      "loss": 3.1676,
      "theoretical_loss": 3.8494266033860103,
      "tokens_seen": 585957376
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004153761283851555,
      "loss": 3.0879,
      "theoretical_loss": 3.849381494399802,
      "tokens_seen": 586022912
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004153660982948846,
      "loss": 2.9697,
      "theoretical_loss": 3.8493363918702457,
      "tokens_seen": 586088448
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041535606820461386,
      "loss": 3.1372,
      "theoretical_loss": 3.8492912957956933,
      "tokens_seen": 586153984
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041534603811434304,
      "loss": 3.0034,
      "theoretical_loss": 3.8492462061745,
      "tokens_seen": 586219520
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004153360080240722,
      "loss": 2.8604,
      "theoretical_loss": 3.849201123005021,
      "tokens_seen": 586285056
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004153259779338014,
      "loss": 3.0208,
      "theoretical_loss": 3.849156046285613,
      "tokens_seen": 586350592
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004153159478435306,
      "loss": 3.1087,
      "theoretical_loss": 3.8491109760146314,
      "tokens_seen": 586416128
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041530591775325976,
      "loss": 2.8284,
      "theoretical_loss": 3.849065912190434,
      "tokens_seen": 586481664
    },
    {
      "epoch": 0.18,
      "objective/train/advantage_avg": 0.49408459663391113,
      "objective/train/docs_used": 342071,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.8466315269470215,
      "objective/train/original_loss": 2.8466310501098633,
      "objective/train/theoretical_loss": 3.849020854811377,
      "objective/train/tokens_used": 607007200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2441895306110382,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506497621536255,
      "objective/train/weighted_lm_loss": 2.9906482696533203,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0304546356201172,
      "theoretical_loss": 3.849020854811377,
      "tokens_seen": 586547200
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.000415295887662989,
      "loss": 3.0672,
      "theoretical_loss": 3.849020854811377,
      "tokens_seen": 586547200
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004152858575727181,
      "loss": 3.0409,
      "theoretical_loss": 3.8489758038758204,
      "tokens_seen": 586612736
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041527582748244736,
      "loss": 3.0651,
      "theoretical_loss": 3.848930759382122,
      "tokens_seen": 586678272
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004152657973921765,
      "loss": 3.1294,
      "theoretical_loss": 3.8488857213286414,
      "tokens_seen": 586743808
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004152557673019057,
      "loss": 2.8821,
      "theoretical_loss": 3.8488406897137386,
      "tokens_seen": 586809344
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004152457372116349,
      "loss": 3.0053,
      "theoretical_loss": 3.8487956645357744,
      "tokens_seen": 586874880
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004152357071213641,
      "loss": 2.9481,
      "theoretical_loss": 3.84875064579311,
      "tokens_seen": 586940416
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041522567703109327,
      "loss": 3.0916,
      "theoretical_loss": 3.848705633484107,
      "tokens_seen": 587005952
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004152156469408225,
      "loss": 3.1136,
      "theoretical_loss": 3.8486606276071287,
      "tokens_seen": 587071488
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041520561685055163,
      "loss": 3.0275,
      "theoretical_loss": 3.8486156281605366,
      "tokens_seen": 587137024
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041519558676028086,
      "loss": 3.2369,
      "theoretical_loss": 3.848570635142696,
      "tokens_seen": 587202560
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041518555667001,
      "loss": 2.7982,
      "theoretical_loss": 3.84852564855197,
      "tokens_seen": 587268096
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004151755265797392,
      "loss": 3.0937,
      "theoretical_loss": 3.848480668386723,
      "tokens_seen": 587333632
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004151654964894684,
      "loss": 2.8783,
      "theoretical_loss": 3.848435694645322,
      "tokens_seen": 587399168
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004151554663991976,
      "loss": 2.972,
      "theoretical_loss": 3.848390727326132,
      "tokens_seen": 587464704
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004151454363089268,
      "loss": 2.9038,
      "theoretical_loss": 3.84834576642752,
      "tokens_seen": 587530240
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041513540621865595,
      "loss": 3.0849,
      "theoretical_loss": 3.848300811947852,
      "tokens_seen": 587595776
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004151253761283852,
      "loss": 2.9321,
      "theoretical_loss": 3.8482558638854982,
      "tokens_seen": 587661312
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041511534603811437,
      "loss": 3.1394,
      "theoretical_loss": 3.8482109222388248,
      "tokens_seen": 587726848
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041510531594784355,
      "loss": 3.0545,
      "theoretical_loss": 3.848165987006202,
      "tokens_seen": 587792384
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041509528585757273,
      "loss": 3.1756,
      "theoretical_loss": 3.848121058185999,
      "tokens_seen": 587857920
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041508525576730196,
      "loss": 2.767,
      "theoretical_loss": 3.8480761357765862,
      "tokens_seen": 587923456
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004150752256770311,
      "loss": 3.0058,
      "theoretical_loss": 3.848031219776334,
      "tokens_seen": 587988992
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004150651955867603,
      "loss": 3.0378,
      "theoretical_loss": 3.8479863101836145,
      "tokens_seen": 588054528
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041505516549648945,
      "loss": 2.7628,
      "theoretical_loss": 3.847941406996799,
      "tokens_seen": 588120064
    },
    {
      "epoch": 0.18,
      "objective/train/advantage_avg": 0.4954984188079834,
      "objective/train/docs_used": 342650,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.9131739139556885,
      "objective/train/original_loss": 2.9131741523742676,
      "objective/train/theoretical_loss": 3.8478965102142606,
      "objective/train/tokens_used": 608645600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24555814266204834,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507981777191162,
      "objective/train/weighted_lm_loss": 3.0610852241516113,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0449190139770508,
      "theoretical_loss": 3.8478965102142606,
      "tokens_seen": 588185600
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004150451354062187,
      "loss": 3.0013,
      "theoretical_loss": 3.8478965102142606,
      "tokens_seen": 588185600
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041503510531594787,
      "loss": 2.9352,
      "theoretical_loss": 3.8478516198343717,
      "tokens_seen": 588251136
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041502507522567705,
      "loss": 3.1833,
      "theoretical_loss": 3.8478067358555075,
      "tokens_seen": 588316672
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041501504513540623,
      "loss": 2.7955,
      "theoretical_loss": 3.8477618582760407,
      "tokens_seen": 588382208
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004150050150451354,
      "loss": 3.164,
      "theoretical_loss": 3.847716987094347,
      "tokens_seen": 588447744
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004149949849548646,
      "loss": 3.0814,
      "theoretical_loss": 3.8476721223088024,
      "tokens_seen": 588513280
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041498495486459383,
      "loss": 2.8312,
      "theoretical_loss": 3.8476272639177824,
      "tokens_seen": 588578816
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041497492477432296,
      "loss": 3.0698,
      "theoretical_loss": 3.8475824119196638,
      "tokens_seen": 588644352
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004149648946840522,
      "loss": 3.067,
      "theoretical_loss": 3.8475375663128246,
      "tokens_seen": 588709888
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004149548645937813,
      "loss": 2.9131,
      "theoretical_loss": 3.8474927270956423,
      "tokens_seen": 588775424
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041494483450351055,
      "loss": 2.7424,
      "theoretical_loss": 3.8474478942664954,
      "tokens_seen": 588840960
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041493480441323974,
      "loss": 2.9721,
      "theoretical_loss": 3.8474030678237625,
      "tokens_seen": 588906496
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004149247743229689,
      "loss": 3.1884,
      "theoretical_loss": 3.8473582477658237,
      "tokens_seen": 588972032
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004149147442326981,
      "loss": 2.9017,
      "theoretical_loss": 3.8473134340910597,
      "tokens_seen": 589037568
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041490471414242733,
      "loss": 3.1136,
      "theoretical_loss": 3.847268626797851,
      "tokens_seen": 589103104
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041489468405215646,
      "loss": 3.1887,
      "theoretical_loss": 3.847223825884579,
      "tokens_seen": 589168640
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004148846539618857,
      "loss": 3.1823,
      "theoretical_loss": 3.847179031349626,
      "tokens_seen": 589234176
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004148746238716148,
      "loss": 3.1803,
      "theoretical_loss": 3.847134243191375,
      "tokens_seen": 589299712
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041486459378134406,
      "loss": 3.1674,
      "theoretical_loss": 3.847089461408208,
      "tokens_seen": 589365248
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041485456369107324,
      "loss": 2.8994,
      "theoretical_loss": 3.8470446859985103,
      "tokens_seen": 589430784
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004148445336008024,
      "loss": 2.8306,
      "theoretical_loss": 3.846999916960666,
      "tokens_seen": 589496320
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004148345035105316,
      "loss": 3.0096,
      "theoretical_loss": 3.8469551542930587,
      "tokens_seen": 589561856
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004148244734202608,
      "loss": 2.8579,
      "theoretical_loss": 3.8469103979940757,
      "tokens_seen": 589627392
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041481444332998996,
      "loss": 3.2163,
      "theoretical_loss": 3.846865648062103,
      "tokens_seen": 589692928
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004148044132397192,
      "loss": 2.8458,
      "theoretical_loss": 3.846820904495526,
      "tokens_seen": 589758464
    },
    {
      "epoch": 0.18,
      "objective/train/advantage_avg": 0.4951511323451996,
      "objective/train/docs_used": 343935,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.6106083393096924,
      "objective/train/original_loss": 2.6106081008911133,
      "objective/train/theoretical_loss": 3.8467761672927336,
      "objective/train/tokens_used": 610284000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2452385425567627,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507616996765137,
      "objective/train/weighted_lm_loss": 2.742924451828003,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0425469875335693,
      "theoretical_loss": 3.8467761672927336,
      "tokens_seen": 589824000
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004147943831494483,
      "loss": 3.1457,
      "theoretical_loss": 3.8467761672927336,
      "tokens_seen": 589824000
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041478435305917756,
      "loss": 3.2341,
      "theoretical_loss": 3.8467314364521132,
      "tokens_seen": 589889536
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004147743229689067,
      "loss": 2.8542,
      "theoretical_loss": 3.8466867119720534,
      "tokens_seen": 589955072
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004147642928786359,
      "loss": 2.8524,
      "theoretical_loss": 3.846641993850943,
      "tokens_seen": 590020608
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004147542627883651,
      "loss": 2.9484,
      "theoretical_loss": 3.846597282087172,
      "tokens_seen": 590086144
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004147442326980943,
      "loss": 2.8778,
      "theoretical_loss": 3.846552576679131,
      "tokens_seen": 590151680
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041473420260782347,
      "loss": 3.149,
      "theoretical_loss": 3.8465078776252097,
      "tokens_seen": 590217216
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004147241725175527,
      "loss": 2.9393,
      "theoretical_loss": 3.8464631849238016,
      "tokens_seen": 590282752
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041471414242728183,
      "loss": 2.8815,
      "theoretical_loss": 3.8464184985732968,
      "tokens_seen": 590348288
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041470411233701106,
      "loss": 3.1955,
      "theoretical_loss": 3.846373818572089,
      "tokens_seen": 590413824
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004146940822467402,
      "loss": 2.9884,
      "theoretical_loss": 3.8463291449185713,
      "tokens_seen": 590479360
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004146840521564694,
      "loss": 2.9491,
      "theoretical_loss": 3.846284477611137,
      "tokens_seen": 590544896
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004146740220661986,
      "loss": 3.1853,
      "theoretical_loss": 3.8462398166481817,
      "tokens_seen": 590610432
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004146639919759278,
      "loss": 3.125,
      "theoretical_loss": 3.846195162028099,
      "tokens_seen": 590675968
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041465396188565697,
      "loss": 3.1654,
      "theoretical_loss": 3.8461505137492855,
      "tokens_seen": 590741504
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041464393179538615,
      "loss": 2.8995,
      "theoretical_loss": 3.8461058718101366,
      "tokens_seen": 590807040
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041463390170511533,
      "loss": 3.1956,
      "theoretical_loss": 3.846061236209049,
      "tokens_seen": 590872576
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041462387161484457,
      "loss": 2.8803,
      "theoretical_loss": 3.8460166069444206,
      "tokens_seen": 590938112
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004146138415245737,
      "loss": 3.1583,
      "theoretical_loss": 3.84597198401465,
      "tokens_seen": 591003648
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041460381143430293,
      "loss": 2.8221,
      "theoretical_loss": 3.845927367418134,
      "tokens_seen": 591069184
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004145937813440321,
      "loss": 2.9167,
      "theoretical_loss": 3.845882757153272,
      "tokens_seen": 591134720
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004145837512537613,
      "loss": 2.7297,
      "theoretical_loss": 3.8458381532184647,
      "tokens_seen": 591200256
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004145737211634905,
      "loss": 3.236,
      "theoretical_loss": 3.8457935556121114,
      "tokens_seen": 591265792
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041456369107321965,
      "loss": 3.1004,
      "theoretical_loss": 3.8457489643326133,
      "tokens_seen": 591331328
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041455366098294884,
      "loss": 2.7284,
      "theoretical_loss": 3.845704379378372,
      "tokens_seen": 591396864
    },
    {
      "epoch": 0.18,
      "objective/train/advantage_avg": 0.49424272775650024,
      "objective/train/docs_used": 344744,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.88754940032959,
      "objective/train/original_loss": 2.8875491619110107,
      "objective/train/theoretical_loss": 3.8456598007477893,
      "objective/train/tokens_used": 611922400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24434421956539154,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506664514541626,
      "objective/train/weighted_lm_loss": 3.0338001251220703,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 3.8456598007477893,
      "tokens_seen": 591462400
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041454363089267807,
      "loss": 2.9559,
      "theoretical_loss": 3.8456598007477893,
      "tokens_seen": 591462400
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004145336008024072,
      "loss": 2.9817,
      "theoretical_loss": 3.845615228439267,
      "tokens_seen": 591527936
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041452357071213643,
      "loss": 2.7385,
      "theoretical_loss": 3.8455706624512094,
      "tokens_seen": 591593472
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041451354062186556,
      "loss": 2.8316,
      "theoretical_loss": 3.84552610278202,
      "tokens_seen": 591659008
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004145035105315948,
      "loss": 3.2164,
      "theoretical_loss": 3.845481549430102,
      "tokens_seen": 591724544
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.000414493480441324,
      "loss": 2.8609,
      "theoretical_loss": 3.8454370023938615,
      "tokens_seen": 591790080
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041448345035105316,
      "loss": 3.0938,
      "theoretical_loss": 3.845392461671704,
      "tokens_seen": 591855616
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041447342026078234,
      "loss": 2.9199,
      "theoretical_loss": 3.8453479272620346,
      "tokens_seen": 591921152
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004144633901705115,
      "loss": 2.7244,
      "theoretical_loss": 3.8453033991632606,
      "tokens_seen": 591986688
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004144533600802407,
      "loss": 2.8757,
      "theoretical_loss": 3.8452588773737886,
      "tokens_seen": 592052224
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041444332998996994,
      "loss": 2.9239,
      "theoretical_loss": 3.845214361892027,
      "tokens_seen": 592117760
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041443329989969906,
      "loss": 3.0304,
      "theoretical_loss": 3.845169852716384,
      "tokens_seen": 592183296
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004144232698094283,
      "loss": 2.8944,
      "theoretical_loss": 3.8451253498452687,
      "tokens_seen": 592248832
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041441323971915753,
      "loss": 2.7255,
      "theoretical_loss": 3.84508085327709,
      "tokens_seen": 592314368
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041440320962888666,
      "loss": 2.9607,
      "theoretical_loss": 3.8450363630102586,
      "tokens_seen": 592379904
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004143931795386159,
      "loss": 2.913,
      "theoretical_loss": 3.8449918790431843,
      "tokens_seen": 592445440
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.000414383149448345,
      "loss": 3.0274,
      "theoretical_loss": 3.84494740137428,
      "tokens_seen": 592510976
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041437311935807426,
      "loss": 2.6217,
      "theoretical_loss": 3.8449029300019557,
      "tokens_seen": 592576512
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041436308926780344,
      "loss": 3.0416,
      "theoretical_loss": 3.8448584649246245,
      "tokens_seen": 592642048
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004143530591775326,
      "loss": 3.1162,
      "theoretical_loss": 3.844814006140699,
      "tokens_seen": 592707584
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004143430290872618,
      "loss": 2.8232,
      "theoretical_loss": 3.844769553648594,
      "tokens_seen": 592773120
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.000414332998996991,
      "loss": 3.287,
      "theoretical_loss": 3.8447251074467226,
      "tokens_seen": 592838656
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041432296890672016,
      "loss": 3.1254,
      "theoretical_loss": 3.844680667533499,
      "tokens_seen": 592904192
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004143129388164494,
      "loss": 2.7245,
      "theoretical_loss": 3.8446362339073397,
      "tokens_seen": 592969728
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004143029087261785,
      "loss": 3.0164,
      "theoretical_loss": 3.84459180656666,
      "tokens_seen": 593035264
    },
    {
      "epoch": 0.18,
      "objective/train/advantage_avg": 0.49422386288642883,
      "objective/train/docs_used": 345976,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.982727289199829,
      "objective/train/original_loss": 2.982726812362671,
      "objective/train/theoretical_loss": 3.844547385509876,
      "objective/train/tokens_used": 613560800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24432706832885742,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506643056869507,
      "objective/train/weighted_lm_loss": 3.1338233947753906,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0253151655197144,
      "theoretical_loss": 3.844547385509876,
      "tokens_seen": 593100800
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041429287863590776,
      "loss": 3.2127,
      "theoretical_loss": 3.844547385509876,
      "tokens_seen": 593100800
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004142828485456369,
      "loss": 3.0185,
      "theoretical_loss": 3.8445029707354053,
      "tokens_seen": 593166336
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004142728184553661,
      "loss": 3.0986,
      "theoretical_loss": 3.8444585622416643,
      "tokens_seen": 593231872
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004142627883650953,
      "loss": 3.0101,
      "theoretical_loss": 3.844414160027072,
      "tokens_seen": 593297408
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004142527582748245,
      "loss": 2.9237,
      "theoretical_loss": 3.844369764090048,
      "tokens_seen": 593362944
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041424272818455367,
      "loss": 2.8899,
      "theoretical_loss": 3.84432537442901,
      "tokens_seen": 593428480
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004142326980942829,
      "loss": 2.8797,
      "theoretical_loss": 3.8442809910423783,
      "tokens_seen": 593494016
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041422266800401203,
      "loss": 3.136,
      "theoretical_loss": 3.844236613928574,
      "tokens_seen": 593559552
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041421263791374126,
      "loss": 2.7573,
      "theoretical_loss": 3.844192243086017,
      "tokens_seen": 593625088
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004142026078234704,
      "loss": 3.0849,
      "theoretical_loss": 3.8441478785131302,
      "tokens_seen": 593690624
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004141925777331996,
      "loss": 3.0881,
      "theoretical_loss": 3.8441035202083347,
      "tokens_seen": 593756160
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004141825476429288,
      "loss": 3.0182,
      "theoretical_loss": 3.844059168170053,
      "tokens_seen": 593821696
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.000414172517552658,
      "loss": 2.9593,
      "theoretical_loss": 3.8440148223967094,
      "tokens_seen": 593887232
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041416248746238717,
      "loss": 2.7364,
      "theoretical_loss": 3.843970482886727,
      "tokens_seen": 593952768
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041415245737211635,
      "loss": 3.3065,
      "theoretical_loss": 3.8439261496385306,
      "tokens_seen": 594018304
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041414242728184553,
      "loss": 3.0362,
      "theoretical_loss": 3.843881822650545,
      "tokens_seen": 594083840
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041413239719157477,
      "loss": 2.8666,
      "theoretical_loss": 3.843837501921196,
      "tokens_seen": 594149376
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004141223671013039,
      "loss": 3.0109,
      "theoretical_loss": 3.8437931874489095,
      "tokens_seen": 594214912
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041411233701103313,
      "loss": 2.9086,
      "theoretical_loss": 3.8437488792321126,
      "tokens_seen": 594280448
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004141023069207623,
      "loss": 3.1336,
      "theoretical_loss": 3.8437045772692313,
      "tokens_seen": 594345984
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004140922768304915,
      "loss": 2.9868,
      "theoretical_loss": 3.8436602815586953,
      "tokens_seen": 594411520
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004140822467402207,
      "loss": 2.9347,
      "theoretical_loss": 3.8436159920989317,
      "tokens_seen": 594477056
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041407221664994985,
      "loss": 3.2388,
      "theoretical_loss": 3.8435717088883696,
      "tokens_seen": 594542592
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041406218655967904,
      "loss": 2.8722,
      "theoretical_loss": 3.8435274319254393,
      "tokens_seen": 594608128
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041405215646940827,
      "loss": 2.6557,
      "theoretical_loss": 3.84348316120857,
      "tokens_seen": 594673664
    },
    {
      "epoch": 0.18,
      "objective/train/advantage_avg": 0.4937964379787445,
      "objective/train/docs_used": 346603,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.9970436096191406,
      "objective/train/original_loss": 2.9970433712005615,
      "objective/train/theoretical_loss": 3.8434388967361928,
      "objective/train/tokens_used": 615199200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24389666318893433,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506194829940796,
      "objective/train/weighted_lm_loss": 3.1488254070281982,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0432155132293701,
      "theoretical_loss": 3.8434388967361928,
      "tokens_seen": 594739200
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004140421263791374,
      "loss": 3.0626,
      "theoretical_loss": 3.8434388967361928,
      "tokens_seen": 594739200
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041403209628886663,
      "loss": 3.0161,
      "theoretical_loss": 3.843394638506739,
      "tokens_seen": 594804736
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041402206619859576,
      "loss": 3.1171,
      "theoretical_loss": 3.8433503865186402,
      "tokens_seen": 594870272
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.000414012036108325,
      "loss": 2.9113,
      "theoretical_loss": 3.8433061407703293,
      "tokens_seen": 594935808
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004140020060180542,
      "loss": 2.8551,
      "theoretical_loss": 3.843261901260239,
      "tokens_seen": 595001344
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041399197592778336,
      "loss": 2.9482,
      "theoretical_loss": 3.843217667986802,
      "tokens_seen": 595066880
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041398194583751254,
      "loss": 2.8973,
      "theoretical_loss": 3.843173440948454,
      "tokens_seen": 595132416
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004139719157472417,
      "loss": 3.0293,
      "theoretical_loss": 3.843129220143628,
      "tokens_seen": 595197952
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004139618856569709,
      "loss": 3.0773,
      "theoretical_loss": 3.8430850055707597,
      "tokens_seen": 595263488
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041395185556670014,
      "loss": 2.9843,
      "theoretical_loss": 3.8430407972282854,
      "tokens_seen": 595329024
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041394182547642926,
      "loss": 2.9135,
      "theoretical_loss": 3.8429965951146414,
      "tokens_seen": 595394560
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004139317953861585,
      "loss": 2.9811,
      "theoretical_loss": 3.842952399228264,
      "tokens_seen": 595460096
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004139217652958877,
      "loss": 2.8953,
      "theoretical_loss": 3.8429082095675913,
      "tokens_seen": 595525632
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041391173520561686,
      "loss": 2.8499,
      "theoretical_loss": 3.842864026131061,
      "tokens_seen": 595591168
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041390170511534604,
      "loss": 3.091,
      "theoretical_loss": 3.8428198489171117,
      "tokens_seen": 595656704
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004138916750250752,
      "loss": 2.9486,
      "theoretical_loss": 3.8427756779241826,
      "tokens_seen": 595722240
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004138816449348044,
      "loss": 3.3724,
      "theoretical_loss": 3.842731513150713,
      "tokens_seen": 595787776
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041387161484453364,
      "loss": 2.9661,
      "theoretical_loss": 3.842687354595144,
      "tokens_seen": 595853312
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041386158475426277,
      "loss": 3.1861,
      "theoretical_loss": 3.842643202255916,
      "tokens_seen": 595918848
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.000413851554663992,
      "loss": 2.9656,
      "theoretical_loss": 3.8425990561314705,
      "tokens_seen": 595984384
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041384152457372113,
      "loss": 2.7901,
      "theoretical_loss": 3.8425549162202493,
      "tokens_seen": 596049920
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041383149448345036,
      "loss": 2.727,
      "theoretical_loss": 3.8425107825206952,
      "tokens_seen": 596115456
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041382146439317954,
      "loss": 2.6459,
      "theoretical_loss": 3.842466655031251,
      "tokens_seen": 596180992
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004138114343029087,
      "loss": 2.9063,
      "theoretical_loss": 3.8424225337503604,
      "tokens_seen": 596246528
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004138014042126379,
      "loss": 3.0183,
      "theoretical_loss": 3.842378418676468,
      "tokens_seen": 596312064
    },
    {
      "epoch": 0.18,
      "objective/train/advantage_avg": 0.49269938468933105,
      "objective/train/docs_used": 348035,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.0026354789733887,
      "objective/train/original_loss": 3.0026357173919678,
      "objective/train/theoretical_loss": 3.8423343098080185,
      "objective/train/tokens_used": 616837600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2428898811340332,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505046844482422,
      "objective/train/weighted_lm_loss": 3.1544291973114014,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0396549701690674,
      "theoretical_loss": 3.8423343098080185,
      "tokens_seen": 596377600
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004137913741223671,
      "loss": 2.9679,
      "theoretical_loss": 3.8423343098080185,
      "tokens_seen": 596377600
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041378134403209627,
      "loss": 2.6532,
      "theoretical_loss": 3.8422902071434564,
      "tokens_seen": 596443136
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004137713139418255,
      "loss": 2.8809,
      "theoretical_loss": 3.842246110681229,
      "tokens_seen": 596508672
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041376128385155463,
      "loss": 3.154,
      "theoretical_loss": 3.8422020204197813,
      "tokens_seen": 596574208
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041375125376128387,
      "loss": 3.0702,
      "theoretical_loss": 3.8421579363575615,
      "tokens_seen": 596639744
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041374122367101305,
      "loss": 2.9801,
      "theoretical_loss": 3.8421138584930166,
      "tokens_seen": 596705280
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041373119358074223,
      "loss": 3.0133,
      "theoretical_loss": 3.8420697868245943,
      "tokens_seen": 596770816
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004137211634904714,
      "loss": 2.9254,
      "theoretical_loss": 3.8420257213507445,
      "tokens_seen": 596836352
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004137111334002006,
      "loss": 3.1234,
      "theoretical_loss": 3.841981662069916,
      "tokens_seen": 596901888
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041370110330992977,
      "loss": 2.7911,
      "theoretical_loss": 3.8419376089805573,
      "tokens_seen": 596967424
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.000413691073219659,
      "loss": 3.1743,
      "theoretical_loss": 3.841893562081121,
      "tokens_seen": 597032960
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041368104312938813,
      "loss": 2.8049,
      "theoretical_loss": 3.8418495213700563,
      "tokens_seen": 597098496
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041367101303911737,
      "loss": 2.9041,
      "theoretical_loss": 3.841805486845815,
      "tokens_seen": 597164032
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041366098294884655,
      "loss": 3.0066,
      "theoretical_loss": 3.8417614585068502,
      "tokens_seen": 597229568
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041365095285857573,
      "loss": 3.1446,
      "theoretical_loss": 3.8417174363516136,
      "tokens_seen": 597295104
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041364092276830497,
      "loss": 2.9185,
      "theoretical_loss": 3.8416734203785587,
      "tokens_seen": 597360640
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004136308926780341,
      "loss": 2.8966,
      "theoretical_loss": 3.8416294105861386,
      "tokens_seen": 597426176
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041362086258776333,
      "loss": 2.9048,
      "theoretical_loss": 3.8415854069728077,
      "tokens_seen": 597491712
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004136108324974925,
      "loss": 2.9324,
      "theoretical_loss": 3.8415414095370215,
      "tokens_seen": 597557248
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004136008024072217,
      "loss": 3.1123,
      "theoretical_loss": 3.8414974182772355,
      "tokens_seen": 597622784
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004135907723169509,
      "loss": 3.0563,
      "theoretical_loss": 3.841453433191904,
      "tokens_seen": 597688320
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041358074222668005,
      "loss": 2.8356,
      "theoretical_loss": 3.8414094542794857,
      "tokens_seen": 597753856
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041357071213640924,
      "loss": 3.1807,
      "theoretical_loss": 3.841365481538436,
      "tokens_seen": 597819392
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041356068204613847,
      "loss": 2.8558,
      "theoretical_loss": 3.841321514967213,
      "tokens_seen": 597884928
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004135506519558676,
      "loss": 2.9408,
      "theoretical_loss": 3.841277554564275,
      "tokens_seen": 597950464
    },
    {
      "epoch": 0.18,
      "objective/train/advantage_avg": 0.49250560998916626,
      "objective/train/docs_used": 348629,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.4963481426239014,
      "objective/train/original_loss": 2.4963483810424805,
      "objective/train/theoretical_loss": 3.8412336003280814,
      "objective/train/tokens_used": 618476000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24267712235450745,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.05048406124115,
      "objective/train/weighted_lm_loss": 2.6226627826690674,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0304546356201172,
      "theoretical_loss": 3.8412336003280814,
      "tokens_seen": 598016000
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041354062186559683,
      "loss": 2.6346,
      "theoretical_loss": 3.8412336003280814,
      "tokens_seen": 598016000
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041353059177532596,
      "loss": 3.0285,
      "theoretical_loss": 3.8411896522570896,
      "tokens_seen": 598081536
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004135205616850552,
      "loss": 2.8425,
      "theoretical_loss": 3.8411457103497613,
      "tokens_seen": 598147072
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004135105315947844,
      "loss": 2.9194,
      "theoretical_loss": 3.8411017746045557,
      "tokens_seen": 598212608
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041350050150451356,
      "loss": 2.8121,
      "theoretical_loss": 3.8410578450199346,
      "tokens_seen": 598278144
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041349047141424274,
      "loss": 3.1357,
      "theoretical_loss": 3.841013921594359,
      "tokens_seen": 598343680
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004134804413239719,
      "loss": 3.0298,
      "theoretical_loss": 3.840970004326291,
      "tokens_seen": 598409216
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004134704112337011,
      "loss": 2.8982,
      "theoretical_loss": 3.8409260932141924,
      "tokens_seen": 598474752
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041346038114343034,
      "loss": 3.1482,
      "theoretical_loss": 3.8408821882565274,
      "tokens_seen": 598540288
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041345035105315946,
      "loss": 3.141,
      "theoretical_loss": 3.84083828945176,
      "tokens_seen": 598605824
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004134403209628887,
      "loss": 2.8413,
      "theoretical_loss": 3.840794396798353,
      "tokens_seen": 598671360
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004134302908726179,
      "loss": 3.0112,
      "theoretical_loss": 3.8407505102947725,
      "tokens_seen": 598736896
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041342026078234706,
      "loss": 3.0589,
      "theoretical_loss": 3.8407066299394836,
      "tokens_seen": 598802432
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041341023069207624,
      "loss": 3.1517,
      "theoretical_loss": 3.8406627557309516,
      "tokens_seen": 598867968
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004134002006018054,
      "loss": 2.7996,
      "theoretical_loss": 3.840618887667643,
      "tokens_seen": 598933504
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004133901705115346,
      "loss": 3.0415,
      "theoretical_loss": 3.8405750257480253,
      "tokens_seen": 598999040
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041338014042126384,
      "loss": 3.0102,
      "theoretical_loss": 3.840531169970566,
      "tokens_seen": 599064576
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041337011033099297,
      "loss": 2.8955,
      "theoretical_loss": 3.840487320333733,
      "tokens_seen": 599130112
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004133600802407222,
      "loss": 2.9177,
      "theoretical_loss": 3.8404434768359943,
      "tokens_seen": 599195648
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041335005015045133,
      "loss": 2.8792,
      "theoretical_loss": 3.8403996394758204,
      "tokens_seen": 599261184
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041334002006018056,
      "loss": 2.9079,
      "theoretical_loss": 3.8403558082516804,
      "tokens_seen": 599326720
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041332998996990975,
      "loss": 3.1494,
      "theoretical_loss": 3.8403119831620445,
      "tokens_seen": 599392256
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004133199598796389,
      "loss": 3.154,
      "theoretical_loss": 3.8402681642053835,
      "tokens_seen": 599457792
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004133099297893681,
      "loss": 2.6964,
      "theoretical_loss": 3.8402243513801695,
      "tokens_seen": 599523328
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004132998996990973,
      "loss": 2.9374,
      "theoretical_loss": 3.8401805446848734,
      "tokens_seen": 599588864
    },
    {
      "epoch": 0.18,
      "objective/train/advantage_avg": 0.4919928014278412,
      "objective/train/docs_used": 349996,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.1843161582946777,
      "objective/train/original_loss": 3.184316396713257,
      "objective/train/theoretical_loss": 3.8401367441179683,
      "objective/train/tokens_used": 620114400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24215196073055267,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050430178642273,
      "objective/train/weighted_lm_loss": 3.3454315662384033,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0396549701690674,
      "theoretical_loss": 3.8401367441179683,
      "tokens_seen": 599654400
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041328986960882647,
      "loss": 3.1445,
      "theoretical_loss": 3.8401367441179683,
      "tokens_seen": 599654400
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004132798395185557,
      "loss": 3.0782,
      "theoretical_loss": 3.840092949677927,
      "tokens_seen": 599719936
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041326980942828483,
      "loss": 2.7462,
      "theoretical_loss": 3.840049161363223,
      "tokens_seen": 599785472
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041325977933801407,
      "loss": 3.0436,
      "theoretical_loss": 3.8400053791723305,
      "tokens_seen": 599851008
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041324974924774325,
      "loss": 3.2286,
      "theoretical_loss": 3.839961603103725,
      "tokens_seen": 599916544
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041323971915747243,
      "loss": 2.8301,
      "theoretical_loss": 3.83991783315588,
      "tokens_seen": 599982080
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004132296890672016,
      "loss": 3.1055,
      "theoretical_loss": 3.8398740693272733,
      "tokens_seen": 600047616
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004132196589769308,
      "loss": 2.7789,
      "theoretical_loss": 3.8398303116163794,
      "tokens_seen": 600113152
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041320962888665997,
      "loss": 2.8868,
      "theoretical_loss": 3.8397865600216763,
      "tokens_seen": 600178688
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004131995987963892,
      "loss": 3.326,
      "theoretical_loss": 3.8397428145416406,
      "tokens_seen": 600244224
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041318956870611833,
      "loss": 2.8209,
      "theoretical_loss": 3.8396990751747513,
      "tokens_seen": 600309760
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041317953861584757,
      "loss": 2.7649,
      "theoretical_loss": 3.839655341919486,
      "tokens_seen": 600375296
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004131695085255767,
      "loss": 3.1701,
      "theoretical_loss": 3.8396116147743236,
      "tokens_seen": 600440832
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041315947843530593,
      "loss": 2.8097,
      "theoretical_loss": 3.8395678937377444,
      "tokens_seen": 600506368
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004131494483450351,
      "loss": 3.1384,
      "theoretical_loss": 3.839524178808228,
      "tokens_seen": 600571904
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004131394182547643,
      "loss": 2.8209,
      "theoretical_loss": 3.839480469984256,
      "tokens_seen": 600637440
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004131293881644935,
      "loss": 2.8981,
      "theoretical_loss": 3.839436767264308,
      "tokens_seen": 600702976
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004131193580742227,
      "loss": 2.6619,
      "theoretical_loss": 3.8393930706468673,
      "tokens_seen": 600768512
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041310932798395184,
      "loss": 3.0827,
      "theoretical_loss": 3.839349380130415,
      "tokens_seen": 600834048
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004130992978936811,
      "loss": 2.891,
      "theoretical_loss": 3.839305695713435,
      "tokens_seen": 600899584
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004130892678034102,
      "loss": 2.8059,
      "theoretical_loss": 3.83926201739441,
      "tokens_seen": 600965120
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041307923771313944,
      "loss": 3.0165,
      "theoretical_loss": 3.8392183451718243,
      "tokens_seen": 601030656
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004130692076228686,
      "loss": 2.6644,
      "theoretical_loss": 3.839174679044162,
      "tokens_seen": 601096192
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004130591775325978,
      "loss": 2.7709,
      "theoretical_loss": 3.839131019009908,
      "tokens_seen": 601161728
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.000413049147442327,
      "loss": 2.9163,
      "theoretical_loss": 3.8390873650675488,
      "tokens_seen": 601227264
    },
    {
      "epoch": 0.18,
      "objective/train/advantage_avg": 0.49476358294487,
      "objective/train/docs_used": 350686,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.6892166137695312,
      "objective/train/original_loss": 2.689216375350952,
      "objective/train/theoretical_loss": 3.839043717215569,
      "objective/train/tokens_used": 621752800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2448757290840149,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507210493087769,
      "objective/train/weighted_lm_loss": 2.8256630897521973,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0425469875335693,
      "theoretical_loss": 3.839043717215569,
      "tokens_seen": 601292800
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041303911735205616,
      "loss": 2.9256,
      "theoretical_loss": 3.839043717215569,
      "tokens_seen": 601292800
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041302908726178534,
      "loss": 2.8058,
      "theoretical_loss": 3.839000075452457,
      "tokens_seen": 601358336
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004130190571715146,
      "loss": 3.0335,
      "theoretical_loss": 3.8389564397766986,
      "tokens_seen": 601423872
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004130090270812437,
      "loss": 2.9922,
      "theoretical_loss": 3.838912810186782,
      "tokens_seen": 601489408
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041299899699097294,
      "loss": 2.9188,
      "theoretical_loss": 3.8388691866811957,
      "tokens_seen": 601554944
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041298896690070207,
      "loss": 2.8292,
      "theoretical_loss": 3.838825569258428,
      "tokens_seen": 601620480
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004129789368104313,
      "loss": 2.9065,
      "theoretical_loss": 3.8387819579169684,
      "tokens_seen": 601686016
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004129689067201605,
      "loss": 2.9821,
      "theoretical_loss": 3.838738352655307,
      "tokens_seen": 601751552
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041295887662988966,
      "loss": 2.9384,
      "theoretical_loss": 3.838694753471934,
      "tokens_seen": 601817088
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041294884653961884,
      "loss": 2.8707,
      "theoretical_loss": 3.838651160365341,
      "tokens_seen": 601882624
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004129388164493481,
      "loss": 3.0539,
      "theoretical_loss": 3.838607573334018,
      "tokens_seen": 601948160
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004129287863590772,
      "loss": 2.8431,
      "theoretical_loss": 3.838563992376459,
      "tokens_seen": 602013696
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041291875626880644,
      "loss": 2.8577,
      "theoretical_loss": 3.8385204174911554,
      "tokens_seen": 602079232
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004129087261785356,
      "loss": 2.8793,
      "theoretical_loss": 3.8384768486765997,
      "tokens_seen": 602144768
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004128986960882648,
      "loss": 2.8166,
      "theoretical_loss": 3.838433285931287,
      "tokens_seen": 602210304
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041288866599799404,
      "loss": 2.9195,
      "theoretical_loss": 3.83838972925371,
      "tokens_seen": 602275840
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041287863590772317,
      "loss": 3.0169,
      "theoretical_loss": 3.8383461786423654,
      "tokens_seen": 602341376
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004128686058174524,
      "loss": 2.7301,
      "theoretical_loss": 3.8383026340957462,
      "tokens_seen": 602406912
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041285857572718153,
      "loss": 2.8883,
      "theoretical_loss": 3.83825909561235,
      "tokens_seen": 602472448
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041284854563691076,
      "loss": 2.819,
      "theoretical_loss": 3.838215563190672,
      "tokens_seen": 602537984
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041283851554663995,
      "loss": 3.1376,
      "theoretical_loss": 3.83817203682921,
      "tokens_seen": 602603520
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004128284854563691,
      "loss": 2.8854,
      "theoretical_loss": 3.83812851652646,
      "tokens_seen": 602669056
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004128184553660983,
      "loss": 2.8926,
      "theoretical_loss": 3.8380850022809216,
      "tokens_seen": 602734592
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004128084252758275,
      "loss": 2.8482,
      "theoretical_loss": 3.838041494091093,
      "tokens_seen": 602800128
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041279839518555667,
      "loss": 2.9126,
      "theoretical_loss": 3.8379979919554716,
      "tokens_seen": 602865664
    },
    {
      "epoch": 0.18,
      "objective/train/advantage_avg": 0.4945717453956604,
      "objective/train/docs_used": 352047,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.8834617137908936,
      "objective/train/original_loss": 2.8834619522094727,
      "objective/train/theoretical_loss": 3.837954495872559,
      "objective/train/tokens_used": 623391200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24465975165367126,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507007837295532,
      "objective/train/weighted_lm_loss": 3.0296597480773926,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0352214574813843,
      "theoretical_loss": 3.837954495872559,
      "tokens_seen": 602931200
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004127883650952859,
      "loss": 2.8988,
      "theoretical_loss": 3.837954495872559,
      "tokens_seen": 602931200
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041277833500501503,
      "loss": 2.7889,
      "theoretical_loss": 3.837911005840854,
      "tokens_seen": 602996736
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041276830491474427,
      "loss": 2.8562,
      "theoretical_loss": 3.8378675218588576,
      "tokens_seen": 603062272
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041275827482447345,
      "loss": 3.0307,
      "theoretical_loss": 3.837824043925071,
      "tokens_seen": 603127808
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041274824473420263,
      "loss": 3.0429,
      "theoretical_loss": 3.837780572037995,
      "tokens_seen": 603193344
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004127382146439318,
      "loss": 2.9563,
      "theoretical_loss": 3.8377371061961343,
      "tokens_seen": 603258880
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.000412728184553661,
      "loss": 3.0735,
      "theoretical_loss": 3.837693646397989,
      "tokens_seen": 603324416
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041271815446339017,
      "loss": 3.1312,
      "theoretical_loss": 3.8376501926420636,
      "tokens_seen": 603389952
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004127081243731194,
      "loss": 2.9716,
      "theoretical_loss": 3.8376067449268616,
      "tokens_seen": 603455488
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041269809428284854,
      "loss": 2.9603,
      "theoretical_loss": 3.837563303250888,
      "tokens_seen": 603521024
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041268806419257777,
      "loss": 2.8295,
      "theoretical_loss": 3.837519867612647,
      "tokens_seen": 603586560
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004126780341023069,
      "loss": 2.9641,
      "theoretical_loss": 3.8374764380106434,
      "tokens_seen": 603652096
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041266800401203613,
      "loss": 2.894,
      "theoretical_loss": 3.837433014443385,
      "tokens_seen": 603717632
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004126579739217653,
      "loss": 2.7361,
      "theoretical_loss": 3.837389596909377,
      "tokens_seen": 603783168
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004126479438314945,
      "loss": 2.8631,
      "theoretical_loss": 3.8373461854071262,
      "tokens_seen": 603848704
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004126379137412237,
      "loss": 2.9649,
      "theoretical_loss": 3.837302779935141,
      "tokens_seen": 603914240
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004126278836509529,
      "loss": 3.1098,
      "theoretical_loss": 3.837259380491929,
      "tokens_seen": 603979776
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041261785356068204,
      "loss": 2.8497,
      "theoretical_loss": 3.8372159870759988,
      "tokens_seen": 604045312
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004126078234704113,
      "loss": 2.9305,
      "theoretical_loss": 3.83717259968586,
      "tokens_seen": 604110848
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004125977933801404,
      "loss": 3.021,
      "theoretical_loss": 3.8371292183200216,
      "tokens_seen": 604176384
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041258776328986964,
      "loss": 3.123,
      "theoretical_loss": 3.837085842976994,
      "tokens_seen": 604241920
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004125777331995988,
      "loss": 3.0998,
      "theoretical_loss": 3.837042473655288,
      "tokens_seen": 604307456
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.000412567703109328,
      "loss": 2.9332,
      "theoretical_loss": 3.836999110353415,
      "tokens_seen": 604372992
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004125576730190572,
      "loss": 3.0441,
      "theoretical_loss": 3.836955753069887,
      "tokens_seen": 604438528
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041254764292878636,
      "loss": 3.0546,
      "theoretical_loss": 3.8369124018032155,
      "tokens_seen": 604504064
    },
    {
      "epoch": 0.18,
      "objective/train/advantage_avg": 0.49409306049346924,
      "objective/train/docs_used": 352465,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.991288900375366,
      "objective/train/original_loss": 2.991288900375366,
      "objective/train/theoretical_loss": 3.8368690565519143,
      "objective/train/tokens_used": 625029600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2442009150981903,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050650715827942,
      "objective/train/weighted_lm_loss": 3.1427950859069824,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0428944826126099,
      "theoretical_loss": 3.8368690565519143,
      "tokens_seen": 604569600
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041253761283851554,
      "loss": 2.9153,
      "theoretical_loss": 3.8368690565519143,
      "tokens_seen": 604569600
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004125275827482448,
      "loss": 2.9188,
      "theoretical_loss": 3.836825717314496,
      "tokens_seen": 604635136
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004125175526579739,
      "loss": 3.2901,
      "theoretical_loss": 3.8367823840894753,
      "tokens_seen": 604700672
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041250752256770314,
      "loss": 2.9556,
      "theoretical_loss": 3.8367390568753654,
      "tokens_seen": 604766208
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041249749247743227,
      "loss": 3.1278,
      "theoretical_loss": 3.8366957356706832,
      "tokens_seen": 604831744
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004124874623871615,
      "loss": 2.972,
      "theoretical_loss": 3.836652420473942,
      "tokens_seen": 604897280
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004124774322968907,
      "loss": 2.8284,
      "theoretical_loss": 3.8366091112836593,
      "tokens_seen": 604962816
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041246740220661986,
      "loss": 3.0764,
      "theoretical_loss": 3.836565808098351,
      "tokens_seen": 605028352
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041245737211634904,
      "loss": 3.1777,
      "theoretical_loss": 3.8365225109165344,
      "tokens_seen": 605093888
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004124473420260783,
      "loss": 2.7575,
      "theoretical_loss": 3.8364792197367272,
      "tokens_seen": 605159424
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004124373119358074,
      "loss": 3.2322,
      "theoretical_loss": 3.8364359345574472,
      "tokens_seen": 605224960
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041242728184553664,
      "loss": 3.0751,
      "theoretical_loss": 3.8363926553772134,
      "tokens_seen": 605290496
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041241725175526577,
      "loss": 2.961,
      "theoretical_loss": 3.8363493821945447,
      "tokens_seen": 605356032
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.000412407221664995,
      "loss": 2.9987,
      "theoretical_loss": 3.8363061150079614,
      "tokens_seen": 605421568
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004123971915747242,
      "loss": 2.8477,
      "theoretical_loss": 3.836262853815983,
      "tokens_seen": 605487104
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041238716148445337,
      "loss": 3.0554,
      "theoretical_loss": 3.83621959861713,
      "tokens_seen": 605552640
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041237713139418255,
      "loss": 3.1716,
      "theoretical_loss": 3.8361763494099246,
      "tokens_seen": 605618176
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041236710130391173,
      "loss": 2.8398,
      "theoretical_loss": 3.836133106192888,
      "tokens_seen": 605683712
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004123570712136409,
      "loss": 3.0087,
      "theoretical_loss": 3.8360898689645433,
      "tokens_seen": 605749248
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041234704112337015,
      "loss": 3.0447,
      "theoretical_loss": 3.836046637723412,
      "tokens_seen": 605814784
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041233701103309927,
      "loss": 2.785,
      "theoretical_loss": 3.8360034124680187,
      "tokens_seen": 605880320
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004123269809428285,
      "loss": 2.7775,
      "theoretical_loss": 3.8359601931968865,
      "tokens_seen": 605945856
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041231695085255763,
      "loss": 3.0185,
      "theoretical_loss": 3.83591697990854,
      "tokens_seen": 606011392
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041230692076228687,
      "loss": 2.6576,
      "theoretical_loss": 3.835873772601505,
      "tokens_seen": 606076928
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041229689067201605,
      "loss": 2.8404,
      "theoretical_loss": 3.8358305712743057,
      "tokens_seen": 606142464
    },
    {
      "epoch": 0.18,
      "objective/train/advantage_avg": 0.492619127035141,
      "objective/train/docs_used": 353634,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.959139108657837,
      "objective/train/original_loss": 2.9591386318206787,
      "objective/train/theoretical_loss": 3.8357873759254693,
      "objective/train/tokens_used": 626668000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24283051490783691,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0504963397979736,
      "objective/train/weighted_lm_loss": 3.1082518100738525,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0363596677780151,
      "theoretical_loss": 3.8357873759254693,
      "tokens_seen": 606208000
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041228686058174523,
      "loss": 2.8593,
      "theoretical_loss": 3.8357873759254693,
      "tokens_seen": 606208000
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004122768304914744,
      "loss": 2.8867,
      "theoretical_loss": 3.835744186553521,
      "tokens_seen": 606273536
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041226680040120365,
      "loss": 2.9337,
      "theoretical_loss": 3.835701003156989,
      "tokens_seen": 606339072
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004122567703109328,
      "loss": 2.7781,
      "theoretical_loss": 3.8356578257344003,
      "tokens_seen": 606404608
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.000412246740220662,
      "loss": 3.123,
      "theoretical_loss": 3.8356146542842833,
      "tokens_seen": 606470144
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041223671013039114,
      "loss": 2.7998,
      "theoretical_loss": 3.835571488805166,
      "tokens_seen": 606535680
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004122266800401204,
      "loss": 3.0589,
      "theoretical_loss": 3.835528329295578,
      "tokens_seen": 606601216
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041221664994984955,
      "loss": 2.9971,
      "theoretical_loss": 3.835485175754049,
      "tokens_seen": 606666752
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041220661985957874,
      "loss": 2.6524,
      "theoretical_loss": 3.8354420281791093,
      "tokens_seen": 606732288
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004121965897693079,
      "loss": 3.2703,
      "theoretical_loss": 3.8353988865692887,
      "tokens_seen": 606797824
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004121865596790371,
      "loss": 3.003,
      "theoretical_loss": 3.835355750923119,
      "tokens_seen": 606863360
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004121765295887663,
      "loss": 3.145,
      "theoretical_loss": 3.835312621239132,
      "tokens_seen": 606928896
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004121664994984955,
      "loss": 2.9327,
      "theoretical_loss": 3.8352694975158608,
      "tokens_seen": 606994432
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004121564694082247,
      "loss": 3.0465,
      "theoretical_loss": 3.8352263797518367,
      "tokens_seen": 607059968
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004121464393179539,
      "loss": 3.1196,
      "theoretical_loss": 3.8351832679455935,
      "tokens_seen": 607125504
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004121364092276831,
      "loss": 3.071,
      "theoretical_loss": 3.835140162095665,
      "tokens_seen": 607191040
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041212637913741224,
      "loss": 2.8521,
      "theoretical_loss": 3.8350970622005858,
      "tokens_seen": 607256576
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004121163490471415,
      "loss": 3.0121,
      "theoretical_loss": 3.8350539682588902,
      "tokens_seen": 607322112
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004121063189568706,
      "loss": 2.9881,
      "theoretical_loss": 3.835010880269114,
      "tokens_seen": 607387648
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041209628886659984,
      "loss": 3.0413,
      "theoretical_loss": 3.834967798229793,
      "tokens_seen": 607453184
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.000412086258776329,
      "loss": 2.9569,
      "theoretical_loss": 3.834924722139464,
      "tokens_seen": 607518720
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004120762286860582,
      "loss": 3.0289,
      "theoretical_loss": 3.834881651996663,
      "tokens_seen": 607584256
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004120661985957874,
      "loss": 2.7218,
      "theoretical_loss": 3.8348385877999283,
      "tokens_seen": 607649792
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041205616850551656,
      "loss": 3.3014,
      "theoretical_loss": 3.8347955295477973,
      "tokens_seen": 607715328
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041204613841524574,
      "loss": 3.1745,
      "theoretical_loss": 3.834752477238809,
      "tokens_seen": 607780864
    },
    {
      "epoch": 0.18,
      "objective/train/advantage_avg": 0.49481290578842163,
      "objective/train/docs_used": 354263,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.5264368057250977,
      "objective/train/original_loss": 2.5264368057250977,
      "objective/train/theoretical_loss": 3.8347094308715017,
      "objective/train/tokens_used": 628306400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2448999285697937,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507261753082275,
      "objective/train/weighted_lm_loss": 2.6545095443725586,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0458475351333618,
      "theoretical_loss": 3.8347094308715017,
      "tokens_seen": 607846400
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.000412036108324975,
      "loss": 2.8539,
      "theoretical_loss": 3.8347094308715017,
      "tokens_seen": 607846400
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004120260782347041,
      "loss": 3.0946,
      "theoretical_loss": 3.834666390444415,
      "tokens_seen": 607911936
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041201604814443334,
      "loss": 3.0349,
      "theoretical_loss": 3.8346233559560896,
      "tokens_seen": 607977472
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041200601805416247,
      "loss": 2.9482,
      "theoretical_loss": 3.8345803274050656,
      "tokens_seen": 608043008
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004119959879638917,
      "loss": 3.0084,
      "theoretical_loss": 3.834537304789884,
      "tokens_seen": 608108544
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004119859578736209,
      "loss": 2.8161,
      "theoretical_loss": 3.834494288109086,
      "tokens_seen": 608174080
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041197592778335006,
      "loss": 2.8162,
      "theoretical_loss": 3.8344512773612154,
      "tokens_seen": 608239616
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041196589769307924,
      "loss": 3.151,
      "theoretical_loss": 3.8344082725448123,
      "tokens_seen": 608305152
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004119558676028085,
      "loss": 2.7836,
      "theoretical_loss": 3.8343652736584213,
      "tokens_seen": 608370688
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004119458375125376,
      "loss": 3.0284,
      "theoretical_loss": 3.8343222807005866,
      "tokens_seen": 608436224
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041193580742226684,
      "loss": 3.0704,
      "theoretical_loss": 3.8342792936698507,
      "tokens_seen": 608501760
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041192577733199597,
      "loss": 3.1985,
      "theoretical_loss": 3.8342363125647587,
      "tokens_seen": 608567296
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004119157472417252,
      "loss": 2.6786,
      "theoretical_loss": 3.834193337383857,
      "tokens_seen": 608632832
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004119057171514544,
      "loss": 2.7307,
      "theoretical_loss": 3.83415036812569,
      "tokens_seen": 608698368
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041189568706118357,
      "loss": 3.3189,
      "theoretical_loss": 3.8341074047888046,
      "tokens_seen": 608763904
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041188565697091275,
      "loss": 3.0942,
      "theoretical_loss": 3.8340644473717465,
      "tokens_seen": 608829440
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041187562688064193,
      "loss": 3.0652,
      "theoretical_loss": 3.8340214958730643,
      "tokens_seen": 608894976
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004118655967903711,
      "loss": 3.0703,
      "theoretical_loss": 3.833978550291305,
      "tokens_seen": 608960512
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041185556670010035,
      "loss": 3.0571,
      "theoretical_loss": 3.833935610625017,
      "tokens_seen": 609026048
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041184553660982947,
      "loss": 2.9045,
      "theoretical_loss": 3.8338926768727486,
      "tokens_seen": 609091584
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004118355065195587,
      "loss": 2.9496,
      "theoretical_loss": 3.8338497490330505,
      "tokens_seen": 609157120
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041182547642928783,
      "loss": 2.7427,
      "theoretical_loss": 3.8338068271044703,
      "tokens_seen": 609222656
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041181544633901707,
      "loss": 2.9116,
      "theoretical_loss": 3.83376391108556,
      "tokens_seen": 609288192
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041180541624874625,
      "loss": 3.1578,
      "theoretical_loss": 3.8337210009748697,
      "tokens_seen": 609353728
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041179538615847543,
      "loss": 2.9417,
      "theoretical_loss": 3.833678096770951,
      "tokens_seen": 609419264
    },
    {
      "epoch": 0.18,
      "objective/train/advantage_avg": 0.49487483501434326,
      "objective/train/docs_used": 355721,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.403420925140381,
      "objective/train/original_loss": 2.40342116355896,
      "objective/train/theoretical_loss": 3.833635198472356,
      "objective/train/tokens_used": 629944800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24495336413383484,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507328510284424,
      "objective/train/weighted_lm_loss": 2.525299549102783,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0363596677780151,
      "theoretical_loss": 3.833635198472356,
      "tokens_seen": 609484800
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004117853560682046,
      "loss": 2.9828,
      "theoretical_loss": 3.833635198472356,
      "tokens_seen": 609484800
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041177532597793385,
      "loss": 3.2071,
      "theoretical_loss": 3.8335923060776365,
      "tokens_seen": 609550336
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.000411765295887663,
      "loss": 2.9593,
      "theoretical_loss": 3.8335494195853452,
      "tokens_seen": 609615872
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004117552657973922,
      "loss": 2.9757,
      "theoretical_loss": 3.833506538994036,
      "tokens_seen": 609681408
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041174523570712134,
      "loss": 3.2515,
      "theoretical_loss": 3.8334636643022626,
      "tokens_seen": 609746944
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004117352056168506,
      "loss": 2.843,
      "theoretical_loss": 3.8334207955085793,
      "tokens_seen": 609812480
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041172517552657975,
      "loss": 3.2277,
      "theoretical_loss": 3.833377932611541,
      "tokens_seen": 609878016
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041171514543630894,
      "loss": 3.0111,
      "theoretical_loss": 3.8333350756097033,
      "tokens_seen": 609943552
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004117051153460381,
      "loss": 3.1555,
      "theoretical_loss": 3.833292224501622,
      "tokens_seen": 610009088
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004116950852557673,
      "loss": 2.9308,
      "theoretical_loss": 3.833249379285853,
      "tokens_seen": 610074624
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004116850551654965,
      "loss": 2.8226,
      "theoretical_loss": 3.833206539960954,
      "tokens_seen": 610140160
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004116750250752257,
      "loss": 2.8797,
      "theoretical_loss": 3.8331637065254824,
      "tokens_seen": 610205696
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041166499498495484,
      "loss": 2.8591,
      "theoretical_loss": 3.8331208789779954,
      "tokens_seen": 610271232
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004116549648946841,
      "loss": 3.0228,
      "theoretical_loss": 3.833078057317052,
      "tokens_seen": 610336768
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.0004116449348044132,
      "loss": 2.9492,
      "theoretical_loss": 3.8330352415412117,
      "tokens_seen": 610402304
    },
    {
      "epoch": 0.18,
      "learning_rate": 0.00041163490471414244,
      "loss": 2.9384,
      "theoretical_loss": 3.832992431649033,
      "tokens_seen": 610467840
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004116248746238716,
      "loss": 2.9245,
      "theoretical_loss": 3.8329496276390764,
      "tokens_seen": 610533376
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004116148445336008,
      "loss": 2.8453,
      "theoretical_loss": 3.832906829509902,
      "tokens_seen": 610598912
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00041160481444333,
      "loss": 2.877,
      "theoretical_loss": 3.832864037260071,
      "tokens_seen": 610664448
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004115947843530592,
      "loss": 2.9677,
      "theoretical_loss": 3.832821250888145,
      "tokens_seen": 610729984
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00041158475426278834,
      "loss": 2.797,
      "theoretical_loss": 3.832778470392686,
      "tokens_seen": 610795520
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004115747241725176,
      "loss": 3.0023,
      "theoretical_loss": 3.8327356957722563,
      "tokens_seen": 610861056
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004115646940822467,
      "loss": 2.9905,
      "theoretical_loss": 3.8326929270254193,
      "tokens_seen": 610926592
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00041155466399197594,
      "loss": 2.8018,
      "theoretical_loss": 3.8326501641507376,
      "tokens_seen": 610992128
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004115446339017051,
      "loss": 3.0205,
      "theoretical_loss": 3.832607407146777,
      "tokens_seen": 611057664
    },
    {
      "epoch": 0.19,
      "objective/train/advantage_avg": 0.49341192841529846,
      "objective/train/docs_used": 356348,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.8695499897003174,
      "objective/train/original_loss": 2.8695504665374756,
      "objective/train/theoretical_loss": 3.8325646560121003,
      "objective/train/tokens_used": 631583200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24351949989795685,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505790710449219,
      "objective/train/weighted_lm_loss": 3.014704704284668,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0432155132293701,
      "theoretical_loss": 3.8325646560121003,
      "tokens_seen": 611123200
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004115346038114343,
      "loss": 3.2144,
      "theoretical_loss": 3.8325646560121003,
      "tokens_seen": 611123200
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004115245737211635,
      "loss": 2.7369,
      "theoretical_loss": 3.832521910745273,
      "tokens_seen": 611188736
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00041151454363089267,
      "loss": 3.0861,
      "theoretical_loss": 3.8324791713448603,
      "tokens_seen": 611254272
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00041150451354062185,
      "loss": 2.891,
      "theoretical_loss": 3.83243643780943,
      "tokens_seen": 611319808
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004114944834503511,
      "loss": 3.0952,
      "theoretical_loss": 3.8323937101375467,
      "tokens_seen": 611385344
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004114844533600802,
      "loss": 3.0197,
      "theoretical_loss": 3.8323509883277778,
      "tokens_seen": 611450880
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00041147442326980944,
      "loss": 2.889,
      "theoretical_loss": 3.8323082723786914,
      "tokens_seen": 611516416
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00041146439317953857,
      "loss": 2.7527,
      "theoretical_loss": 3.832265562288856,
      "tokens_seen": 611581952
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004114543630892678,
      "loss": 2.8419,
      "theoretical_loss": 3.8322228580568387,
      "tokens_seen": 611647488
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.000411444332998997,
      "loss": 2.9321,
      "theoretical_loss": 3.8321801596812097,
      "tokens_seen": 611713024
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00041143430290872617,
      "loss": 3.0371,
      "theoretical_loss": 3.832137467160538,
      "tokens_seen": 611778560
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00041142427281845535,
      "loss": 3.1545,
      "theoretical_loss": 3.8320947804933945,
      "tokens_seen": 611844096
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004114142427281846,
      "loss": 3.1331,
      "theoretical_loss": 3.8320520996783483,
      "tokens_seen": 611909632
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00041140421263791377,
      "loss": 2.8731,
      "theoretical_loss": 3.832009424713972,
      "tokens_seen": 611975168
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00041139418254764295,
      "loss": 3.0396,
      "theoretical_loss": 3.831966755598837,
      "tokens_seen": 612040704
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00041138415245737213,
      "loss": 2.7858,
      "theoretical_loss": 3.831924092331514,
      "tokens_seen": 612106240
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004113741223671013,
      "loss": 3.0872,
      "theoretical_loss": 3.831881434910577,
      "tokens_seen": 612171776
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00041136409227683055,
      "loss": 2.9538,
      "theoretical_loss": 3.8318387833345984,
      "tokens_seen": 612237312
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00041135406218655967,
      "loss": 2.8256,
      "theoretical_loss": 3.8317961376021517,
      "tokens_seen": 612302848
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004113440320962889,
      "loss": 2.8364,
      "theoretical_loss": 3.8317534977118117,
      "tokens_seen": 612368384
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00041133400200601803,
      "loss": 2.9313,
      "theoretical_loss": 3.8317108636621526,
      "tokens_seen": 612433920
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00041132397191574727,
      "loss": 2.9137,
      "theoretical_loss": 3.831668235451749,
      "tokens_seen": 612499456
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00041131394182547645,
      "loss": 2.9223,
      "theoretical_loss": 3.8316256130791775,
      "tokens_seen": 612564992
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00041130391173520563,
      "loss": 3.0074,
      "theoretical_loss": 3.8315829965430126,
      "tokens_seen": 612630528
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004112938816449348,
      "loss": 3.3702,
      "theoretical_loss": 3.8315403858418327,
      "tokens_seen": 612696064
    },
    {
      "epoch": 0.19,
      "objective/train/advantage_avg": 0.49583494663238525,
      "objective/train/docs_used": 356825,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.6894307136535645,
      "objective/train/original_loss": 2.6894302368164062,
      "objective/train/theoretical_loss": 3.831497780974214,
      "objective/train/tokens_used": 633221600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24590255320072174,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0508334636688232,
      "objective/train/weighted_lm_loss": 2.82619571685791,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0396549701690674,
      "theoretical_loss": 3.831497780974214,
      "tokens_seen": 612761600
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00041128385155466405,
      "loss": 2.9407,
      "theoretical_loss": 3.831497780974214,
      "tokens_seen": 612761600
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004112738214643932,
      "loss": 3.0441,
      "theoretical_loss": 3.8314551819387335,
      "tokens_seen": 612827136
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004112637913741224,
      "loss": 2.9445,
      "theoretical_loss": 3.8314125887339703,
      "tokens_seen": 612892672
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00041125376128385154,
      "loss": 3.1243,
      "theoretical_loss": 3.8313700013585024,
      "tokens_seen": 612958208
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004112437311935808,
      "loss": 3.3294,
      "theoretical_loss": 3.8313274198109095,
      "tokens_seen": 613023744
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00041123370110330995,
      "loss": 2.971,
      "theoretical_loss": 3.8312848440897707,
      "tokens_seen": 613089280
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00041122367101303914,
      "loss": 2.9607,
      "theoretical_loss": 3.8312422741936656,
      "tokens_seen": 613154816
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004112136409227683,
      "loss": 3.0692,
      "theoretical_loss": 3.831199710121176,
      "tokens_seen": 613220352
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004112036108324975,
      "loss": 2.8414,
      "theoretical_loss": 3.8311571518708813,
      "tokens_seen": 613285888
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004111935807422267,
      "loss": 2.8239,
      "theoretical_loss": 3.8311145994413653,
      "tokens_seen": 613351424
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004111835506519559,
      "loss": 3.0198,
      "theoretical_loss": 3.8310720528312077,
      "tokens_seen": 613416960
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00041117352056168504,
      "loss": 2.9307,
      "theoretical_loss": 3.831029512038993,
      "tokens_seen": 613482496
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004111634904714143,
      "loss": 2.8955,
      "theoretical_loss": 3.8309869770633025,
      "tokens_seen": 613548032
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004111534603811434,
      "loss": 3.1274,
      "theoretical_loss": 3.8309444479027217,
      "tokens_seen": 613613568
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00041114343029087264,
      "loss": 2.9662,
      "theoretical_loss": 3.8309019245558327,
      "tokens_seen": 613679104
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004111334002006018,
      "loss": 3.2154,
      "theoretical_loss": 3.8308594070212214,
      "tokens_seen": 613744640
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.000411123370110331,
      "loss": 3.1499,
      "theoretical_loss": 3.830816895297472,
      "tokens_seen": 613810176
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004111133400200602,
      "loss": 3.0148,
      "theoretical_loss": 3.830774389383171,
      "tokens_seen": 613875712
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004111033099297894,
      "loss": 2.9567,
      "theoretical_loss": 3.830731889276904,
      "tokens_seen": 613941248
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00041109327983951854,
      "loss": 2.874,
      "theoretical_loss": 3.8306893949772567,
      "tokens_seen": 614006784
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004110832497492478,
      "loss": 3.1267,
      "theoretical_loss": 3.830646906482817,
      "tokens_seen": 614072320
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004110732196589769,
      "loss": 3.0421,
      "theoretical_loss": 3.8306044237921726,
      "tokens_seen": 614137856
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00041106318956870614,
      "loss": 3.0334,
      "theoretical_loss": 3.830561946903911,
      "tokens_seen": 614203392
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004110531594784353,
      "loss": 2.9171,
      "theoretical_loss": 3.8305194758166206,
      "tokens_seen": 614268928
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004110431293881645,
      "loss": 3.0744,
      "theoretical_loss": 3.830477010528891,
      "tokens_seen": 614334464
    },
    {
      "epoch": 0.19,
      "objective/train/advantage_avg": 0.49538320302963257,
      "objective/train/docs_used": 357491,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.561123847961426,
      "objective/train/original_loss": 2.561123847961426,
      "objective/train/theoretical_loss": 3.8304345510393105,
      "objective/train/tokens_used": 634860000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24548153579235077,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507862567901611,
      "objective/train/weighted_lm_loss": 2.6910343170166016,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0363596677780151,
      "theoretical_loss": 3.8304345510393105,
      "tokens_seen": 614400000
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004110330992978937,
      "loss": 2.7969,
      "theoretical_loss": 3.8304345510393105,
      "tokens_seen": 614400000
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00041102306920762287,
      "loss": 3.2158,
      "theoretical_loss": 3.830392097346471,
      "tokens_seen": 614465536
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00041101303911735205,
      "loss": 3.0213,
      "theoretical_loss": 3.8303496494489613,
      "tokens_seen": 614531072
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004110030090270813,
      "loss": 2.9254,
      "theoretical_loss": 3.830307207345373,
      "tokens_seen": 614596608
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004109929789368104,
      "loss": 3.1167,
      "theoretical_loss": 3.8302647710342983,
      "tokens_seen": 614662144
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00041098294884653965,
      "loss": 3.0004,
      "theoretical_loss": 3.830222340514328,
      "tokens_seen": 614727680
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00041097291875626877,
      "loss": 2.97,
      "theoretical_loss": 3.830179915784054,
      "tokens_seen": 614793216
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.000410962888665998,
      "loss": 2.959,
      "theoretical_loss": 3.830137496842071,
      "tokens_seen": 614858752
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004109528585757272,
      "loss": 2.8121,
      "theoretical_loss": 3.830095083686972,
      "tokens_seen": 614924288
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00041094282848545637,
      "loss": 3.0612,
      "theoretical_loss": 3.83005267631735,
      "tokens_seen": 614989824
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00041093279839518555,
      "loss": 3.2098,
      "theoretical_loss": 3.8300102747318006,
      "tokens_seen": 615055360
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004109227683049148,
      "loss": 3.1771,
      "theoretical_loss": 3.829967878928917,
      "tokens_seen": 615120896
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004109127382146439,
      "loss": 2.9618,
      "theoretical_loss": 3.8299254889072962,
      "tokens_seen": 615186432
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00041090270812437315,
      "loss": 2.8472,
      "theoretical_loss": 3.8298831046655337,
      "tokens_seen": 615251968
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004108926780341023,
      "loss": 2.7992,
      "theoretical_loss": 3.8298407262022263,
      "tokens_seen": 615317504
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004108826479438315,
      "loss": 2.9715,
      "theoretical_loss": 3.8297983535159696,
      "tokens_seen": 615383040
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004108726178535607,
      "loss": 2.782,
      "theoretical_loss": 3.8297559866053614,
      "tokens_seen": 615448576
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00041086258776328987,
      "loss": 3.1105,
      "theoretical_loss": 3.8297136254690005,
      "tokens_seen": 615514112
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00041085255767301905,
      "loss": 2.8506,
      "theoretical_loss": 3.8296712701054845,
      "tokens_seen": 615579648
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00041084252758274823,
      "loss": 3.1018,
      "theoretical_loss": 3.8296289205134117,
      "tokens_seen": 615645184
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004108324974924774,
      "loss": 2.9386,
      "theoretical_loss": 3.8295865766913826,
      "tokens_seen": 615710720
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00041082246740220665,
      "loss": 3.0194,
      "theoretical_loss": 3.8295442386379968,
      "tokens_seen": 615776256
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004108124373119358,
      "loss": 2.8927,
      "theoretical_loss": 3.8295019063518536,
      "tokens_seen": 615841792
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.000410802407221665,
      "loss": 2.7635,
      "theoretical_loss": 3.8294595798315543,
      "tokens_seen": 615907328
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004107923771313942,
      "loss": 2.8216,
      "theoretical_loss": 3.8294172590757007,
      "tokens_seen": 615972864
    },
    {
      "epoch": 0.19,
      "objective/train/advantage_avg": 0.49148598313331604,
      "objective/train/docs_used": 358741,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.7719364166259766,
      "objective/train/original_loss": 2.7719359397888184,
      "objective/train/theoretical_loss": 3.829374944082894,
      "objective/train/tokens_used": 636498400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24167288839817047,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0503770112991333,
      "objective/train/weighted_lm_loss": 2.9115805625915527,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0408108234405518,
      "theoretical_loss": 3.829374944082894,
      "tokens_seen": 616038400
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004107823470411234,
      "loss": 3.0128,
      "theoretical_loss": 3.829374944082894,
      "tokens_seen": 616038400
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00041077231695085256,
      "loss": 2.7906,
      "theoretical_loss": 3.8293326348517365,
      "tokens_seen": 616103936
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00041076228686058174,
      "loss": 3.0478,
      "theoretical_loss": 3.8292903313808306,
      "tokens_seen": 616169472
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004107522567703109,
      "loss": 2.9501,
      "theoretical_loss": 3.829248033668781,
      "tokens_seen": 616235008
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00041074222668004015,
      "loss": 2.9196,
      "theoretical_loss": 3.8292057417141896,
      "tokens_seen": 616300544
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004107321965897693,
      "loss": 2.8731,
      "theoretical_loss": 3.8291634555156615,
      "tokens_seen": 616366080
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004107221664994985,
      "loss": 2.7984,
      "theoretical_loss": 3.8291211750718013,
      "tokens_seen": 616431616
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00041071213640922764,
      "loss": 2.7985,
      "theoretical_loss": 3.8290789003812145,
      "tokens_seen": 616497152
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004107021063189569,
      "loss": 3.1624,
      "theoretical_loss": 3.829036631442506,
      "tokens_seen": 616562688
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00041069207622868606,
      "loss": 2.7907,
      "theoretical_loss": 3.828994368254283,
      "tokens_seen": 616628224
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00041068204613841524,
      "loss": 3.2193,
      "theoretical_loss": 3.8289521108151505,
      "tokens_seen": 616693760
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004106720160481444,
      "loss": 3.1124,
      "theoretical_loss": 3.8289098591237174,
      "tokens_seen": 616759296
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004106619859578736,
      "loss": 2.978,
      "theoretical_loss": 3.8288676131785895,
      "tokens_seen": 616824832
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00041065195586760284,
      "loss": 3.1188,
      "theoretical_loss": 3.8288253729783763,
      "tokens_seen": 616890368
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.000410641925777332,
      "loss": 2.9387,
      "theoretical_loss": 3.828783138521686,
      "tokens_seen": 616955904
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004106318956870612,
      "loss": 3.0985,
      "theoretical_loss": 3.828740909807128,
      "tokens_seen": 617021440
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004106218655967904,
      "loss": 3.1113,
      "theoretical_loss": 3.8286986868333113,
      "tokens_seen": 617086976
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004106118355065196,
      "loss": 2.7493,
      "theoretical_loss": 3.8286564695988456,
      "tokens_seen": 617152512
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00041060180541624874,
      "loss": 2.9905,
      "theoretical_loss": 3.828614258102342,
      "tokens_seen": 617218048
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.000410591775325978,
      "loss": 3.1543,
      "theoretical_loss": 3.8285720523424116,
      "tokens_seen": 617283584
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004105817452357071,
      "loss": 2.9215,
      "theoretical_loss": 3.828529852317666,
      "tokens_seen": 617349120
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00041057171514543634,
      "loss": 3.1864,
      "theoretical_loss": 3.8284876580267158,
      "tokens_seen": 617414656
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004105616850551655,
      "loss": 3.2267,
      "theoretical_loss": 3.8284454694681753,
      "tokens_seen": 617480192
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004105516549648947,
      "loss": 2.9614,
      "theoretical_loss": 3.8284032866406563,
      "tokens_seen": 617545728
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004105416248746239,
      "loss": 3.0195,
      "theoretical_loss": 3.8283611095427723,
      "tokens_seen": 617611264
    },
    {
      "epoch": 0.19,
      "objective/train/advantage_avg": 0.49466273188591003,
      "objective/train/docs_used": 359404,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.728306531906128,
      "objective/train/original_loss": 2.728306770324707,
      "objective/train/theoretical_loss": 3.8283189381731377,
      "objective/train/tokens_used": 638136800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24474026262760162,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507103204727173,
      "objective/train/weighted_lm_loss": 2.8665640354156494,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0432155132293701,
      "theoretical_loss": 3.8283189381731377,
      "tokens_seen": 617676800
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00041053159478435307,
      "loss": 2.9453,
      "theoretical_loss": 3.8283189381731377,
      "tokens_seen": 617676800
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00041052156469408225,
      "loss": 3.0666,
      "theoretical_loss": 3.8282767725303666,
      "tokens_seen": 617742336
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004105115346038115,
      "loss": 3.0843,
      "theoretical_loss": 3.8282346126130733,
      "tokens_seen": 617807872
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004105015045135406,
      "loss": 2.9531,
      "theoretical_loss": 3.828192458419874,
      "tokens_seen": 617873408
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00041049147442326985,
      "loss": 3.0632,
      "theoretical_loss": 3.8281503099493843,
      "tokens_seen": 617938944
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00041048144433299897,
      "loss": 3.1819,
      "theoretical_loss": 3.82810816720022,
      "tokens_seen": 618004480
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004104714142427282,
      "loss": 3.0659,
      "theoretical_loss": 3.828066030170999,
      "tokens_seen": 618070016
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004104613841524574,
      "loss": 2.9982,
      "theoretical_loss": 3.8280238988603363,
      "tokens_seen": 618135552
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00041045135406218657,
      "loss": 3.0423,
      "theoretical_loss": 3.8279817732668526,
      "tokens_seen": 618201088
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00041044132397191575,
      "loss": 3.0149,
      "theoretical_loss": 3.827939653389164,
      "tokens_seen": 618266624
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.000410431293881645,
      "loss": 2.9741,
      "theoretical_loss": 3.8278975392258903,
      "tokens_seen": 618332160
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004104212637913741,
      "loss": 3.141,
      "theoretical_loss": 3.82785543077565,
      "tokens_seen": 618397696
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00041041123370110335,
      "loss": 2.8396,
      "theoretical_loss": 3.827813328037063,
      "tokens_seen": 618463232
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004104012036108325,
      "loss": 3.2096,
      "theoretical_loss": 3.8277712310087493,
      "tokens_seen": 618528768
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004103911735205617,
      "loss": 2.8117,
      "theoretical_loss": 3.82772913968933,
      "tokens_seen": 618594304
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004103811434302909,
      "loss": 2.9111,
      "theoretical_loss": 3.827687054077426,
      "tokens_seen": 618659840
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00041037111334002007,
      "loss": 2.82,
      "theoretical_loss": 3.827644974171659,
      "tokens_seen": 618725376
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00041036108324974925,
      "loss": 2.9438,
      "theoretical_loss": 3.8276028999706506,
      "tokens_seen": 618790912
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00041035105315947844,
      "loss": 3.073,
      "theoretical_loss": 3.8275608314730234,
      "tokens_seen": 618856448
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004103410230692076,
      "loss": 2.8658,
      "theoretical_loss": 3.8275187686774013,
      "tokens_seen": 618921984
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00041033099297893685,
      "loss": 2.7217,
      "theoretical_loss": 3.827476711582406,
      "tokens_seen": 618987520
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.000410320962888666,
      "loss": 3.1318,
      "theoretical_loss": 3.8274346601866633,
      "tokens_seen": 619053056
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004103109327983952,
      "loss": 3.0759,
      "theoretical_loss": 3.8273926144887973,
      "tokens_seen": 619118592
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004103009027081244,
      "loss": 3.1984,
      "theoretical_loss": 3.827350574487432,
      "tokens_seen": 619184128
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004102908726178536,
      "loss": 3.0183,
      "theoretical_loss": 3.8273085401811935,
      "tokens_seen": 619249664
    },
    {
      "epoch": 0.19,
      "objective/train/advantage_avg": 0.49511563777923584,
      "objective/train/docs_used": 360794,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.007056951522827,
      "objective/train/original_loss": 3.007056474685669,
      "objective/train/theoretical_loss": 3.8272665115687077,
      "objective/train/tokens_used": 639775200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24521978199481964,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507581233978271,
      "objective/train/weighted_lm_loss": 3.159679412841797,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0304546356201172,
      "theoretical_loss": 3.8272665115687077,
      "tokens_seen": 619315200
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00041028084252758276,
      "loss": 3.1158,
      "theoretical_loss": 3.8272665115687077,
      "tokens_seen": 619315200
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00041027081243731194,
      "loss": 2.9234,
      "theoretical_loss": 3.827224488648601,
      "tokens_seen": 619380736
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004102607823470411,
      "loss": 2.9772,
      "theoretical_loss": 3.8271824714194995,
      "tokens_seen": 619446272
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00041025075225677035,
      "loss": 2.7482,
      "theoretical_loss": 3.8271404598800314,
      "tokens_seen": 619511808
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004102407221664995,
      "loss": 3.0024,
      "theoretical_loss": 3.8270984540288238,
      "tokens_seen": 619577344
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004102306920762287,
      "loss": 3.0638,
      "theoretical_loss": 3.8270564538645058,
      "tokens_seen": 619642880
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00041022066198595784,
      "loss": 3.0174,
      "theoretical_loss": 3.8270144593857056,
      "tokens_seen": 619708416
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004102106318956871,
      "loss": 3.1448,
      "theoretical_loss": 3.826972470591053,
      "tokens_seen": 619773952
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00041020060180541626,
      "loss": 2.8871,
      "theoretical_loss": 3.8269304874791765,
      "tokens_seen": 619839488
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00041019057171514544,
      "loss": 3.0776,
      "theoretical_loss": 3.826888510048707,
      "tokens_seen": 619905024
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004101805416248746,
      "loss": 2.9602,
      "theoretical_loss": 3.8268465382982755,
      "tokens_seen": 619970560
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004101705115346038,
      "loss": 2.9772,
      "theoretical_loss": 3.8268045722265125,
      "tokens_seen": 620036096
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.000410160481444333,
      "loss": 3.2579,
      "theoretical_loss": 3.8267626118320504,
      "tokens_seen": 620101632
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004101504513540622,
      "loss": 2.945,
      "theoretical_loss": 3.82672065711352,
      "tokens_seen": 620167168
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00041014042126379135,
      "loss": 2.9826,
      "theoretical_loss": 3.826678708069555,
      "tokens_seen": 620232704
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004101303911735206,
      "loss": 2.6889,
      "theoretical_loss": 3.8266367646987876,
      "tokens_seen": 620298240
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00041012036108324976,
      "loss": 3.0328,
      "theoretical_loss": 3.8265948269998518,
      "tokens_seen": 620363776
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00041011033099297894,
      "loss": 2.9427,
      "theoretical_loss": 3.826552894971381,
      "tokens_seen": 620429312
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004101003009027081,
      "loss": 2.9984,
      "theoretical_loss": 3.826510968612011,
      "tokens_seen": 620494848
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004100902708124373,
      "loss": 3.0901,
      "theoretical_loss": 3.8264690479203747,
      "tokens_seen": 620560384
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004100802407221665,
      "loss": 3.1372,
      "theoretical_loss": 3.826427132895109,
      "tokens_seen": 620625920
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004100702106318957,
      "loss": 3.0007,
      "theoretical_loss": 3.8263852235348494,
      "tokens_seen": 620691456
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00041006018054162485,
      "loss": 2.9421,
      "theoretical_loss": 3.8263433198382324,
      "tokens_seen": 620756992
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004100501504513541,
      "loss": 2.755,
      "theoretical_loss": 3.826301421803894,
      "tokens_seen": 620822528
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004100401203610832,
      "loss": 3.1102,
      "theoretical_loss": 3.826259529430472,
      "tokens_seen": 620888064
    },
    {
      "epoch": 0.19,
      "objective/train/advantage_avg": 0.49440309405326843,
      "objective/train/docs_used": 361341,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.111715793609619,
      "objective/train/original_loss": 3.11171555519104,
      "objective/train/theoretical_loss": 3.8262176427166046,
      "objective/train/tokens_used": 641413600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24449469149112701,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506831407546997,
      "objective/train/weighted_lm_loss": 3.269418716430664,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0425469875335693,
      "theoretical_loss": 3.8262176427166046,
      "tokens_seen": 620953600
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00041003009027081245,
      "loss": 2.971,
      "theoretical_loss": 3.8262176427166046,
      "tokens_seen": 620953600
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00041002006018054163,
      "loss": 2.6765,
      "theoretical_loss": 3.826175761660929,
      "tokens_seen": 621019136
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004100100300902708,
      "loss": 3.0897,
      "theoretical_loss": 3.826133886262085,
      "tokens_seen": 621084672
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00041,
      "loss": 2.9315,
      "theoretical_loss": 3.826092016518711,
      "tokens_seen": 621150208
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040998996990972917,
      "loss": 2.9051,
      "theoretical_loss": 3.826050152429447,
      "tokens_seen": 621215744
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040997993981945835,
      "loss": 3.0736,
      "theoretical_loss": 3.8260082939929325,
      "tokens_seen": 621281280
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004099699097291876,
      "loss": 2.9934,
      "theoretical_loss": 3.825966441207809,
      "tokens_seen": 621346816
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004099598796389167,
      "loss": 2.948,
      "theoretical_loss": 3.825924594072717,
      "tokens_seen": 621412352
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040994984954864595,
      "loss": 2.8166,
      "theoretical_loss": 3.8258827525862973,
      "tokens_seen": 621477888
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040993981945837513,
      "loss": 3.1924,
      "theoretical_loss": 3.825840916747193,
      "tokens_seen": 621543424
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004099297893681043,
      "loss": 2.8433,
      "theoretical_loss": 3.8257990865540465,
      "tokens_seen": 621608960
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004099197592778335,
      "loss": 2.9888,
      "theoretical_loss": 3.8257572620055003,
      "tokens_seen": 621674496
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004099097291875627,
      "loss": 3.0746,
      "theoretical_loss": 3.8257154431001976,
      "tokens_seen": 621740032
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004098996990972919,
      "loss": 3.2196,
      "theoretical_loss": 3.825673629836783,
      "tokens_seen": 621805568
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004098896690070211,
      "loss": 3.0216,
      "theoretical_loss": 3.8256318222139,
      "tokens_seen": 621871104
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004098796389167503,
      "loss": 3.0213,
      "theoretical_loss": 3.8255900202301936,
      "tokens_seen": 621936640
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040986960882647945,
      "loss": 2.8886,
      "theoretical_loss": 3.8255482238843097,
      "tokens_seen": 622002176
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040985957873620864,
      "loss": 3.1866,
      "theoretical_loss": 3.825506433174893,
      "tokens_seen": 622067712
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004098495486459378,
      "loss": 3.0999,
      "theoretical_loss": 3.825464648100591,
      "tokens_seen": 622133248
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040983951855566705,
      "loss": 3.0143,
      "theoretical_loss": 3.8254228686600493,
      "tokens_seen": 622198784
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004098294884653962,
      "loss": 3.047,
      "theoretical_loss": 3.825381094851915,
      "tokens_seen": 622264320
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004098194583751254,
      "loss": 2.8767,
      "theoretical_loss": 3.8253393266748366,
      "tokens_seen": 622329856
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004098094282848546,
      "loss": 2.8967,
      "theoretical_loss": 3.825297564127461,
      "tokens_seen": 622395392
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004097993981945838,
      "loss": 2.8285,
      "theoretical_loss": 3.825255807208438,
      "tokens_seen": 622460928
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040978936810431296,
      "loss": 3.0944,
      "theoretical_loss": 3.825214055916416,
      "tokens_seen": 622526464
    },
    {
      "epoch": 0.19,
      "objective/train/advantage_avg": 0.49671006202697754,
      "objective/train/docs_used": 362042,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.9464352130889893,
      "objective/train/original_loss": 2.9464352130889893,
      "objective/train/theoretical_loss": 3.8251723102500437,
      "objective/train/tokens_used": 643052000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24677026271820068,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0509254932403564,
      "objective/train/weighted_lm_loss": 3.0963797569274902,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0304546356201172,
      "theoretical_loss": 3.8251723102500437,
      "tokens_seen": 622592000
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040977933801404214,
      "loss": 2.7691,
      "theoretical_loss": 3.8251723102500437,
      "tokens_seen": 622592000
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004097693079237713,
      "loss": 3.1065,
      "theoretical_loss": 3.8251305702079725,
      "tokens_seen": 622657536
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040975927783350055,
      "loss": 3.0354,
      "theoretical_loss": 3.8250888357888515,
      "tokens_seen": 622723072
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004097492477432297,
      "loss": 3.0829,
      "theoretical_loss": 3.8250471069913328,
      "tokens_seen": 622788608
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004097392176529589,
      "loss": 2.7627,
      "theoretical_loss": 3.8250053838140663,
      "tokens_seen": 622854144
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040972918756268804,
      "loss": 2.9241,
      "theoretical_loss": 3.824963666255705,
      "tokens_seen": 622919680
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004097191574724173,
      "loss": 2.81,
      "theoretical_loss": 3.8249219543149007,
      "tokens_seen": 622985216
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040970912738214646,
      "loss": 3.0531,
      "theoretical_loss": 3.824880247990306,
      "tokens_seen": 623050752
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040969909729187564,
      "loss": 2.9973,
      "theoretical_loss": 3.8248385472805744,
      "tokens_seen": 623116288
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004096890672016048,
      "loss": 3.1137,
      "theoretical_loss": 3.8247968521843596,
      "tokens_seen": 623181824
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.000409679037111334,
      "loss": 2.8728,
      "theoretical_loss": 3.824755162700315,
      "tokens_seen": 623247360
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004096690070210632,
      "loss": 2.935,
      "theoretical_loss": 3.8247134788270962,
      "tokens_seen": 623312896
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004096589769307924,
      "loss": 2.8482,
      "theoretical_loss": 3.8246718005633573,
      "tokens_seen": 623378432
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040964894684052155,
      "loss": 2.911,
      "theoretical_loss": 3.8246301279077546,
      "tokens_seen": 623443968
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004096389167502508,
      "loss": 3.095,
      "theoretical_loss": 3.8245884608589438,
      "tokens_seen": 623509504
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040962888665997996,
      "loss": 2.8863,
      "theoretical_loss": 3.824546799415581,
      "tokens_seen": 623575040
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040961885656970914,
      "loss": 2.9437,
      "theoretical_loss": 3.8245051435763235,
      "tokens_seen": 623640576
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004096088264794383,
      "loss": 3.2054,
      "theoretical_loss": 3.824463493339829,
      "tokens_seen": 623706112
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004095987963891675,
      "loss": 2.9714,
      "theoretical_loss": 3.8244218487047545,
      "tokens_seen": 623771648
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004095887662988967,
      "loss": 3.1625,
      "theoretical_loss": 3.8243802096697586,
      "tokens_seen": 623837184
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004095787362086259,
      "loss": 3.1701,
      "theoretical_loss": 3.8243385762335,
      "tokens_seen": 623902720
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040956870611835505,
      "loss": 2.8891,
      "theoretical_loss": 3.8242969483946387,
      "tokens_seen": 623968256
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004095586760280843,
      "loss": 2.9852,
      "theoretical_loss": 3.8242553261518335,
      "tokens_seen": 624033792
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004095486459378134,
      "loss": 2.9742,
      "theoretical_loss": 3.8242137095037445,
      "tokens_seen": 624099328
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040953861584754265,
      "loss": 3.0141,
      "theoretical_loss": 3.824172098449033,
      "tokens_seen": 624164864
    },
    {
      "epoch": 0.19,
      "objective/train/advantage_avg": 0.49275341629981995,
      "objective/train/docs_used": 363016,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.0655364990234375,
      "objective/train/original_loss": 3.0655369758605957,
      "objective/train/theoretical_loss": 3.824130492986359,
      "objective/train/tokens_used": 644690400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24289196729660034,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050510048866272,
      "objective/train/weighted_lm_loss": 3.2203409671783447,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0304546356201172,
      "theoretical_loss": 3.824130492986359,
      "tokens_seen": 624230400
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040952858575727183,
      "loss": 3.1657,
      "theoretical_loss": 3.824130492986359,
      "tokens_seen": 624230400
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.000409518555667001,
      "loss": 3.0501,
      "theoretical_loss": 3.824088893114385,
      "tokens_seen": 624295936
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004095085255767302,
      "loss": 3.0588,
      "theoretical_loss": 3.824047298831773,
      "tokens_seen": 624361472
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040949849548645937,
      "loss": 2.8385,
      "theoretical_loss": 3.8240057101371843,
      "tokens_seen": 624427008
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040948846539618855,
      "loss": 3.097,
      "theoretical_loss": 3.8239641270292832,
      "tokens_seen": 624492544
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004094784353059178,
      "loss": 3.1058,
      "theoretical_loss": 3.823922549506732,
      "tokens_seen": 624558080
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004094684052156469,
      "loss": 3.0423,
      "theoretical_loss": 3.8238809775681952,
      "tokens_seen": 624623616
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040945837512537615,
      "loss": 2.929,
      "theoretical_loss": 3.823839411212337,
      "tokens_seen": 624689152
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040944834503510533,
      "loss": 3.1735,
      "theoretical_loss": 3.8237978504378223,
      "tokens_seen": 624754688
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004094383149448345,
      "loss": 2.821,
      "theoretical_loss": 3.8237562952433155,
      "tokens_seen": 624820224
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004094282848545637,
      "loss": 2.8029,
      "theoretical_loss": 3.823714745627483,
      "tokens_seen": 624885760
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004094182547642929,
      "loss": 3.1124,
      "theoretical_loss": 3.8236732015889903,
      "tokens_seen": 624951296
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040940822467402206,
      "loss": 3.0276,
      "theoretical_loss": 3.823631663126505,
      "tokens_seen": 625016832
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004093981945837513,
      "loss": 2.9483,
      "theoretical_loss": 3.823590130238693,
      "tokens_seen": 625082368
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004093881644934804,
      "loss": 2.835,
      "theoretical_loss": 3.823548602924223,
      "tokens_seen": 625147904
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040937813440320965,
      "loss": 3.3181,
      "theoretical_loss": 3.8235070811817615,
      "tokens_seen": 625213440
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004093681043129388,
      "loss": 3.0448,
      "theoretical_loss": 3.823465565009978,
      "tokens_seen": 625278976
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.000409358074222668,
      "loss": 2.8252,
      "theoretical_loss": 3.8234240544075417,
      "tokens_seen": 625344512
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004093480441323972,
      "loss": 2.9188,
      "theoretical_loss": 3.8233825493731204,
      "tokens_seen": 625410048
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004093380140421264,
      "loss": 3.0201,
      "theoretical_loss": 3.823341049905385,
      "tokens_seen": 625475584
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040932798395185556,
      "loss": 2.9653,
      "theoretical_loss": 3.823299556003006,
      "tokens_seen": 625541120
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004093179538615848,
      "loss": 3.13,
      "theoretical_loss": 3.823258067664653,
      "tokens_seen": 625606656
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004093079237713139,
      "loss": 2.8461,
      "theoretical_loss": 3.823216584888998,
      "tokens_seen": 625672192
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040929789368104316,
      "loss": 3.1238,
      "theoretical_loss": 3.8231751076747127,
      "tokens_seen": 625737728
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004092878635907723,
      "loss": 2.9948,
      "theoretical_loss": 3.8231336360204686,
      "tokens_seen": 625803264
    },
    {
      "epoch": 0.19,
      "objective/train/advantage_avg": 0.49530017375946045,
      "objective/train/docs_used": 363672,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.7158381938934326,
      "objective/train/original_loss": 2.715837240219116,
      "objective/train/theoretical_loss": 3.823092169924938,
      "objective/train/tokens_used": 646328800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24537445604801178,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507774353027344,
      "objective/train/weighted_lm_loss": 2.853473663330078,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0467102527618408,
      "theoretical_loss": 3.823092169924938,
      "tokens_seen": 625868800
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004092778335005015,
      "loss": 2.8959,
      "theoretical_loss": 3.823092169924938,
      "tokens_seen": 625868800
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004092678034102307,
      "loss": 2.9641,
      "theoretical_loss": 3.8230507093867954,
      "tokens_seen": 625934336
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004092577733199599,
      "loss": 2.9304,
      "theoretical_loss": 3.8230092544047123,
      "tokens_seen": 625999872
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040924774322968906,
      "loss": 3.0644,
      "theoretical_loss": 3.822967804977364,
      "tokens_seen": 626065408
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040923771313941824,
      "loss": 3.0352,
      "theoretical_loss": 3.8229263611034243,
      "tokens_seen": 626130944
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004092276830491474,
      "loss": 2.9103,
      "theoretical_loss": 3.822884922781568,
      "tokens_seen": 626196480
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040921765295887666,
      "loss": 2.9107,
      "theoretical_loss": 3.8228434900104706,
      "tokens_seen": 626262016
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004092076228686058,
      "loss": 2.9826,
      "theoretical_loss": 3.8228020627888073,
      "tokens_seen": 626327552
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.000409197592778335,
      "loss": 2.7634,
      "theoretical_loss": 3.822760641115255,
      "tokens_seen": 626393088
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040918756268806415,
      "loss": 2.9769,
      "theoretical_loss": 3.8227192249884894,
      "tokens_seen": 626458624
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004091775325977934,
      "loss": 3.2949,
      "theoretical_loss": 3.8226778144071885,
      "tokens_seen": 626524160
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040916750250752257,
      "loss": 3.1077,
      "theoretical_loss": 3.8226364093700296,
      "tokens_seen": 626589696
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040915747241725175,
      "loss": 3.2067,
      "theoretical_loss": 3.82259500987569,
      "tokens_seen": 626655232
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.000409147442326981,
      "loss": 2.6774,
      "theoretical_loss": 3.822553615922849,
      "tokens_seen": 626720768
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040913741223671016,
      "loss": 2.8655,
      "theoretical_loss": 3.8225122275101846,
      "tokens_seen": 626786304
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040912738214643934,
      "loss": 3.0833,
      "theoretical_loss": 3.8224708446363778,
      "tokens_seen": 626851840
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004091173520561685,
      "loss": 2.9303,
      "theoretical_loss": 3.8224294673001062,
      "tokens_seen": 626917376
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004091073219658977,
      "loss": 3.162,
      "theoretical_loss": 3.8223880955000515,
      "tokens_seen": 626982912
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004090972918756269,
      "loss": 2.8827,
      "theoretical_loss": 3.8223467292348943,
      "tokens_seen": 627048448
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004090872617853561,
      "loss": 3.0232,
      "theoretical_loss": 3.822305368503315,
      "tokens_seen": 627113984
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040907723169508525,
      "loss": 3.1041,
      "theoretical_loss": 3.822264013303996,
      "tokens_seen": 627179520
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004090672016048145,
      "loss": 3.3352,
      "theoretical_loss": 3.822222663635619,
      "tokens_seen": 627245056
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004090571715145436,
      "loss": 3.1181,
      "theoretical_loss": 3.822181319496867,
      "tokens_seen": 627310592
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040904714142427285,
      "loss": 3.0825,
      "theoretical_loss": 3.822139980886422,
      "tokens_seen": 627376128
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040903711133400203,
      "loss": 2.9488,
      "theoretical_loss": 3.8220986478029677,
      "tokens_seen": 627441664
    },
    {
      "epoch": 0.19,
      "objective/train/advantage_avg": 0.492232084274292,
      "objective/train/docs_used": 364826,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.5110385417938232,
      "objective/train/original_loss": 2.5110385417938232,
      "objective/train/theoretical_loss": 3.8220573202451886,
      "objective/train/tokens_used": 647967200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24244053661823273,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0504554510116577,
      "objective/train/weighted_lm_loss": 2.637688159942627,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0408108234405518,
      "theoretical_loss": 3.8220573202451886,
      "tokens_seen": 627507200
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004090270812437312,
      "loss": 2.8294,
      "theoretical_loss": 3.8220573202451886,
      "tokens_seen": 627507200
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004090170511534604,
      "loss": 3.1261,
      "theoretical_loss": 3.822015998211768,
      "tokens_seen": 627572736
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040900702106318957,
      "loss": 2.9388,
      "theoretical_loss": 3.821974681701392,
      "tokens_seen": 627638272
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040899699097291875,
      "loss": 3.1221,
      "theoretical_loss": 3.8219333707127445,
      "tokens_seen": 627703808
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.000408986960882648,
      "loss": 3.0322,
      "theoretical_loss": 3.8218920652445125,
      "tokens_seen": 627769344
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004089769307923771,
      "loss": 3.0576,
      "theoretical_loss": 3.8218507652953813,
      "tokens_seen": 627834880
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040896690070210635,
      "loss": 3.2305,
      "theoretical_loss": 3.8218094708640367,
      "tokens_seen": 627900416
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040895687061183553,
      "loss": 3.1117,
      "theoretical_loss": 3.8217681819491673,
      "tokens_seen": 627965952
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004089468405215647,
      "loss": 3.1197,
      "theoretical_loss": 3.8217268985494597,
      "tokens_seen": 628031488
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004089368104312939,
      "loss": 3.1226,
      "theoretical_loss": 3.8216856206636014,
      "tokens_seen": 628097024
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004089267803410231,
      "loss": 3.0344,
      "theoretical_loss": 3.821644348290282,
      "tokens_seen": 628162560
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040891675025075226,
      "loss": 3.0813,
      "theoretical_loss": 3.821603081428189,
      "tokens_seen": 628228096
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004089067201604815,
      "loss": 3.0493,
      "theoretical_loss": 3.8215618200760124,
      "tokens_seen": 628293632
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004088966900702106,
      "loss": 3.0773,
      "theoretical_loss": 3.8215205642324417,
      "tokens_seen": 628359168
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040888665997993985,
      "loss": 2.9093,
      "theoretical_loss": 3.8214793138961674,
      "tokens_seen": 628424704
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.000408876629889669,
      "loss": 3.0997,
      "theoretical_loss": 3.82143806906588,
      "tokens_seen": 628490240
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004088665997993982,
      "loss": 2.893,
      "theoretical_loss": 3.82139682974027,
      "tokens_seen": 628555776
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004088565697091274,
      "loss": 3.193,
      "theoretical_loss": 3.821355595918029,
      "tokens_seen": 628621312
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004088465396188566,
      "loss": 2.9186,
      "theoretical_loss": 3.8213143675978496,
      "tokens_seen": 628686848
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040883650952858576,
      "loss": 3.1285,
      "theoretical_loss": 3.821273144778424,
      "tokens_seen": 628752384
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.000408826479438315,
      "loss": 2.9994,
      "theoretical_loss": 3.8212319274584448,
      "tokens_seen": 628817920
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004088164493480441,
      "loss": 3.1583,
      "theoretical_loss": 3.8211907156366047,
      "tokens_seen": 628883456
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040880641925777336,
      "loss": 3.0408,
      "theoretical_loss": 3.821149509311599,
      "tokens_seen": 628948992
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004087963891675025,
      "loss": 3.0779,
      "theoretical_loss": 3.8211083084821205,
      "tokens_seen": 629014528
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004087863590772317,
      "loss": 3.2651,
      "theoretical_loss": 3.821067113146864,
      "tokens_seen": 629080064
    },
    {
      "epoch": 0.19,
      "objective/train/advantage_avg": 0.4947104752063751,
      "objective/train/docs_used": 365375,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.0285804271698,
      "objective/train/original_loss": 3.028580665588379,
      "objective/train/theoretical_loss": 3.8210259233045254,
      "objective/train/tokens_used": 649605600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2447933405637741,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507153272628784,
      "objective/train/weighted_lm_loss": 3.182091236114502,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0363596677780151,
      "theoretical_loss": 3.8210259233045254,
      "tokens_seen": 629145600
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004087763289869609,
      "loss": 2.911,
      "theoretical_loss": 3.8210259233045254,
      "tokens_seen": 629145600
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004087662988966901,
      "loss": 3.055,
      "theoretical_loss": 3.8209847389537996,
      "tokens_seen": 629211136
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040875626880641926,
      "loss": 2.9683,
      "theoretical_loss": 3.820943560093382,
      "tokens_seen": 629276672
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040874623871614844,
      "loss": 3.2464,
      "theoretical_loss": 3.8209023867219702,
      "tokens_seen": 629342208
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004087362086258776,
      "loss": 3.2791,
      "theoretical_loss": 3.8208612188382607,
      "tokens_seen": 629407744
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040872617853560686,
      "loss": 3.1835,
      "theoretical_loss": 3.8208200564409505,
      "tokens_seen": 629473280
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.000408716148445336,
      "loss": 2.9428,
      "theoretical_loss": 3.8207788995287375,
      "tokens_seen": 629538816
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004087061183550652,
      "loss": 2.918,
      "theoretical_loss": 3.8207377481003197,
      "tokens_seen": 629604352
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040869608826479435,
      "loss": 3.1981,
      "theoretical_loss": 3.8206966021543955,
      "tokens_seen": 629669888
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004086860581745236,
      "loss": 2.9027,
      "theoretical_loss": 3.8206554616896646,
      "tokens_seen": 629735424
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040867602808425277,
      "loss": 3.0015,
      "theoretical_loss": 3.8206143267048267,
      "tokens_seen": 629800960
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040866599799398195,
      "loss": 3.0634,
      "theoretical_loss": 3.820573197198581,
      "tokens_seen": 629866496
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040865596790371113,
      "loss": 3.1038,
      "theoretical_loss": 3.8205320731696277,
      "tokens_seen": 629932032
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040864593781344036,
      "loss": 2.9408,
      "theoretical_loss": 3.820490954616669,
      "tokens_seen": 629997568
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004086359077231695,
      "loss": 2.8465,
      "theoretical_loss": 3.820449841538405,
      "tokens_seen": 630063104
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004086258776328987,
      "loss": 3.3048,
      "theoretical_loss": 3.8204087339335384,
      "tokens_seen": 630128640
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040861584754262785,
      "loss": 2.8984,
      "theoretical_loss": 3.8203676318007704,
      "tokens_seen": 630194176
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004086058174523571,
      "loss": 3.1948,
      "theoretical_loss": 3.8203265351388045,
      "tokens_seen": 630259712
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040859578736208627,
      "loss": 2.7919,
      "theoretical_loss": 3.820285443946343,
      "tokens_seen": 630325248
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040858575727181545,
      "loss": 2.9034,
      "theoretical_loss": 3.820244358222089,
      "tokens_seen": 630390784
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040857572718154463,
      "loss": 3.1614,
      "theoretical_loss": 3.8202032779647483,
      "tokens_seen": 630456320
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004085656970912738,
      "loss": 2.8932,
      "theoretical_loss": 3.820162203173024,
      "tokens_seen": 630521856
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.000408555667001003,
      "loss": 2.9798,
      "theoretical_loss": 3.8201211338456202,
      "tokens_seen": 630587392
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040854563691073223,
      "loss": 2.9327,
      "theoretical_loss": 3.820080069981244,
      "tokens_seen": 630652928
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040853560682046136,
      "loss": 3.1197,
      "theoretical_loss": 3.8200390115786003,
      "tokens_seen": 630718464
    },
    {
      "epoch": 0.19,
      "objective/train/advantage_avg": 0.4957661032676697,
      "objective/train/docs_used": 366612,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.354647397994995,
      "objective/train/original_loss": 2.354647159576416,
      "objective/train/theoretical_loss": 3.8199979586363946,
      "objective/train/tokens_used": 651244000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24582457542419434,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0508263111114502,
      "objective/train/weighted_lm_loss": 2.474064826965332,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0425469875335693,
      "theoretical_loss": 3.8199979586363946,
      "tokens_seen": 630784000
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004085255767301906,
      "loss": 2.8543,
      "theoretical_loss": 3.8199979586363946,
      "tokens_seen": 630784000
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004085155466399197,
      "loss": 3.2024,
      "theoretical_loss": 3.8199569111533345,
      "tokens_seen": 630849536
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040850551654964895,
      "loss": 3.0524,
      "theoretical_loss": 3.8199158691281263,
      "tokens_seen": 630915072
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040849548645937814,
      "loss": 2.8654,
      "theoretical_loss": 3.819874832559478,
      "tokens_seen": 630980608
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004084854563691073,
      "loss": 3.1622,
      "theoretical_loss": 3.819833801446097,
      "tokens_seen": 631046144
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004084754262788365,
      "loss": 3.216,
      "theoretical_loss": 3.8197927757866923,
      "tokens_seen": 631111680
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040846539618856573,
      "loss": 2.838,
      "theoretical_loss": 3.819751755579972,
      "tokens_seen": 631177216
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040845536609829486,
      "loss": 3.0869,
      "theoretical_loss": 3.819710740824646,
      "tokens_seen": 631242752
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004084453360080241,
      "loss": 3.0296,
      "theoretical_loss": 3.819669731519424,
      "tokens_seen": 631308288
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004084353059177532,
      "loss": 2.9512,
      "theoretical_loss": 3.819628727663015,
      "tokens_seen": 631373824
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040842527582748246,
      "loss": 3.0527,
      "theoretical_loss": 3.819587729254131,
      "tokens_seen": 631439360
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040841524573721164,
      "loss": 3.028,
      "theoretical_loss": 3.819546736291483,
      "tokens_seen": 631504896
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004084052156469408,
      "loss": 3.0688,
      "theoretical_loss": 3.8195057487737802,
      "tokens_seen": 631570432
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040839518555667005,
      "loss": 3.0566,
      "theoretical_loss": 3.8194647666997374,
      "tokens_seen": 631635968
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004083851554663992,
      "loss": 3.2341,
      "theoretical_loss": 3.8194237900680648,
      "tokens_seen": 631701504
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004083751253761284,
      "loss": 2.9805,
      "theoretical_loss": 3.819382818877477,
      "tokens_seen": 631767040
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004083650952858576,
      "loss": 2.9937,
      "theoretical_loss": 3.819341853126685,
      "tokens_seen": 631832576
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004083550651955868,
      "loss": 2.9938,
      "theoretical_loss": 3.819300892814404,
      "tokens_seen": 631898112
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040834503510531596,
      "loss": 3.026,
      "theoretical_loss": 3.8192599379393477,
      "tokens_seen": 631963648
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004083350050150452,
      "loss": 3.0801,
      "theoretical_loss": 3.819218988500231,
      "tokens_seen": 632029184
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004083249749247743,
      "loss": 2.9466,
      "theoretical_loss": 3.8191780444957675,
      "tokens_seen": 632094720
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040831494483450356,
      "loss": 3.0511,
      "theoretical_loss": 3.819137105924674,
      "tokens_seen": 632160256
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004083049147442327,
      "loss": 3.1435,
      "theoretical_loss": 3.819096172785666,
      "tokens_seen": 632225792
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004082948846539619,
      "loss": 3.0067,
      "theoretical_loss": 3.8190552450774584,
      "tokens_seen": 632291328
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004082848545636911,
      "loss": 2.9919,
      "theoretical_loss": 3.81901432279877,
      "tokens_seen": 632356864
    },
    {
      "epoch": 0.19,
      "objective/train/advantage_avg": 0.4962116777896881,
      "objective/train/docs_used": 367323,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.1482865810394287,
      "objective/train/original_loss": 3.1482863426208496,
      "objective/train/theoretical_loss": 3.8189734059483165,
      "objective/train/tokens_used": 652882400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24628770351409912,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0508732795715332,
      "objective/train/weighted_lm_loss": 3.3084802627563477,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0396549701690674,
      "theoretical_loss": 3.8189734059483165,
      "tokens_seen": 632422400
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004082748244734203,
      "loss": 3.2113,
      "theoretical_loss": 3.8189734059483165,
      "tokens_seen": 632422400
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040826479438314946,
      "loss": 2.7556,
      "theoretical_loss": 3.8189324945248164,
      "tokens_seen": 632487936
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040825476429287864,
      "loss": 2.8185,
      "theoretical_loss": 3.818891588526987,
      "tokens_seen": 632553472
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004082447342026078,
      "loss": 3.0757,
      "theoretical_loss": 3.818850687953547,
      "tokens_seen": 632619008
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040823470411233706,
      "loss": 2.8998,
      "theoretical_loss": 3.8188097928032145,
      "tokens_seen": 632684544
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004082246740220662,
      "loss": 3.1264,
      "theoretical_loss": 3.81876890307471,
      "tokens_seen": 632750080
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004082146439317954,
      "loss": 3.1038,
      "theoretical_loss": 3.8187280187667527,
      "tokens_seen": 632815616
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040820461384152455,
      "loss": 3.111,
      "theoretical_loss": 3.8186871398780626,
      "tokens_seen": 632881152
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004081945837512538,
      "loss": 2.8071,
      "theoretical_loss": 3.8186462664073613,
      "tokens_seen": 632946688
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040818455366098297,
      "loss": 2.958,
      "theoretical_loss": 3.818605398353368,
      "tokens_seen": 633012224
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040817452357071215,
      "loss": 2.7234,
      "theoretical_loss": 3.8185645357148057,
      "tokens_seen": 633077760
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040816449348044133,
      "loss": 2.9931,
      "theoretical_loss": 3.8185236784903958,
      "tokens_seen": 633143296
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040815446339017056,
      "loss": 3.1107,
      "theoretical_loss": 3.8184828266788604,
      "tokens_seen": 633208832
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004081444332998997,
      "loss": 2.7649,
      "theoretical_loss": 3.818441980278923,
      "tokens_seen": 633274368
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004081344032096289,
      "loss": 2.6766,
      "theoretical_loss": 3.818401139289306,
      "tokens_seen": 633339904
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040812437311935805,
      "loss": 2.8995,
      "theoretical_loss": 3.818360303708733,
      "tokens_seen": 633405440
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004081143430290873,
      "loss": 2.9138,
      "theoretical_loss": 3.8183194735359294,
      "tokens_seen": 633470976
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040810431293881647,
      "loss": 3.221,
      "theoretical_loss": 3.8182786487696183,
      "tokens_seen": 633536512
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040809428284854565,
      "loss": 3.2016,
      "theoretical_loss": 3.818237829408525,
      "tokens_seen": 633602048
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040808425275827483,
      "loss": 2.8493,
      "theoretical_loss": 3.818197015451375,
      "tokens_seen": 633667584
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.000408074222668004,
      "loss": 2.931,
      "theoretical_loss": 3.8181562068968944,
      "tokens_seen": 633733120
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004080641925777332,
      "loss": 2.9177,
      "theoretical_loss": 3.8181154037438088,
      "tokens_seen": 633798656
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040805416248746243,
      "loss": 3.0448,
      "theoretical_loss": 3.8180746059908457,
      "tokens_seen": 633864192
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040804413239719156,
      "loss": 2.9269,
      "theoretical_loss": 3.818033813636731,
      "tokens_seen": 633929728
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004080341023069208,
      "loss": 2.9388,
      "theoretical_loss": 3.8179930266801936,
      "tokens_seen": 633995264
    },
    {
      "epoch": 0.19,
      "objective/train/advantage_avg": 0.4939691722393036,
      "objective/train/docs_used": 367883,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.0024900436401367,
      "objective/train/original_loss": 3.0024900436401367,
      "objective/train/theoretical_loss": 3.817952245119961,
      "objective/train/tokens_used": 654520800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24407844245433807,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506377220153809,
      "objective/train/weighted_lm_loss": 3.1544413566589355,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 3.817952245119961,
      "tokens_seen": 634060800
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004080240722166499,
      "loss": 3.0987,
      "theoretical_loss": 3.817952245119961,
      "tokens_seen": 634060800
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040801404212637915,
      "loss": 2.9136,
      "theoretical_loss": 3.8179114689547604,
      "tokens_seen": 634126336
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040800401203610834,
      "loss": 3.1235,
      "theoretical_loss": 3.8178706981833224,
      "tokens_seen": 634191872
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004079939819458375,
      "loss": 2.941,
      "theoretical_loss": 3.8178299328043757,
      "tokens_seen": 634257408
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004079839518555667,
      "loss": 2.9743,
      "theoretical_loss": 3.817789172816649,
      "tokens_seen": 634322944
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040797392176529593,
      "loss": 2.9791,
      "theoretical_loss": 3.8177484182188737,
      "tokens_seen": 634388480
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040796389167502506,
      "loss": 2.7063,
      "theoretical_loss": 3.8177076690097795,
      "tokens_seen": 634454016
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004079538615847543,
      "loss": 3.0466,
      "theoretical_loss": 3.817666925188098,
      "tokens_seen": 634519552
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004079438314944834,
      "loss": 2.9151,
      "theoretical_loss": 3.8176261867525607,
      "tokens_seen": 634585088
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040793380140421266,
      "loss": 2.8291,
      "theoretical_loss": 3.8175854537018985,
      "tokens_seen": 634650624
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040792377131394184,
      "loss": 3.1059,
      "theoretical_loss": 3.817544726034844,
      "tokens_seen": 634716160
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.000407913741223671,
      "loss": 3.001,
      "theoretical_loss": 3.8175040037501304,
      "tokens_seen": 634781696
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004079037111334002,
      "loss": 2.9786,
      "theoretical_loss": 3.8174632868464915,
      "tokens_seen": 634847232
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004078936810431294,
      "loss": 2.9977,
      "theoretical_loss": 3.817422575322659,
      "tokens_seen": 634912768
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040788365095285856,
      "loss": 2.9808,
      "theoretical_loss": 3.817381869177368,
      "tokens_seen": 634978304
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004078736208625878,
      "loss": 3.0868,
      "theoretical_loss": 3.817341168409353,
      "tokens_seen": 635043840
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004078635907723169,
      "loss": 3.2782,
      "theoretical_loss": 3.8173004730173483,
      "tokens_seen": 635109376
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040785356068204616,
      "loss": 3.1407,
      "theoretical_loss": 3.8172597830000896,
      "tokens_seen": 635174912
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004078435305917753,
      "loss": 3.1211,
      "theoretical_loss": 3.817219098356312,
      "tokens_seen": 635240448
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004078335005015045,
      "loss": 3.0272,
      "theoretical_loss": 3.8171784190847524,
      "tokens_seen": 635305984
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004078234704112337,
      "loss": 3.0236,
      "theoretical_loss": 3.817137745184147,
      "tokens_seen": 635371520
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004078134403209629,
      "loss": 2.99,
      "theoretical_loss": 3.8170970766532326,
      "tokens_seen": 635437056
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040780341023069207,
      "loss": 2.936,
      "theoretical_loss": 3.8170564134907474,
      "tokens_seen": 635502592
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004077933801404213,
      "loss": 2.7339,
      "theoretical_loss": 3.8170157556954285,
      "tokens_seen": 635568128
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040778335005015043,
      "loss": 3.1537,
      "theoretical_loss": 3.8169751032660137,
      "tokens_seen": 635633664
    },
    {
      "epoch": 0.19,
      "objective/train/advantage_avg": 0.4920635223388672,
      "objective/train/docs_used": 369288,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.0266594886779785,
      "objective/train/original_loss": 3.0266597270965576,
      "objective/train/theoretical_loss": 3.816934456201243,
      "objective/train/tokens_used": 656159200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24223260581493378,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050437569618225,
      "objective/train/weighted_lm_loss": 3.179224967956543,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 3.816934456201243,
      "tokens_seen": 635699200
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040777331995987966,
      "loss": 3.0592,
      "theoretical_loss": 3.816934456201243,
      "tokens_seen": 635699200
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004077632898696088,
      "loss": 2.7893,
      "theoretical_loss": 3.8168938144998545,
      "tokens_seen": 635764736
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.000407753259779338,
      "loss": 2.9319,
      "theoretical_loss": 3.8168531781605877,
      "tokens_seen": 635830272
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004077432296890672,
      "loss": 2.9742,
      "theoretical_loss": 3.8168125471821837,
      "tokens_seen": 635895808
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004077331995987964,
      "loss": 3.1295,
      "theoretical_loss": 3.8167719215633817,
      "tokens_seen": 635961344
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040772316950852557,
      "loss": 2.6777,
      "theoretical_loss": 3.816731301302923,
      "tokens_seen": 636026880
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040771313941825475,
      "loss": 2.9079,
      "theoretical_loss": 3.816690686399549,
      "tokens_seen": 636092416
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040770310932798393,
      "loss": 2.9297,
      "theoretical_loss": 3.8166500768520004,
      "tokens_seen": 636157952
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040769307923771317,
      "loss": 2.961,
      "theoretical_loss": 3.816609472659021,
      "tokens_seen": 636223488
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004076830491474423,
      "loss": 2.9706,
      "theoretical_loss": 3.8165688738193517,
      "tokens_seen": 636289024
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040767301905717153,
      "loss": 3.1058,
      "theoretical_loss": 3.8165282803317364,
      "tokens_seen": 636354560
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004076629889669007,
      "loss": 3.3057,
      "theoretical_loss": 3.8164876921949182,
      "tokens_seen": 636420096
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004076529588766299,
      "loss": 3.0244,
      "theoretical_loss": 3.816447109407641,
      "tokens_seen": 636485632
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004076429287863591,
      "loss": 2.8081,
      "theoretical_loss": 3.8164065319686484,
      "tokens_seen": 636551168
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040763289869608825,
      "loss": 3.0751,
      "theoretical_loss": 3.816365959876686,
      "tokens_seen": 636616704
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004076228686058175,
      "loss": 3.0045,
      "theoretical_loss": 3.8163253931304983,
      "tokens_seen": 636682240
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040761283851554667,
      "loss": 3.0623,
      "theoretical_loss": 3.816284831728831,
      "tokens_seen": 636747776
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040760280842527585,
      "loss": 2.9913,
      "theoretical_loss": 3.81624427567043,
      "tokens_seen": 636813312
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040759277833500503,
      "loss": 3.0024,
      "theoretical_loss": 3.816203724954041,
      "tokens_seen": 636878848
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004075827482447342,
      "loss": 2.8,
      "theoretical_loss": 3.8161631795784117,
      "tokens_seen": 636944384
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004075727181544634,
      "loss": 2.7457,
      "theoretical_loss": 3.816122639542289,
      "tokens_seen": 637009920
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040756268806419263,
      "loss": 2.9282,
      "theoretical_loss": 3.81608210484442,
      "tokens_seen": 637075456
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040755265797392176,
      "loss": 3.0657,
      "theoretical_loss": 3.8160415754835535,
      "tokens_seen": 637140992
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.000407542627883651,
      "loss": 2.856,
      "theoretical_loss": 3.8160010514584375,
      "tokens_seen": 637206528
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004075325977933801,
      "loss": 2.8694,
      "theoretical_loss": 3.815960532767821,
      "tokens_seen": 637272064
    },
    {
      "epoch": 0.19,
      "objective/train/advantage_avg": 0.49385586380958557,
      "objective/train/docs_used": 369983,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.322895050048828,
      "objective/train/original_loss": 3.322894811630249,
      "objective/train/theoretical_loss": 3.815920019410453,
      "objective/train/tokens_used": 657797600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24395935237407684,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506256818771362,
      "objective/train/weighted_lm_loss": 3.4910480976104736,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0432155132293701,
      "theoretical_loss": 3.815920019410453,
      "tokens_seen": 637337600
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040752256770310935,
      "loss": 3.3538,
      "theoretical_loss": 3.815920019410453,
      "tokens_seen": 637337600
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040751253761283854,
      "loss": 2.8461,
      "theoretical_loss": 3.8158795113850834,
      "tokens_seen": 637403136
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004075025075225677,
      "loss": 3.064,
      "theoretical_loss": 3.815839008690463,
      "tokens_seen": 637468672
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004074924774322969,
      "loss": 2.8104,
      "theoretical_loss": 3.815798511325341,
      "tokens_seen": 637534208
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040748244734202613,
      "loss": 3.098,
      "theoretical_loss": 3.8157580192884692,
      "tokens_seen": 637599744
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040747241725175526,
      "loss": 2.8878,
      "theoretical_loss": 3.8157175325785997,
      "tokens_seen": 637665280
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004074623871614845,
      "loss": 3.0175,
      "theoretical_loss": 3.8156770511944833,
      "tokens_seen": 637730816
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004074523570712136,
      "loss": 2.9095,
      "theoretical_loss": 3.8156365751348718,
      "tokens_seen": 637796352
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040744232698094286,
      "loss": 2.9389,
      "theoretical_loss": 3.8155961043985194,
      "tokens_seen": 637861888
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040743229689067204,
      "loss": 3.2684,
      "theoretical_loss": 3.8155556389841783,
      "tokens_seen": 637927424
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004074222668004012,
      "loss": 2.8674,
      "theoretical_loss": 3.8155151788906014,
      "tokens_seen": 637992960
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004074122367101304,
      "loss": 2.8272,
      "theoretical_loss": 3.8154747241165436,
      "tokens_seen": 638058496
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004074022066198596,
      "loss": 3.2811,
      "theoretical_loss": 3.815434274660759,
      "tokens_seen": 638124032
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040739217652958876,
      "loss": 3.254,
      "theoretical_loss": 3.8153938305220025,
      "tokens_seen": 638189568
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.000407382146439318,
      "loss": 3.2367,
      "theoretical_loss": 3.8153533916990288,
      "tokens_seen": 638255104
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004073721163490471,
      "loss": 2.9719,
      "theoretical_loss": 3.815312958190594,
      "tokens_seen": 638320640
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040736208625877636,
      "loss": 2.771,
      "theoretical_loss": 3.815272529995454,
      "tokens_seen": 638386176
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004073520561685055,
      "loss": 2.7862,
      "theoretical_loss": 3.8152321071123643,
      "tokens_seen": 638451712
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004073420260782347,
      "loss": 2.8181,
      "theoretical_loss": 3.8151916895400833,
      "tokens_seen": 638517248
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004073319959879639,
      "loss": 2.8391,
      "theoretical_loss": 3.8151512772773675,
      "tokens_seen": 638582784
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004073219658976931,
      "loss": 3.0907,
      "theoretical_loss": 3.8151108703229744,
      "tokens_seen": 638648320
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040731193580742227,
      "loss": 3.1267,
      "theoretical_loss": 3.8150704686756627,
      "tokens_seen": 638713856
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004073019057171515,
      "loss": 2.8035,
      "theoretical_loss": 3.8150300723341903,
      "tokens_seen": 638779392
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040729187562688063,
      "loss": 3.0481,
      "theoretical_loss": 3.814989681297317,
      "tokens_seen": 638844928
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040728184553660986,
      "loss": 3.1043,
      "theoretical_loss": 3.8149492955638014,
      "tokens_seen": 638910464
    },
    {
      "epoch": 0.19,
      "objective/train/advantage_avg": 0.4932076036930084,
      "objective/train/docs_used": 370882,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.0491318702697754,
      "objective/train/original_loss": 3.0491316318511963,
      "objective/train/theoretical_loss": 3.8149089151324036,
      "objective/train/tokens_used": 659436000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24334897100925446,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505577325820923,
      "objective/train/weighted_lm_loss": 3.203434705734253,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0417574644088745,
      "theoretical_loss": 3.8149089151324036,
      "tokens_seen": 638976000
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.000407271815446339,
      "loss": 2.9542,
      "theoretical_loss": 3.8149089151324036,
      "tokens_seen": 638976000
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004072617853560682,
      "loss": 2.9565,
      "theoretical_loss": 3.814868540001883,
      "tokens_seen": 639041536
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004072517552657974,
      "loss": 3.1415,
      "theoretical_loss": 3.814828170171002,
      "tokens_seen": 639107072
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004072417251755266,
      "loss": 3.0967,
      "theoretical_loss": 3.8147878056385203,
      "tokens_seen": 639172608
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040723169508525577,
      "loss": 3.2724,
      "theoretical_loss": 3.814747446403199,
      "tokens_seen": 639238144
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040722166499498495,
      "loss": 3.1796,
      "theoretical_loss": 3.814707092463801,
      "tokens_seen": 639303680
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040721163490471413,
      "loss": 3.1635,
      "theoretical_loss": 3.814666743819088,
      "tokens_seen": 639369216
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040720160481444337,
      "loss": 3.133,
      "theoretical_loss": 3.814626400467823,
      "tokens_seen": 639434752
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004071915747241725,
      "loss": 2.8343,
      "theoretical_loss": 3.814586062408769,
      "tokens_seen": 639500288
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040718154463390173,
      "loss": 2.9663,
      "theoretical_loss": 3.81454572964069,
      "tokens_seen": 639565824
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004071715145436309,
      "loss": 2.8242,
      "theoretical_loss": 3.814505402162349,
      "tokens_seen": 639631360
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004071614844533601,
      "loss": 3.2247,
      "theoretical_loss": 3.8144650799725106,
      "tokens_seen": 639696896
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040715145436308927,
      "loss": 3.0267,
      "theoretical_loss": 3.8144247630699404,
      "tokens_seen": 639762432
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040714142427281845,
      "loss": 3.3019,
      "theoretical_loss": 3.8143844514534027,
      "tokens_seen": 639827968
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040713139418254763,
      "loss": 3.1385,
      "theoretical_loss": 3.8143441451216633,
      "tokens_seen": 639893504
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040712136409227687,
      "loss": 3.1209,
      "theoretical_loss": 3.814303844073488,
      "tokens_seen": 639959040
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.000407111334002006,
      "loss": 2.9054,
      "theoretical_loss": 3.8142635483076446,
      "tokens_seen": 640024576
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040710130391173523,
      "loss": 3.2456,
      "theoretical_loss": 3.8142232578228983,
      "tokens_seen": 640090112
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040709127382146436,
      "loss": 2.9878,
      "theoretical_loss": 3.814182972618017,
      "tokens_seen": 640155648
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004070812437311936,
      "loss": 2.9156,
      "theoretical_loss": 3.8141426926917683,
      "tokens_seen": 640221184
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004070712136409228,
      "loss": 3.1361,
      "theoretical_loss": 3.8141024180429204,
      "tokens_seen": 640286720
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040706118355065196,
      "loss": 3.0272,
      "theoretical_loss": 3.814062148670242,
      "tokens_seen": 640352256
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040705115346038114,
      "loss": 2.843,
      "theoretical_loss": 3.814021884572502,
      "tokens_seen": 640417792
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004070411233701103,
      "loss": 3.1477,
      "theoretical_loss": 3.813981625748469,
      "tokens_seen": 640483328
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004070310932798395,
      "loss": 3.0275,
      "theoretical_loss": 3.813941372196914,
      "tokens_seen": 640548864
    },
    {
      "epoch": 0.19,
      "objective/train/advantage_avg": 0.49450546503067017,
      "objective/train/docs_used": 371543,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.856275796890259,
      "objective/train/original_loss": 2.856276273727417,
      "objective/train/theoretical_loss": 3.8139011239166054,
      "objective/train/tokens_used": 661074400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24459320306777954,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506938695907593,
      "objective/train/weighted_lm_loss": 3.001183271408081,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0413066148757935,
      "theoretical_loss": 3.8139011239166054,
      "tokens_seen": 640614400
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040702106318956874,
      "loss": 3.0965,
      "theoretical_loss": 3.8139011239166054,
      "tokens_seen": 640614400
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040701103309929786,
      "loss": 2.9623,
      "theoretical_loss": 3.813860880906316,
      "tokens_seen": 640679936
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004070010030090271,
      "loss": 3.0618,
      "theoretical_loss": 3.8138206431648145,
      "tokens_seen": 640745472
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004069909729187563,
      "loss": 3.1166,
      "theoretical_loss": 3.8137804106908737,
      "tokens_seen": 640811008
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040698094282848546,
      "loss": 2.9099,
      "theoretical_loss": 3.8137401834832656,
      "tokens_seen": 640876544
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040697091273821464,
      "loss": 3.0518,
      "theoretical_loss": 3.8136999615407614,
      "tokens_seen": 640942080
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004069608826479438,
      "loss": 2.958,
      "theoretical_loss": 3.8136597448621345,
      "tokens_seen": 641007616
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.000406950852557673,
      "loss": 3.3231,
      "theoretical_loss": 3.8136195334461576,
      "tokens_seen": 641073152
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040694082246740224,
      "loss": 2.9672,
      "theoretical_loss": 3.8135793272916043,
      "tokens_seen": 641138688
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040693079237713137,
      "loss": 3.0803,
      "theoretical_loss": 3.8135391263972487,
      "tokens_seen": 641204224
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004069207622868606,
      "loss": 2.8453,
      "theoretical_loss": 3.813498930761864,
      "tokens_seen": 641269760
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040691073219658973,
      "loss": 2.965,
      "theoretical_loss": 3.8134587403842266,
      "tokens_seen": 641335296
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040690070210631896,
      "loss": 3.1938,
      "theoretical_loss": 3.8134185552631097,
      "tokens_seen": 641400832
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004068906720160482,
      "loss": 3.0429,
      "theoretical_loss": 3.8133783753972903,
      "tokens_seen": 641466368
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004068806419257773,
      "loss": 3.1776,
      "theoretical_loss": 3.8133382007855445,
      "tokens_seen": 641531904
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040687061183550656,
      "loss": 3.1824,
      "theoretical_loss": 3.8132980314266476,
      "tokens_seen": 641597440
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004068605817452357,
      "loss": 3.0044,
      "theoretical_loss": 3.8132578673193755,
      "tokens_seen": 641662976
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004068505516549649,
      "loss": 3.0477,
      "theoretical_loss": 3.813217708462508,
      "tokens_seen": 641728512
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004068405215646941,
      "loss": 2.9274,
      "theoretical_loss": 3.8131775548548204,
      "tokens_seen": 641794048
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004068304914744233,
      "loss": 3.0869,
      "theoretical_loss": 3.813137406495092,
      "tokens_seen": 641859584
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040682046138415247,
      "loss": 2.8224,
      "theoretical_loss": 3.8130972633821,
      "tokens_seen": 641925120
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004068104312938817,
      "loss": 2.9872,
      "theoretical_loss": 3.8130571255146246,
      "tokens_seen": 641990656
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040680040120361083,
      "loss": 2.9837,
      "theoretical_loss": 3.813016992891444,
      "tokens_seen": 642056192
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040679037111334006,
      "loss": 3.0829,
      "theoretical_loss": 3.812976865511337,
      "tokens_seen": 642121728
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004067803410230692,
      "loss": 2.7923,
      "theoretical_loss": 3.8129367433730863,
      "tokens_seen": 642187264
    },
    {
      "epoch": 0.19,
      "objective/train/advantage_avg": 0.49417826533317566,
      "objective/train/docs_used": 373013,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.0631229877471924,
      "objective/train/original_loss": 3.0631234645843506,
      "objective/train/theoretical_loss": 3.81289662647547,
      "objective/train/tokens_used": 662712800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24427801370620728,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506595373153687,
      "objective/train/weighted_lm_loss": 3.2182188034057617,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0396549701690674,
      "theoretical_loss": 3.81289662647547,
      "tokens_seen": 642252800
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004067703109327984,
      "loss": 3.0777,
      "theoretical_loss": 3.81289662647547,
      "tokens_seen": 642252800
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004067602808425276,
      "loss": 3.1291,
      "theoretical_loss": 3.8128565148172693,
      "tokens_seen": 642318336
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004067502507522568,
      "loss": 3.1252,
      "theoretical_loss": 3.812816408397266,
      "tokens_seen": 642383872
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040674022066198597,
      "loss": 3.0459,
      "theoretical_loss": 3.812776307214241,
      "tokens_seen": 642449408
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040673019057171515,
      "loss": 2.9913,
      "theoretical_loss": 3.8127362112669774,
      "tokens_seen": 642514944
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040672016048144433,
      "loss": 2.7712,
      "theoretical_loss": 3.812696120554257,
      "tokens_seen": 642580480
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040671013039117357,
      "loss": 2.8677,
      "theoretical_loss": 3.8126560350748626,
      "tokens_seen": 642646016
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004067001003009027,
      "loss": 3.0131,
      "theoretical_loss": 3.812615954827578,
      "tokens_seen": 642711552
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040669007021063193,
      "loss": 3.3288,
      "theoretical_loss": 3.8125758798111864,
      "tokens_seen": 642777088
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004066800401203611,
      "loss": 2.6377,
      "theoretical_loss": 3.8125358100244715,
      "tokens_seen": 642842624
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004066700100300903,
      "loss": 2.9553,
      "theoretical_loss": 3.8124957454662187,
      "tokens_seen": 642908160
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040665997993981947,
      "loss": 2.991,
      "theoretical_loss": 3.8124556861352126,
      "tokens_seen": 642973696
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040664994984954865,
      "loss": 3.0554,
      "theoretical_loss": 3.8124156320302385,
      "tokens_seen": 643039232
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040663991975927783,
      "loss": 2.9247,
      "theoretical_loss": 3.8123755831500814,
      "tokens_seen": 643104768
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040662988966900707,
      "loss": 2.9518,
      "theoretical_loss": 3.8123355394935285,
      "tokens_seen": 643170304
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004066198595787362,
      "loss": 3.0329,
      "theoretical_loss": 3.812295501059366,
      "tokens_seen": 643235840
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040660982948846543,
      "loss": 3.3443,
      "theoretical_loss": 3.8122554678463807,
      "tokens_seen": 643301376
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.00040659979939819456,
      "loss": 2.8862,
      "theoretical_loss": 3.8122154398533596,
      "tokens_seen": 643366912
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.0004065897693079238,
      "loss": 3.0606,
      "theoretical_loss": 3.812175417079091,
      "tokens_seen": 643432448
    },
    {
      "epoch": 0.19,
      "learning_rate": 0.000406579739217653,
      "loss": 2.9849,
      "theoretical_loss": 3.8121353995223624,
      "tokens_seen": 643497984
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040656970912738216,
      "loss": 3.0809,
      "theoretical_loss": 3.812095387181963,
      "tokens_seen": 643563520
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040655967903711134,
      "loss": 2.92,
      "theoretical_loss": 3.812055380056682,
      "tokens_seen": 643629056
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004065496489468405,
      "loss": 2.9178,
      "theoretical_loss": 3.812015378145308,
      "tokens_seen": 643694592
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004065396188565697,
      "loss": 3.0435,
      "theoretical_loss": 3.811975381446631,
      "tokens_seen": 643760128
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040652958876629894,
      "loss": 3.2576,
      "theoretical_loss": 3.8119353899594413,
      "tokens_seen": 643825664
    },
    {
      "epoch": 0.2,
      "objective/train/advantage_avg": 0.4939141571521759,
      "objective/train/docs_used": 373533,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.698076009750366,
      "objective/train/original_loss": 2.6980762481689453,
      "objective/train/theoretical_loss": 3.811895403682529,
      "objective/train/tokens_used": 664351200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2440139204263687,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506318807601929,
      "objective/train/weighted_lm_loss": 2.8346660137176514,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0338950157165527,
      "theoretical_loss": 3.811895403682529,
      "tokens_seen": 643891200
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040651955867602806,
      "loss": 3.0744,
      "theoretical_loss": 3.811895403682529,
      "tokens_seen": 643891200
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004065095285857573,
      "loss": 2.7825,
      "theoretical_loss": 3.811855422614686,
      "tokens_seen": 643956736
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004064994984954865,
      "loss": 2.9321,
      "theoretical_loss": 3.811815446754703,
      "tokens_seen": 644022272
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040648946840521566,
      "loss": 2.9798,
      "theoretical_loss": 3.8117754761013725,
      "tokens_seen": 644087808
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040647943831494484,
      "loss": 2.8923,
      "theoretical_loss": 3.8117355106534854,
      "tokens_seen": 644153344
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.000406469408224674,
      "loss": 3.0501,
      "theoretical_loss": 3.811695550409835,
      "tokens_seen": 644218880
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004064593781344032,
      "loss": 2.9954,
      "theoretical_loss": 3.8116555953692144,
      "tokens_seen": 644284416
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040644934804413244,
      "loss": 3.0747,
      "theoretical_loss": 3.811615645530417,
      "tokens_seen": 644349952
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040643931795386157,
      "loss": 3.1948,
      "theoretical_loss": 3.811575700892236,
      "tokens_seen": 644415488
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004064292878635908,
      "loss": 3.0149,
      "theoretical_loss": 3.8115357614534666,
      "tokens_seen": 644481024
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040641925777331993,
      "loss": 3.0287,
      "theoretical_loss": 3.8114958272129034,
      "tokens_seen": 644546560
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040640922768304916,
      "loss": 2.905,
      "theoretical_loss": 3.8114558981693403,
      "tokens_seen": 644612096
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040639919759277834,
      "loss": 2.9702,
      "theoretical_loss": 3.8114159743215734,
      "tokens_seen": 644677632
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004063891675025075,
      "loss": 2.9662,
      "theoretical_loss": 3.811376055668398,
      "tokens_seen": 644743168
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004063791374122367,
      "loss": 2.9989,
      "theoretical_loss": 3.8113361422086114,
      "tokens_seen": 644808704
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004063691073219659,
      "loss": 3.0102,
      "theoretical_loss": 3.8112962339410092,
      "tokens_seen": 644874240
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040635907723169507,
      "loss": 2.8812,
      "theoretical_loss": 3.8112563308643885,
      "tokens_seen": 644939776
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004063490471414243,
      "loss": 3.0345,
      "theoretical_loss": 3.8112164329775475,
      "tokens_seen": 645005312
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040633901705115343,
      "loss": 3.0368,
      "theoretical_loss": 3.811176540279283,
      "tokens_seen": 645070848
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040632898696088267,
      "loss": 3.2544,
      "theoretical_loss": 3.8111366527683943,
      "tokens_seen": 645136384
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040631895687061185,
      "loss": 3.0538,
      "theoretical_loss": 3.8110967704436787,
      "tokens_seen": 645201920
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040630892678034103,
      "loss": 2.8662,
      "theoretical_loss": 3.8110568933039364,
      "tokens_seen": 645267456
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004062988966900702,
      "loss": 3.1226,
      "theoretical_loss": 3.811017021347966,
      "tokens_seen": 645332992
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004062888665997994,
      "loss": 3.1838,
      "theoretical_loss": 3.810977154574568,
      "tokens_seen": 645398528
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040627883650952857,
      "loss": 2.972,
      "theoretical_loss": 3.8109372929825422,
      "tokens_seen": 645464064
    },
    {
      "epoch": 0.2,
      "objective/train/advantage_avg": 0.4958713948726654,
      "objective/train/docs_used": 374553,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.4572575092315674,
      "objective/train/original_loss": 3.4572575092315674,
      "objective/train/theoretical_loss": 3.8108974365706887,
      "objective/train/tokens_used": 665989600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2459346503019333,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0508372783660889,
      "objective/train/weighted_lm_loss": 3.6329286098480225,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0468999147415161,
      "theoretical_loss": 3.8108974365706887,
      "tokens_seen": 645529600
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004062688064192578,
      "loss": 3.1292,
      "theoretical_loss": 3.8108974365706887,
      "tokens_seen": 645529600
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040625877632898693,
      "loss": 2.8937,
      "theoretical_loss": 3.8108575853378097,
      "tokens_seen": 645595136
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040624874623871617,
      "loss": 3.2604,
      "theoretical_loss": 3.8108177392827063,
      "tokens_seen": 645660672
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004062387161484453,
      "loss": 3.0076,
      "theoretical_loss": 3.810777898404179,
      "tokens_seen": 645726208
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040622868605817453,
      "loss": 2.9791,
      "theoretical_loss": 3.8107380627010317,
      "tokens_seen": 645791744
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004062186559679037,
      "loss": 3.169,
      "theoretical_loss": 3.810698232172066,
      "tokens_seen": 645857280
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004062086258776329,
      "loss": 2.999,
      "theoretical_loss": 3.810658406816085,
      "tokens_seen": 645922816
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004061985957873621,
      "loss": 2.8678,
      "theoretical_loss": 3.810618586631893,
      "tokens_seen": 645988352
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040618856569709126,
      "loss": 3.1855,
      "theoretical_loss": 3.810578771618293,
      "tokens_seen": 646053888
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040617853560682044,
      "loss": 3.1866,
      "theoretical_loss": 3.810538961774089,
      "tokens_seen": 646119424
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040616850551654967,
      "loss": 3.1117,
      "theoretical_loss": 3.8104991570980857,
      "tokens_seen": 646184960
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004061584754262788,
      "loss": 3.1846,
      "theoretical_loss": 3.810459357589089,
      "tokens_seen": 646250496
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040614844533600804,
      "loss": 3.1276,
      "theoretical_loss": 3.8104195632459033,
      "tokens_seen": 646316032
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040613841524573727,
      "loss": 2.9176,
      "theoretical_loss": 3.8103797740673344,
      "tokens_seen": 646381568
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004061283851554664,
      "loss": 2.9858,
      "theoretical_loss": 3.8103399900521895,
      "tokens_seen": 646447104
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040611835506519563,
      "loss": 3.4177,
      "theoretical_loss": 3.810300211199274,
      "tokens_seen": 646512640
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040610832497492476,
      "loss": 3.0091,
      "theoretical_loss": 3.810260437507395,
      "tokens_seen": 646578176
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.000406098294884654,
      "loss": 3.3793,
      "theoretical_loss": 3.810220668975361,
      "tokens_seen": 646643712
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004060882647943832,
      "loss": 3.2083,
      "theoretical_loss": 3.810180905601979,
      "tokens_seen": 646709248
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040607823470411236,
      "loss": 3.006,
      "theoretical_loss": 3.810141147386057,
      "tokens_seen": 646774784
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040606820461384154,
      "loss": 3.2925,
      "theoretical_loss": 3.8101013943264035,
      "tokens_seen": 646840320
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004060581745235707,
      "loss": 3.034,
      "theoretical_loss": 3.810061646421828,
      "tokens_seen": 646905856
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004060481444332999,
      "loss": 3.288,
      "theoretical_loss": 3.8100219036711396,
      "tokens_seen": 646971392
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040603811434302914,
      "loss": 3.0253,
      "theoretical_loss": 3.809982166073148,
      "tokens_seen": 647036928
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040602808425275826,
      "loss": 3.1188,
      "theoretical_loss": 3.8099424336266634,
      "tokens_seen": 647102464
    },
    {
      "epoch": 0.2,
      "objective/train/advantage_avg": 0.4925914704799652,
      "objective/train/docs_used": 375146,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.901201009750366,
      "objective/train/original_loss": 2.901201009750366,
      "objective/train/theoretical_loss": 3.8099027063304964,
      "objective/train/tokens_used": 667628000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24278397858142853,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0504932403564453,
      "objective/train/weighted_lm_loss": 3.047834634780884,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0304546356201172,
      "theoretical_loss": 3.8099027063304964,
      "tokens_seen": 647168000
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004060180541624875,
      "loss": 3.1507,
      "theoretical_loss": 3.8099027063304964,
      "tokens_seen": 647168000
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004060080240722167,
      "loss": 3.1944,
      "theoretical_loss": 3.8098629841834577,
      "tokens_seen": 647233536
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040599799398194586,
      "loss": 3.0543,
      "theoretical_loss": 3.809823267184359,
      "tokens_seen": 647299072
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040598796389167504,
      "loss": 2.8861,
      "theoretical_loss": 3.809783555332012,
      "tokens_seen": 647364608
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004059779338014042,
      "loss": 3.207,
      "theoretical_loss": 3.809743848625228,
      "tokens_seen": 647430144
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004059679037111334,
      "loss": 3.1532,
      "theoretical_loss": 3.8097041470628206,
      "tokens_seen": 647495680
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040595787362086264,
      "loss": 3.1635,
      "theoretical_loss": 3.809664450643602,
      "tokens_seen": 647561216
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040594784353059177,
      "loss": 2.9077,
      "theoretical_loss": 3.8096247593663866,
      "tokens_seen": 647626752
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.000405937813440321,
      "loss": 3.1802,
      "theoretical_loss": 3.8095850732299867,
      "tokens_seen": 647692288
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040592778335005013,
      "loss": 3.0506,
      "theoretical_loss": 3.8095453922332174,
      "tokens_seen": 647757824
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040591775325977936,
      "loss": 2.9637,
      "theoretical_loss": 3.8095057163748924,
      "tokens_seen": 647823360
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040590772316950854,
      "loss": 3.0319,
      "theoretical_loss": 3.809466045653827,
      "tokens_seen": 647888896
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004058976930792377,
      "loss": 3.0497,
      "theoretical_loss": 3.8094263800688366,
      "tokens_seen": 647954432
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004058876629889669,
      "loss": 3.1754,
      "theoretical_loss": 3.809386719618737,
      "tokens_seen": 648019968
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004058776328986961,
      "loss": 3.2111,
      "theoretical_loss": 3.8093470643023433,
      "tokens_seen": 648085504
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040586760280842527,
      "loss": 3.1755,
      "theoretical_loss": 3.809307414118473,
      "tokens_seen": 648151040
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004058575727181545,
      "loss": 2.9784,
      "theoretical_loss": 3.8092677690659427,
      "tokens_seen": 648216576
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040584754262788363,
      "loss": 3.3791,
      "theoretical_loss": 3.809228129143569,
      "tokens_seen": 648282112
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040583751253761287,
      "loss": 3.0592,
      "theoretical_loss": 3.8091884943501704,
      "tokens_seen": 648347648
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040582748244734205,
      "loss": 2.7524,
      "theoretical_loss": 3.8091488646845644,
      "tokens_seen": 648413184
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040581745235707123,
      "loss": 3.1614,
      "theoretical_loss": 3.80910924014557,
      "tokens_seen": 648478720
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004058074222668004,
      "loss": 2.8345,
      "theoretical_loss": 3.8090696207320054,
      "tokens_seen": 648544256
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004057973921765296,
      "loss": 3.012,
      "theoretical_loss": 3.80903000644269,
      "tokens_seen": 648609792
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040578736208625877,
      "loss": 3.2751,
      "theoretical_loss": 3.808990397276443,
      "tokens_seen": 648675328
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.000405777331995988,
      "loss": 3.1333,
      "theoretical_loss": 3.808950793232085,
      "tokens_seen": 648740864
    },
    {
      "epoch": 0.2,
      "objective/train/advantage_avg": 0.4957117438316345,
      "objective/train/docs_used": 375860,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.1292247772216797,
      "objective/train/original_loss": 3.1292245388031006,
      "objective/train/theoretical_loss": 3.808911194308436,
      "objective/train/tokens_used": 669266400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24579770863056183,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0508207082748413,
      "objective/train/weighted_lm_loss": 3.2881851196289062,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0304546356201172,
      "theoretical_loss": 3.808911194308436,
      "tokens_seen": 648806400
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040576730190571713,
      "loss": 3.0904,
      "theoretical_loss": 3.808911194308436,
      "tokens_seen": 648806400
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040575727181544637,
      "loss": 3.14,
      "theoretical_loss": 3.8088716005043173,
      "tokens_seen": 648871936
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004057472417251755,
      "loss": 3.063,
      "theoretical_loss": 3.8088320118185495,
      "tokens_seen": 648937472
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040573721163490473,
      "loss": 3.218,
      "theoretical_loss": 3.808792428249954,
      "tokens_seen": 649003008
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004057271815446339,
      "loss": 3.3105,
      "theoretical_loss": 3.808752849797353,
      "tokens_seen": 649068544
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004057171514543631,
      "loss": 3.1336,
      "theoretical_loss": 3.808713276459569,
      "tokens_seen": 649134080
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004057071213640923,
      "loss": 3.2372,
      "theoretical_loss": 3.8086737082354247,
      "tokens_seen": 649199616
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040569709127382146,
      "loss": 3.2103,
      "theoretical_loss": 3.808634145123743,
      "tokens_seen": 649265152
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040568706118355064,
      "loss": 2.8491,
      "theoretical_loss": 3.8085945871233475,
      "tokens_seen": 649330688
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004056770310932799,
      "loss": 3.2064,
      "theoretical_loss": 3.8085550342330614,
      "tokens_seen": 649396224
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.000405667001003009,
      "loss": 3.1033,
      "theoretical_loss": 3.8085154864517103,
      "tokens_seen": 649461760
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040565697091273824,
      "loss": 3.211,
      "theoretical_loss": 3.808475943778118,
      "tokens_seen": 649527296
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004056469408224674,
      "loss": 3.1136,
      "theoretical_loss": 3.8084364062111096,
      "tokens_seen": 649592832
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004056369107321966,
      "loss": 3.0811,
      "theoretical_loss": 3.8083968737495106,
      "tokens_seen": 649658368
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004056268806419258,
      "loss": 3.1017,
      "theoretical_loss": 3.808357346392147,
      "tokens_seen": 649723904
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040561685055165496,
      "loss": 3.2541,
      "theoretical_loss": 3.808317824137845,
      "tokens_seen": 649789440
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040560682046138414,
      "loss": 2.8173,
      "theoretical_loss": 3.8082783069854305,
      "tokens_seen": 649854976
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004055967903711134,
      "loss": 2.9396,
      "theoretical_loss": 3.8082387949337315,
      "tokens_seen": 649920512
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004055867602808425,
      "loss": 3.0381,
      "theoretical_loss": 3.8081992879815747,
      "tokens_seen": 649986048
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040557673019057174,
      "loss": 3.1864,
      "theoretical_loss": 3.8081597861277885,
      "tokens_seen": 650051584
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040556670010030087,
      "loss": 3.0723,
      "theoretical_loss": 3.8081202893712005,
      "tokens_seen": 650117120
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004055566700100301,
      "loss": 3.0542,
      "theoretical_loss": 3.8080807977106397,
      "tokens_seen": 650182656
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004055466399197593,
      "loss": 3.2076,
      "theoretical_loss": 3.808041311144934,
      "tokens_seen": 650248192
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040553660982948846,
      "loss": 2.9668,
      "theoretical_loss": 3.8080018296729143,
      "tokens_seen": 650313728
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040552657973921764,
      "loss": 2.9198,
      "theoretical_loss": 3.8079623532934086,
      "tokens_seen": 650379264
    },
    {
      "epoch": 0.2,
      "objective/train/advantage_avg": 0.49582216143608093,
      "objective/train/docs_used": 377374,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.65570330619812,
      "objective/train/original_loss": 2.655703067779541,
      "objective/train/theoretical_loss": 3.807922882005249,
      "objective/train/tokens_used": 670904800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2458837330341339,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0508320331573486,
      "objective/train/weighted_lm_loss": 2.7905735969543457,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0425469875335693,
      "theoretical_loss": 3.807922882005249,
      "tokens_seen": 650444800
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004055165496489469,
      "loss": 2.8678,
      "theoretical_loss": 3.807922882005249,
      "tokens_seen": 650444800
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.000405506519558676,
      "loss": 3.0724,
      "theoretical_loss": 3.807883415807264,
      "tokens_seen": 650510336
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040549648946840524,
      "loss": 3.1168,
      "theoretical_loss": 3.8078439546982854,
      "tokens_seen": 650575872
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040548645937813437,
      "loss": 2.7986,
      "theoretical_loss": 3.807804498677145,
      "tokens_seen": 650641408
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004054764292878636,
      "loss": 3.0233,
      "theoretical_loss": 3.8077650477426728,
      "tokens_seen": 650706944
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004054663991975928,
      "loss": 2.9552,
      "theoretical_loss": 3.8077256018937025,
      "tokens_seen": 650772480
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040545636910732197,
      "loss": 3.3128,
      "theoretical_loss": 3.807686161129065,
      "tokens_seen": 650838016
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040544633901705115,
      "loss": 3.0617,
      "theoretical_loss": 3.807646725447595,
      "tokens_seen": 650903552
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040543630892678033,
      "loss": 3.0236,
      "theoretical_loss": 3.8076072948481237,
      "tokens_seen": 650969088
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004054262788365095,
      "loss": 2.9449,
      "theoretical_loss": 3.8075678693294863,
      "tokens_seen": 651034624
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040541624874623874,
      "loss": 3.0561,
      "theoretical_loss": 3.8075284488905154,
      "tokens_seen": 651100160
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040540621865596787,
      "loss": 3.1633,
      "theoretical_loss": 3.807489033530046,
      "tokens_seen": 651165696
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004053961885656971,
      "loss": 2.8894,
      "theoretical_loss": 3.807449623246913,
      "tokens_seen": 651231232
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004053861584754263,
      "loss": 3.1328,
      "theoretical_loss": 3.8074102180399514,
      "tokens_seen": 651296768
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040537612838515547,
      "loss": 2.9372,
      "theoretical_loss": 3.807370817907996,
      "tokens_seen": 651362304
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004053660982948847,
      "loss": 3.0735,
      "theoretical_loss": 3.8073314228498836,
      "tokens_seen": 651427840
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040535606820461383,
      "loss": 3.1576,
      "theoretical_loss": 3.80729203286445,
      "tokens_seen": 651493376
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040534603811434307,
      "loss": 3.0321,
      "theoretical_loss": 3.8072526479505315,
      "tokens_seen": 651558912
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040533600802407225,
      "loss": 3.0831,
      "theoretical_loss": 3.807213268106966,
      "tokens_seen": 651624448
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040532597793380143,
      "loss": 3.0577,
      "theoretical_loss": 3.8071738933325903,
      "tokens_seen": 651689984
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004053159478435306,
      "loss": 2.9893,
      "theoretical_loss": 3.8071345236262424,
      "tokens_seen": 651755520
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004053059177532598,
      "loss": 2.8375,
      "theoretical_loss": 3.8070951589867605,
      "tokens_seen": 651821056
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040529588766298897,
      "loss": 3.2334,
      "theoretical_loss": 3.807055799412983,
      "tokens_seen": 651886592
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004052858575727182,
      "loss": 2.8973,
      "theoretical_loss": 3.8070164449037485,
      "tokens_seen": 651952128
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040527582748244733,
      "loss": 3.1853,
      "theoretical_loss": 3.8069770954578974,
      "tokens_seen": 652017664
    },
    {
      "epoch": 0.2,
      "objective/train/advantage_avg": 0.4939846098423004,
      "objective/train/docs_used": 377873,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.000206708908081,
      "objective/train/original_loss": 3.000206470489502,
      "objective/train/theoretical_loss": 3.806937751074268,
      "objective/train/tokens_used": 672543200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24408704042434692,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050639271736145,
      "objective/train/weighted_lm_loss": 3.151867389678955,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0445106029510498,
      "theoretical_loss": 3.806937751074268,
      "tokens_seen": 652083200
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040526579739217657,
      "loss": 3.1751,
      "theoretical_loss": 3.806937751074268,
      "tokens_seen": 652083200
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004052557673019057,
      "loss": 3.1766,
      "theoretical_loss": 3.8068984117517015,
      "tokens_seen": 652148736
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040524573721163493,
      "loss": 3.2398,
      "theoretical_loss": 3.806859077489038,
      "tokens_seen": 652214272
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004052357071213641,
      "loss": 2.9537,
      "theoretical_loss": 3.8068197482851183,
      "tokens_seen": 652279808
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004052256770310933,
      "loss": 2.818,
      "theoretical_loss": 3.806780424138784,
      "tokens_seen": 652345344
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004052156469408225,
      "loss": 3.0505,
      "theoretical_loss": 3.806741105048876,
      "tokens_seen": 652410880
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040520561685055166,
      "loss": 3.0991,
      "theoretical_loss": 3.8067017910142367,
      "tokens_seen": 652476416
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040519558676028084,
      "loss": 3.2127,
      "theoretical_loss": 3.8066624820337083,
      "tokens_seen": 652541952
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004051855566700101,
      "loss": 3.2775,
      "theoretical_loss": 3.8066231781061344,
      "tokens_seen": 652607488
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004051755265797392,
      "loss": 2.7746,
      "theoretical_loss": 3.806583879230357,
      "tokens_seen": 652673024
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040516549648946844,
      "loss": 2.8663,
      "theoretical_loss": 3.80654458540522,
      "tokens_seen": 652738560
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004051554663991976,
      "loss": 2.7501,
      "theoretical_loss": 3.806505296629567,
      "tokens_seen": 652804096
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004051454363089268,
      "loss": 3.1158,
      "theoretical_loss": 3.8064660129022436,
      "tokens_seen": 652869632
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.000405135406218656,
      "loss": 2.8884,
      "theoretical_loss": 3.806426734222093,
      "tokens_seen": 652935168
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040512537612838516,
      "loss": 3.155,
      "theoretical_loss": 3.806387460587961,
      "tokens_seen": 653000704
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040511534603811434,
      "loss": 3.2342,
      "theoretical_loss": 3.806348191998693,
      "tokens_seen": 653066240
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004051053159478436,
      "loss": 2.8873,
      "theoretical_loss": 3.806308928453134,
      "tokens_seen": 653131776
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004050952858575727,
      "loss": 3.096,
      "theoretical_loss": 3.8062696699501313,
      "tokens_seen": 653197312
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040508525576730194,
      "loss": 3.1619,
      "theoretical_loss": 3.806230416488531,
      "tokens_seen": 653262848
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040507522567703107,
      "loss": 3.064,
      "theoretical_loss": 3.80619116806718,
      "tokens_seen": 653328384
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004050651955867603,
      "loss": 3.0323,
      "theoretical_loss": 3.806151924684925,
      "tokens_seen": 653393920
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004050551654964895,
      "loss": 3.0558,
      "theoretical_loss": 3.8061126863406147,
      "tokens_seen": 653459456
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040504513540621866,
      "loss": 3.1922,
      "theoretical_loss": 3.806073453033097,
      "tokens_seen": 653524992
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040503510531594784,
      "loss": 2.7584,
      "theoretical_loss": 3.8060342247612207,
      "tokens_seen": 653590528
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004050250752256771,
      "loss": 3.1123,
      "theoretical_loss": 3.805995001523833,
      "tokens_seen": 653656064
    },
    {
      "epoch": 0.2,
      "objective/train/advantage_avg": 0.495535284280777,
      "objective/train/docs_used": 379015,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.9806227684020996,
      "objective/train/original_loss": 2.9806227684020996,
      "objective/train/theoretical_loss": 3.805955783319785,
      "objective/train/tokens_used": 674181600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24561432003974915,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0508019924163818,
      "objective/train/weighted_lm_loss": 3.131927013397217,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0396549701690674,
      "theoretical_loss": 3.805955783319785,
      "tokens_seen": 653721600
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004050150451354062,
      "loss": 2.9773,
      "theoretical_loss": 3.805955783319785,
      "tokens_seen": 653721600
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040500501504513544,
      "loss": 3.0198,
      "theoretical_loss": 3.805916570147925,
      "tokens_seen": 653787136
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040499498495486457,
      "loss": 3.0641,
      "theoretical_loss": 3.805877362007104,
      "tokens_seen": 653852672
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004049849548645938,
      "loss": 2.9541,
      "theoretical_loss": 3.8058381588961714,
      "tokens_seen": 653918208
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.000404974924774323,
      "loss": 3.1192,
      "theoretical_loss": 3.805798960813979,
      "tokens_seen": 653983744
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040496489468405217,
      "loss": 2.7449,
      "theoretical_loss": 3.805759767759377,
      "tokens_seen": 654049280
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040495486459378135,
      "loss": 2.9401,
      "theoretical_loss": 3.805720579731217,
      "tokens_seen": 654114816
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040494483450351053,
      "loss": 3.0493,
      "theoretical_loss": 3.8056813967283514,
      "tokens_seen": 654180352
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004049348044132397,
      "loss": 2.7665,
      "theoretical_loss": 3.8056422187496315,
      "tokens_seen": 654245888
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040492477432296894,
      "loss": 3.0583,
      "theoretical_loss": 3.8056030457939114,
      "tokens_seen": 654311424
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040491474423269807,
      "loss": 2.9546,
      "theoretical_loss": 3.8055638778600427,
      "tokens_seen": 654376960
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004049047141424273,
      "loss": 3.0619,
      "theoretical_loss": 3.805524714946879,
      "tokens_seen": 654442496
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040489468405215643,
      "loss": 2.9912,
      "theoretical_loss": 3.8054855570532746,
      "tokens_seen": 654508032
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040488465396188567,
      "loss": 3.1537,
      "theoretical_loss": 3.8054464041780838,
      "tokens_seen": 654573568
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040487462387161485,
      "loss": 3.1525,
      "theoretical_loss": 3.80540725632016,
      "tokens_seen": 654639104
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040486459378134403,
      "loss": 2.8802,
      "theoretical_loss": 3.8053681134783592,
      "tokens_seen": 654704640
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004048545636910732,
      "loss": 3.1794,
      "theoretical_loss": 3.805328975651536,
      "tokens_seen": 654770176
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040484453360080245,
      "loss": 3.0124,
      "theoretical_loss": 3.8052898428385458,
      "tokens_seen": 654835712
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004048345035105316,
      "loss": 2.9746,
      "theoretical_loss": 3.8052507150382455,
      "tokens_seen": 654901248
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004048244734202608,
      "loss": 3.1231,
      "theoretical_loss": 3.8052115922494907,
      "tokens_seen": 654966784
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040481444332998994,
      "loss": 2.981,
      "theoretical_loss": 3.8051724744711386,
      "tokens_seen": 655032320
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040480441323971917,
      "loss": 2.9789,
      "theoretical_loss": 3.805133361702046,
      "tokens_seen": 655097856
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040479438314944835,
      "loss": 3.0543,
      "theoretical_loss": 3.805094253941071,
      "tokens_seen": 655163392
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040478435305917753,
      "loss": 3.0275,
      "theoretical_loss": 3.8050551511870707,
      "tokens_seen": 655228928
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004047743229689067,
      "loss": 2.9345,
      "theoretical_loss": 3.805016053438904,
      "tokens_seen": 655294464
    },
    {
      "debugging/Self-BLEU-5": 0.28723315400076466,
      "debugging/distinct-1-grams": 0.831755762171831,
      "debugging/distinct-2-grams": 0.9583018392542203,
      "debugging/entropy-1-grams": 5.453202391069032,
      "debugging/entropy-2-grams": 5.995093667215148,
      "debugging/length": 442.0,
      "debugging/num_segments": 7,
      "debugging/raw_token_scores_avg": 0.006729413755238056,
      "debugging/raw_token_scores_std": 0.008488241583108902,
      "debugging/score": 0.009230784774488613,
      "debugging/score_std": 0.005285791546055549,
      "epoch": 0.2,
      "objective/train/advantage_avg": 0.4932686984539032,
      "objective/train/docs_used": 379813,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.8795816898345947,
      "objective/train/original_loss": 2.879581928253174,
      "objective/train/theoretical_loss": 3.804976960695429,
      "objective/train/tokens_used": 675820000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24338610470294952,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505640506744385,
      "objective/train/weighted_lm_loss": 3.025219678878784,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0338950157165527,
      "theoretical_loss": 3.804976960695429,
      "tokens_seen": 655360000
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004047642928786359,
      "loss": 3.0975,
      "theoretical_loss": 3.804976960695429,
      "tokens_seen": 655360000
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004047542627883651,
      "loss": 2.903,
      "theoretical_loss": 3.804937872955505,
      "tokens_seen": 655425536
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004047442326980943,
      "loss": 3.4094,
      "theoretical_loss": 3.804898790217991,
      "tokens_seen": 655491072
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040473420260782344,
      "loss": 3.1951,
      "theoretical_loss": 3.804859712481748,
      "tokens_seen": 655556608
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004047241725175527,
      "loss": 2.9988,
      "theoretical_loss": 3.804820639745634,
      "tokens_seen": 655622144
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004047141424272818,
      "loss": 3.2313,
      "theoretical_loss": 3.804781572008511,
      "tokens_seen": 655687680
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040470411233701104,
      "loss": 2.9846,
      "theoretical_loss": 3.80474250926924,
      "tokens_seen": 655753216
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004046940822467402,
      "loss": 2.929,
      "theoretical_loss": 3.8047034515266818,
      "tokens_seen": 655818752
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004046840521564694,
      "loss": 3.1283,
      "theoretical_loss": 3.804664398779698,
      "tokens_seen": 655884288
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004046740220661986,
      "loss": 3.3101,
      "theoretical_loss": 3.80462535102715,
      "tokens_seen": 655949824
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004046639919759278,
      "loss": 3.3082,
      "theoretical_loss": 3.8045863082679015,
      "tokens_seen": 656015360
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.000404653961885657,
      "loss": 2.9339,
      "theoretical_loss": 3.8045472705008145,
      "tokens_seen": 656080896
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004046439317953862,
      "loss": 3.1908,
      "theoretical_loss": 3.804508237724752,
      "tokens_seen": 656146432
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040463390170511536,
      "loss": 3.0642,
      "theoretical_loss": 3.8044692099385777,
      "tokens_seen": 656211968
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040462387161484454,
      "loss": 2.9785,
      "theoretical_loss": 3.8044301871411554,
      "tokens_seen": 656277504
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004046138415245738,
      "loss": 3.1343,
      "theoretical_loss": 3.804391169331349,
      "tokens_seen": 656343040
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004046038114343029,
      "loss": 3.1976,
      "theoretical_loss": 3.8043521565080236,
      "tokens_seen": 656408576
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040459378134403214,
      "loss": 3.0393,
      "theoretical_loss": 3.804313148670044,
      "tokens_seen": 656474112
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040458375125376127,
      "loss": 3.0965,
      "theoretical_loss": 3.804274145816276,
      "tokens_seen": 656539648
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004045737211634905,
      "loss": 3.1929,
      "theoretical_loss": 3.804235147945584,
      "tokens_seen": 656605184
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004045636910732197,
      "loss": 3.2185,
      "theoretical_loss": 3.804196155056835,
      "tokens_seen": 656670720
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040455366098294886,
      "loss": 3.12,
      "theoretical_loss": 3.8041571671488956,
      "tokens_seen": 656736256
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040454363089267804,
      "loss": 2.9061,
      "theoretical_loss": 3.8041181842206324,
      "tokens_seen": 656801792
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004045336008024073,
      "loss": 3.1219,
      "theoretical_loss": 3.804079206270913,
      "tokens_seen": 656867328
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004045235707121364,
      "loss": 3.2063,
      "theoretical_loss": 3.804040233298604,
      "tokens_seen": 656932864
    },
    {
      "epoch": 0.2,
      "objective/train/advantage_avg": 0.492044061422348,
      "objective/train/docs_used": 381022,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.819871664047241,
      "objective/train/original_loss": 2.819871425628662,
      "objective/train/theoretical_loss": 3.804001265302574,
      "objective/train/tokens_used": 677458400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24219727516174316,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0504354238510132,
      "objective/train/weighted_lm_loss": 2.962068557739258,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0417574644088745,
      "theoretical_loss": 3.804001265302574,
      "tokens_seen": 656998400
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040451354062186564,
      "loss": 2.9399,
      "theoretical_loss": 3.804001265302574,
      "tokens_seen": 656998400
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040450351053159477,
      "loss": 3.2077,
      "theoretical_loss": 3.8039623022816915,
      "tokens_seen": 657063936
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.000404493480441324,
      "loss": 3.0602,
      "theoretical_loss": 3.803923344234825,
      "tokens_seen": 657129472
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004044834503510532,
      "loss": 3.0596,
      "theoretical_loss": 3.803884391160843,
      "tokens_seen": 657195008
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040447342026078237,
      "loss": 3.1008,
      "theoretical_loss": 3.8038454430586155,
      "tokens_seen": 657260544
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040446339017051155,
      "loss": 3.1746,
      "theoretical_loss": 3.803806499927012,
      "tokens_seen": 657326080
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040445336008024073,
      "loss": 3.1285,
      "theoretical_loss": 3.8037675617649036,
      "tokens_seen": 657391616
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004044433299899699,
      "loss": 3.4122,
      "theoretical_loss": 3.803728628571159,
      "tokens_seen": 657457152
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040443329989969915,
      "loss": 3.2138,
      "theoretical_loss": 3.8036897003446506,
      "tokens_seen": 657522688
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040442326980942827,
      "loss": 3.0279,
      "theoretical_loss": 3.8036507770842496,
      "tokens_seen": 657588224
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004044132397191575,
      "loss": 2.9981,
      "theoretical_loss": 3.803611858788827,
      "tokens_seen": 657653760
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040440320962888663,
      "loss": 3.3259,
      "theoretical_loss": 3.8035729454572547,
      "tokens_seen": 657719296
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040439317953861587,
      "loss": 2.7568,
      "theoretical_loss": 3.8035340370884057,
      "tokens_seen": 657784832
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040438314944834505,
      "loss": 2.9091,
      "theoretical_loss": 3.8034951336811527,
      "tokens_seen": 657850368
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040437311935807423,
      "loss": 3.3398,
      "theoretical_loss": 3.803456235234368,
      "tokens_seen": 657915904
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004043630892678034,
      "loss": 2.6882,
      "theoretical_loss": 3.8034173417469264,
      "tokens_seen": 657981440
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040435305917753265,
      "loss": 3.1224,
      "theoretical_loss": 3.8033784532177,
      "tokens_seen": 658046976
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004043430290872618,
      "loss": 2.9507,
      "theoretical_loss": 3.803339569645565,
      "tokens_seen": 658112512
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.000404332998996991,
      "loss": 3.2165,
      "theoretical_loss": 3.8033006910293947,
      "tokens_seen": 658178048
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040432296890672014,
      "loss": 2.744,
      "theoretical_loss": 3.8032618173680643,
      "tokens_seen": 658243584
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040431293881644937,
      "loss": 3.1914,
      "theoretical_loss": 3.803222948660449,
      "tokens_seen": 658309120
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040430290872617855,
      "loss": 2.9914,
      "theoretical_loss": 3.803184084905425,
      "tokens_seen": 658374656
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040429287863590773,
      "loss": 3.0442,
      "theoretical_loss": 3.8031452261018677,
      "tokens_seen": 658440192
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004042828485456369,
      "loss": 3.0255,
      "theoretical_loss": 3.803106372248654,
      "tokens_seen": 658505728
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004042728184553661,
      "loss": 3.0498,
      "theoretical_loss": 3.8030675233446605,
      "tokens_seen": 658571264
    },
    {
      "epoch": 0.2,
      "objective/train/advantage_avg": 0.4937957227230072,
      "objective/train/docs_used": 381585,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.9325308799743652,
      "objective/train/original_loss": 2.9325308799743652,
      "objective/train/theoretical_loss": 3.8030286793887647,
      "objective/train/tokens_used": 679096800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24391229450702667,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506194829940796,
      "objective/train/weighted_lm_loss": 3.0808846950531006,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 3.8030286793887647,
      "tokens_seen": 658636800
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004042627883650953,
      "loss": 3.163,
      "theoretical_loss": 3.8030286793887647,
      "tokens_seen": 658636800
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004042527582748245,
      "loss": 3.0067,
      "theoretical_loss": 3.8029898403798432,
      "tokens_seen": 658702336
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040424272818455364,
      "loss": 2.8798,
      "theoretical_loss": 3.802951006316775,
      "tokens_seen": 658767872
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004042326980942829,
      "loss": 3.0937,
      "theoretical_loss": 3.8029121771984373,
      "tokens_seen": 658833408
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.000404222668004012,
      "loss": 3.0434,
      "theoretical_loss": 3.8028733530237093,
      "tokens_seen": 658898944
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040421263791374124,
      "loss": 3.0142,
      "theoretical_loss": 3.8028345337914704,
      "tokens_seen": 658964480
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004042026078234704,
      "loss": 3.0591,
      "theoretical_loss": 3.8027957195005992,
      "tokens_seen": 659030016
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004041925777331996,
      "loss": 3.1099,
      "theoretical_loss": 3.8027569101499754,
      "tokens_seen": 659095552
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004041825476429288,
      "loss": 3.1014,
      "theoretical_loss": 3.8027181057384793,
      "tokens_seen": 659161088
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.000404172517552658,
      "loss": 2.9469,
      "theoretical_loss": 3.8026793062649915,
      "tokens_seen": 659226624
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040416248746238714,
      "loss": 2.9447,
      "theoretical_loss": 3.8026405117283932,
      "tokens_seen": 659292160
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004041524573721164,
      "loss": 3.1694,
      "theoretical_loss": 3.802601722127565,
      "tokens_seen": 659357696
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004041424272818455,
      "loss": 2.8377,
      "theoretical_loss": 3.8025629374613876,
      "tokens_seen": 659423232
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040413239719157474,
      "loss": 2.9566,
      "theoretical_loss": 3.8025241577287447,
      "tokens_seen": 659488768
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004041223671013039,
      "loss": 3.1717,
      "theoretical_loss": 3.8024853829285172,
      "tokens_seen": 659554304
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004041123370110331,
      "loss": 2.8841,
      "theoretical_loss": 3.8024466130595886,
      "tokens_seen": 659619840
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004041023069207623,
      "loss": 2.8657,
      "theoretical_loss": 3.8024078481208416,
      "tokens_seen": 659685376
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040409227683049147,
      "loss": 3.081,
      "theoretical_loss": 3.802369088111159,
      "tokens_seen": 659750912
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040408224674022065,
      "loss": 3.0602,
      "theoretical_loss": 3.802330333029426,
      "tokens_seen": 659816448
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004040722166499499,
      "loss": 3.1383,
      "theoretical_loss": 3.802291582874525,
      "tokens_seen": 659881984
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.000404062186559679,
      "loss": 2.9285,
      "theoretical_loss": 3.802252837645341,
      "tokens_seen": 659947520
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040405215646940824,
      "loss": 3.0986,
      "theoretical_loss": 3.8022140973407588,
      "tokens_seen": 660013056
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040404212637913737,
      "loss": 3.0216,
      "theoretical_loss": 3.802175361959664,
      "tokens_seen": 660078592
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004040320962888666,
      "loss": 3.2062,
      "theoretical_loss": 3.8021366315009417,
      "tokens_seen": 660144128
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004040220661985958,
      "loss": 2.9607,
      "theoretical_loss": 3.8020979059634783,
      "tokens_seen": 660209664
    },
    {
      "epoch": 0.2,
      "objective/train/advantage_avg": 0.4946053922176361,
      "objective/train/docs_used": 382875,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.2414348125457764,
      "objective/train/original_loss": 3.2414348125457764,
      "objective/train/theoretical_loss": 3.8020591853461596,
      "objective/train/tokens_used": 680735200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2447008639574051,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507043600082397,
      "objective/train/weighted_lm_loss": 3.4058923721313477,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0417574644088745,
      "theoretical_loss": 3.8020591853461596,
      "tokens_seen": 660275200
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040401203610832497,
      "loss": 3.1123,
      "theoretical_loss": 3.8020591853461596,
      "tokens_seen": 660275200
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040400200601805415,
      "loss": 2.946,
      "theoretical_loss": 3.8020204696478723,
      "tokens_seen": 660340736
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004039919759277834,
      "loss": 2.8726,
      "theoretical_loss": 3.8019817588675036,
      "tokens_seen": 660406272
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004039819458375125,
      "loss": 2.985,
      "theoretical_loss": 3.8019430530039404,
      "tokens_seen": 660471808
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040397191574724175,
      "loss": 3.0087,
      "theoretical_loss": 3.801904352056071,
      "tokens_seen": 660537344
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004039618856569709,
      "loss": 2.9163,
      "theoretical_loss": 3.801865656022783,
      "tokens_seen": 660602880
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004039518555667001,
      "loss": 3.1254,
      "theoretical_loss": 3.8018269649029652,
      "tokens_seen": 660668416
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004039418254764293,
      "loss": 2.9816,
      "theoretical_loss": 3.801788278695507,
      "tokens_seen": 660733952
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040393179538615847,
      "loss": 3.0612,
      "theoretical_loss": 3.8017495973992954,
      "tokens_seen": 660799488
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040392176529588765,
      "loss": 3.0356,
      "theoretical_loss": 3.801710921013222,
      "tokens_seen": 660865024
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040391173520561683,
      "loss": 2.8365,
      "theoretical_loss": 3.801672249536176,
      "tokens_seen": 660930560
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040390170511534607,
      "loss": 3.027,
      "theoretical_loss": 3.801633582967048,
      "tokens_seen": 660996096
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040389167502507525,
      "loss": 3.0734,
      "theoretical_loss": 3.8015949213047278,
      "tokens_seen": 661061632
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040388164493480443,
      "loss": 2.8974,
      "theoretical_loss": 3.8015562645481067,
      "tokens_seen": 661127168
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004038716148445336,
      "loss": 2.9467,
      "theoretical_loss": 3.8015176126960766,
      "tokens_seen": 661192704
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040386158475426285,
      "loss": 2.9534,
      "theoretical_loss": 3.8014789657475285,
      "tokens_seen": 661258240
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.000403851554663992,
      "loss": 2.8802,
      "theoretical_loss": 3.801440323701355,
      "tokens_seen": 661323776
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004038415245737212,
      "loss": 2.9029,
      "theoretical_loss": 3.801401686556448,
      "tokens_seen": 661389312
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040383149448345034,
      "loss": 2.9695,
      "theoretical_loss": 3.8013630543117003,
      "tokens_seen": 661454848
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040382146439317957,
      "loss": 3.1867,
      "theoretical_loss": 3.801324426966006,
      "tokens_seen": 661520384
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040381143430290875,
      "loss": 3.0933,
      "theoretical_loss": 3.8012858045182565,
      "tokens_seen": 661585920
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040380140421263794,
      "loss": 3.0414,
      "theoretical_loss": 3.801247186967348,
      "tokens_seen": 661651456
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004037913741223671,
      "loss": 3.1277,
      "theoretical_loss": 3.8012085743121733,
      "tokens_seen": 661716992
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004037813440320963,
      "loss": 3.1822,
      "theoretical_loss": 3.801169966551627,
      "tokens_seen": 661782528
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004037713139418255,
      "loss": 2.8984,
      "theoretical_loss": 3.801131363684605,
      "tokens_seen": 661848064
    },
    {
      "epoch": 0.2,
      "objective/train/advantage_avg": 0.4948192536830902,
      "objective/train/docs_used": 383560,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.8632445335388184,
      "objective/train/original_loss": 2.8632445335388184,
      "objective/train/theoretical_loss": 3.8010927657100013,
      "objective/train/tokens_used": 682373600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24492135643959045,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507270097732544,
      "objective/train/weighted_lm_loss": 3.0084314346313477,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0304546356201172,
      "theoretical_loss": 3.8010927657100013,
      "tokens_seen": 661913600
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004037612838515547,
      "loss": 2.8754,
      "theoretical_loss": 3.8010927657100013,
      "tokens_seen": 661913600
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040375125376128384,
      "loss": 2.9871,
      "theoretical_loss": 3.801054172626712,
      "tokens_seen": 661979136
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004037412236710131,
      "loss": 3.3264,
      "theoretical_loss": 3.8010155844336344,
      "tokens_seen": 662044672
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004037311935807422,
      "loss": 2.9979,
      "theoretical_loss": 3.8009770011296626,
      "tokens_seen": 662110208
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040372116349047144,
      "loss": 2.8163,
      "theoretical_loss": 3.8009384227136946,
      "tokens_seen": 662175744
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004037111334002006,
      "loss": 2.8757,
      "theoretical_loss": 3.8008998491846278,
      "tokens_seen": 662241280
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004037011033099298,
      "loss": 3.0545,
      "theoretical_loss": 3.8008612805413593,
      "tokens_seen": 662306816
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.000403691073219659,
      "loss": 2.8795,
      "theoretical_loss": 3.800822716782786,
      "tokens_seen": 662372352
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004036810431293882,
      "loss": 2.9356,
      "theoretical_loss": 3.800784157907807,
      "tokens_seen": 662437888
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040367101303911734,
      "loss": 2.8041,
      "theoretical_loss": 3.800745603915321,
      "tokens_seen": 662503424
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004036609829488466,
      "loss": 2.9462,
      "theoretical_loss": 3.800707054804226,
      "tokens_seen": 662568960
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004036509528585757,
      "loss": 3.408,
      "theoretical_loss": 3.800668510573422,
      "tokens_seen": 662634496
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040364092276830494,
      "loss": 3.1141,
      "theoretical_loss": 3.8006299712218086,
      "tokens_seen": 662700032
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004036308926780341,
      "loss": 2.8129,
      "theoretical_loss": 3.800591436748285,
      "tokens_seen": 662765568
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004036208625877633,
      "loss": 3.119,
      "theoretical_loss": 3.8005529071517516,
      "tokens_seen": 662831104
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004036108324974925,
      "loss": 2.7775,
      "theoretical_loss": 3.8005143824311096,
      "tokens_seen": 662896640
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040360080240722167,
      "loss": 2.9591,
      "theoretical_loss": 3.8004758625852597,
      "tokens_seen": 662962176
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040359077231695085,
      "loss": 2.9146,
      "theoretical_loss": 3.8004373476131037,
      "tokens_seen": 663027712
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004035807422266801,
      "loss": 2.9326,
      "theoretical_loss": 3.8003988375135425,
      "tokens_seen": 663093248
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004035707121364092,
      "loss": 3.1618,
      "theoretical_loss": 3.8003603322854786,
      "tokens_seen": 663158784
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040356068204613844,
      "loss": 2.7856,
      "theoretical_loss": 3.800321831927815,
      "tokens_seen": 663224320
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040355065195586757,
      "loss": 3.1655,
      "theoretical_loss": 3.8002833364394535,
      "tokens_seen": 663289856
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004035406218655968,
      "loss": 2.6962,
      "theoretical_loss": 3.8002448458192974,
      "tokens_seen": 663355392
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.000403530591775326,
      "loss": 3.0421,
      "theoretical_loss": 3.800206360066251,
      "tokens_seen": 663420928
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040352056168505517,
      "loss": 2.6329,
      "theoretical_loss": 3.8001678791792175,
      "tokens_seen": 663486464
    },
    {
      "epoch": 0.2,
      "objective/train/advantage_avg": 0.49639981985092163,
      "objective/train/docs_used": 384103,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.4779796600341797,
      "objective/train/original_loss": 2.4779791831970215,
      "objective/train/theoretical_loss": 3.8001294031571007,
      "objective/train/tokens_used": 684012000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24644888937473297,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0508928298950195,
      "objective/train/weighted_lm_loss": 2.603922128677368,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 3.8001294031571007,
      "tokens_seen": 663552000
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040351053159478435,
      "loss": 2.7374,
      "theoretical_loss": 3.8001294031571007,
      "tokens_seen": 663552000
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004035005015045136,
      "loss": 3.0577,
      "theoretical_loss": 3.800090931998806,
      "tokens_seen": 663617536
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004034904714142427,
      "loss": 3.0575,
      "theoretical_loss": 3.800052465703238,
      "tokens_seen": 663683072
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040348044132397195,
      "loss": 3.0442,
      "theoretical_loss": 3.8000140042693022,
      "tokens_seen": 663748608
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004034704112337011,
      "loss": 2.836,
      "theoretical_loss": 3.7999755476959036,
      "tokens_seen": 663814144
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004034603811434303,
      "loss": 2.8807,
      "theoretical_loss": 3.7999370959819494,
      "tokens_seen": 663879680
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004034503510531595,
      "loss": 2.8949,
      "theoretical_loss": 3.799898649126344,
      "tokens_seen": 663945216
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040344032096288867,
      "loss": 3.0699,
      "theoretical_loss": 3.799860207127996,
      "tokens_seen": 664010752
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040343029087261785,
      "loss": 2.8431,
      "theoretical_loss": 3.7998217699858112,
      "tokens_seen": 664076288
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040342026078234703,
      "loss": 2.985,
      "theoretical_loss": 3.799783337698697,
      "tokens_seen": 664141824
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004034102306920762,
      "loss": 3.1604,
      "theoretical_loss": 3.799744910265563,
      "tokens_seen": 664207360
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040340020060180545,
      "loss": 2.7191,
      "theoretical_loss": 3.799706487685315,
      "tokens_seen": 664272896
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004033901705115346,
      "loss": 3.0714,
      "theoretical_loss": 3.799668069956862,
      "tokens_seen": 664338432
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004033801404212638,
      "loss": 2.8224,
      "theoretical_loss": 3.7996296570791133,
      "tokens_seen": 664403968
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.000403370110330993,
      "loss": 3.0953,
      "theoretical_loss": 3.7995912490509784,
      "tokens_seen": 664469504
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004033600802407222,
      "loss": 3.1904,
      "theoretical_loss": 3.799552845871366,
      "tokens_seen": 664535040
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040335005015045136,
      "loss": 2.9258,
      "theoretical_loss": 3.7995144475391864,
      "tokens_seen": 664600576
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040334002006018054,
      "loss": 2.7903,
      "theoretical_loss": 3.79947605405335,
      "tokens_seen": 664666112
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004033299899699097,
      "loss": 3.0581,
      "theoretical_loss": 3.799437665412767,
      "tokens_seen": 664731648
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040331995987963895,
      "loss": 2.8679,
      "theoretical_loss": 3.799399281616348,
      "tokens_seen": 664797184
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004033099297893681,
      "loss": 2.8,
      "theoretical_loss": 3.799360902663005,
      "tokens_seen": 664862720
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004032998996990973,
      "loss": 3.0536,
      "theoretical_loss": 3.7993225285516496,
      "tokens_seen": 664928256
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040328986960882644,
      "loss": 3.0081,
      "theoretical_loss": 3.799284159281193,
      "tokens_seen": 664993792
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004032798395185557,
      "loss": 3.1578,
      "theoretical_loss": 3.799245794850548,
      "tokens_seen": 665059328
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040326980942828486,
      "loss": 2.8766,
      "theoretical_loss": 3.799207435258628,
      "tokens_seen": 665124864
    },
    {
      "epoch": 0.2,
      "objective/train/advantage_avg": 0.4897213280200958,
      "objective/train/docs_used": 385247,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.056739568710327,
      "objective/train/original_loss": 3.056739568710327,
      "objective/train/theoretical_loss": 3.7991690805043445,
      "objective/train/tokens_used": 685650400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2399553656578064,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0501917600631714,
      "objective/train/weighted_lm_loss": 3.2104380130767822,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0458475351333618,
      "theoretical_loss": 3.7991690805043445,
      "tokens_seen": 665190400
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040325977933801404,
      "loss": 3.1806,
      "theoretical_loss": 3.7991690805043445,
      "tokens_seen": 665190400
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004032497492477432,
      "loss": 3.0699,
      "theoretical_loss": 3.7991307305866124,
      "tokens_seen": 665255936
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004032397191574724,
      "loss": 2.9636,
      "theoretical_loss": 3.799092385504345,
      "tokens_seen": 665321472
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004032296890672016,
      "loss": 3.0132,
      "theoretical_loss": 3.7990540452564554,
      "tokens_seen": 665387008
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004032196589769308,
      "loss": 3.057,
      "theoretical_loss": 3.799015709841859,
      "tokens_seen": 665452544
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040320962888665995,
      "loss": 3.1139,
      "theoretical_loss": 3.7989773792594708,
      "tokens_seen": 665518080
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004031995987963892,
      "loss": 2.887,
      "theoretical_loss": 3.7989390535082057,
      "tokens_seen": 665583616
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040318956870611836,
      "loss": 2.7698,
      "theoretical_loss": 3.7989007325869784,
      "tokens_seen": 665649152
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040317953861584754,
      "loss": 3.1462,
      "theoretical_loss": 3.7988624164947056,
      "tokens_seen": 665714688
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004031695085255767,
      "loss": 2.8112,
      "theoretical_loss": 3.7988241052303033,
      "tokens_seen": 665780224
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004031594784353059,
      "loss": 3.1859,
      "theoretical_loss": 3.798785798792688,
      "tokens_seen": 665845760
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040314944834503514,
      "loss": 2.9564,
      "theoretical_loss": 3.798747497180777,
      "tokens_seen": 665911296
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004031394182547643,
      "loss": 2.9924,
      "theoretical_loss": 3.7987092003934864,
      "tokens_seen": 665976832
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004031293881644935,
      "loss": 2.85,
      "theoretical_loss": 3.7986709084297354,
      "tokens_seen": 666042368
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004031193580742227,
      "loss": 3.0063,
      "theoretical_loss": 3.7986326212884407,
      "tokens_seen": 666107904
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040310932798395187,
      "loss": 2.9339,
      "theoretical_loss": 3.7985943389685213,
      "tokens_seen": 666173440
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040309929789368105,
      "loss": 2.8512,
      "theoretical_loss": 3.7985560614688945,
      "tokens_seen": 666238976
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004030892678034103,
      "loss": 2.7489,
      "theoretical_loss": 3.798517788788481,
      "tokens_seen": 666304512
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004030792377131394,
      "loss": 3.2419,
      "theoretical_loss": 3.7984795209261994,
      "tokens_seen": 666370048
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040306920762286864,
      "loss": 3.0189,
      "theoretical_loss": 3.79844125788097,
      "tokens_seen": 666435584
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040305917753259777,
      "loss": 3.1061,
      "theoretical_loss": 3.7984029996517115,
      "tokens_seen": 666501120
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.000403049147442327,
      "loss": 2.7693,
      "theoretical_loss": 3.798364746237345,
      "tokens_seen": 666566656
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004030391173520562,
      "loss": 3.0,
      "theoretical_loss": 3.7983264976367916,
      "tokens_seen": 666632192
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040302908726178537,
      "loss": 2.9546,
      "theoretical_loss": 3.798288253848972,
      "tokens_seen": 666697728
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040301905717151455,
      "loss": 3.0049,
      "theoretical_loss": 3.7982500148728073,
      "tokens_seen": 666763264
    },
    {
      "epoch": 0.2,
      "objective/train/advantage_avg": 0.49440762400627136,
      "objective/train/docs_used": 385855,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.845923662185669,
      "objective/train/original_loss": 2.845924139022827,
      "objective/train/theoretical_loss": 3.79821178070722,
      "objective/train/tokens_used": 687288800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24448974430561066,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050683617591858,
      "objective/train/weighted_lm_loss": 2.9901020526885986,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0432155132293701,
      "theoretical_loss": 3.79821178070722,
      "tokens_seen": 666828800
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004030090270812438,
      "loss": 3.147,
      "theoretical_loss": 3.79821178070722,
      "tokens_seen": 666828800
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004029989969909729,
      "loss": 2.9542,
      "theoretical_loss": 3.798173551351132,
      "tokens_seen": 666894336
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040298896690070215,
      "loss": 3.1726,
      "theoretical_loss": 3.7981353268034654,
      "tokens_seen": 666959872
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004029789368104313,
      "loss": 2.9534,
      "theoretical_loss": 3.7980971070631435,
      "tokens_seen": 667025408
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004029689067201605,
      "loss": 2.9719,
      "theoretical_loss": 3.7980588921290885,
      "tokens_seen": 667090944
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004029588766298897,
      "loss": 3.2675,
      "theoretical_loss": 3.7980206820002254,
      "tokens_seen": 667156480
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040294884653961887,
      "loss": 2.9364,
      "theoretical_loss": 3.797982476675477,
      "tokens_seen": 667222016
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040293881644934805,
      "loss": 3.2078,
      "theoretical_loss": 3.7979442761537676,
      "tokens_seen": 667287552
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040292878635907723,
      "loss": 3.1735,
      "theoretical_loss": 3.7979060804340223,
      "tokens_seen": 667353088
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004029187562688064,
      "loss": 3.1162,
      "theoretical_loss": 3.797867889515165,
      "tokens_seen": 667418624
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040290872617853565,
      "loss": 2.9004,
      "theoretical_loss": 3.797829703396122,
      "tokens_seen": 667484160
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004028986960882648,
      "loss": 3.2785,
      "theoretical_loss": 3.7977915220758183,
      "tokens_seen": 667549696
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.000402888665997994,
      "loss": 3.0448,
      "theoretical_loss": 3.79775334555318,
      "tokens_seen": 667615232
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004028786359077232,
      "loss": 2.8503,
      "theoretical_loss": 3.7977151738271333,
      "tokens_seen": 667680768
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004028686058174524,
      "loss": 3.1814,
      "theoretical_loss": 3.7976770068966053,
      "tokens_seen": 667746304
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040285857572718156,
      "loss": 2.9457,
      "theoretical_loss": 3.797638844760522,
      "tokens_seen": 667811840
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040284854563691074,
      "loss": 2.7752,
      "theoretical_loss": 3.7976006874178116,
      "tokens_seen": 667877376
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004028385155466399,
      "loss": 2.8435,
      "theoretical_loss": 3.797562534867401,
      "tokens_seen": 667942912
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040282848545636915,
      "loss": 2.9226,
      "theoretical_loss": 3.7975243871082194,
      "tokens_seen": 668008448
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004028184553660983,
      "loss": 3.084,
      "theoretical_loss": 3.797486244139194,
      "tokens_seen": 668073984
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004028084252758275,
      "loss": 2.8214,
      "theoretical_loss": 3.7974481059592535,
      "tokens_seen": 668139520
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040279839518555664,
      "loss": 2.965,
      "theoretical_loss": 3.7974099725673276,
      "tokens_seen": 668205056
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004027883650952859,
      "loss": 2.782,
      "theoretical_loss": 3.7973718439623454,
      "tokens_seen": 668270592
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040277833500501506,
      "loss": 3.037,
      "theoretical_loss": 3.7973337201432367,
      "tokens_seen": 668336128
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040276830491474424,
      "loss": 3.0237,
      "theoretical_loss": 3.7972956011089316,
      "tokens_seen": 668401664
    },
    {
      "epoch": 0.2,
      "objective/train/advantage_avg": 0.49422749876976013,
      "objective/train/docs_used": 387212,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.062035083770752,
      "objective/train/original_loss": 3.062035322189331,
      "objective/train/theoretical_loss": 3.797257486858361,
      "objective/train/tokens_used": 688927200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2443234920501709,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506647825241089,
      "objective/train/weighted_lm_loss": 3.2171730995178223,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0304546356201172,
      "theoretical_loss": 3.797257486858361,
      "tokens_seen": 668467200
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004027582748244734,
      "loss": 3.1354,
      "theoretical_loss": 3.797257486858361,
      "tokens_seen": 668467200
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004027482447342026,
      "loss": 2.8529,
      "theoretical_loss": 3.797219377390454,
      "tokens_seen": 668532736
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004027382146439318,
      "loss": 2.8196,
      "theoretical_loss": 3.797181272704143,
      "tokens_seen": 668598272
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.000402728184553661,
      "loss": 2.8875,
      "theoretical_loss": 3.79714317279836,
      "tokens_seen": 668663808
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040271815446339015,
      "loss": 2.8678,
      "theoretical_loss": 3.797105077672035,
      "tokens_seen": 668729344
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004027081243731194,
      "loss": 2.9458,
      "theoretical_loss": 3.797066987324102,
      "tokens_seen": 668794880
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040269809428284856,
      "loss": 2.9354,
      "theoretical_loss": 3.7970289017534924,
      "tokens_seen": 668860416
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040268806419257774,
      "loss": 2.98,
      "theoretical_loss": 3.796990820959139,
      "tokens_seen": 668925952
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004026780341023069,
      "loss": 2.8653,
      "theoretical_loss": 3.796952744939976,
      "tokens_seen": 668991488
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004026680040120361,
      "loss": 2.9133,
      "theoretical_loss": 3.7969146736949355,
      "tokens_seen": 669057024
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004026579739217653,
      "loss": 2.9588,
      "theoretical_loss": 3.796876607222952,
      "tokens_seen": 669122560
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004026479438314945,
      "loss": 2.9819,
      "theoretical_loss": 3.7968385455229603,
      "tokens_seen": 669188096
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040263791374122365,
      "loss": 2.895,
      "theoretical_loss": 3.7968004885938935,
      "tokens_seen": 669253632
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004026278836509529,
      "loss": 2.9575,
      "theoretical_loss": 3.7967624364346877,
      "tokens_seen": 669319168
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.000402617853560682,
      "loss": 3.0432,
      "theoretical_loss": 3.796724389044278,
      "tokens_seen": 669384704
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040260782347041125,
      "loss": 3.012,
      "theoretical_loss": 3.796686346421599,
      "tokens_seen": 669450240
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040259779338014043,
      "loss": 2.6819,
      "theoretical_loss": 3.7966483085655884,
      "tokens_seen": 669515776
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004025877632898696,
      "loss": 2.9816,
      "theoretical_loss": 3.7966102754751803,
      "tokens_seen": 669581312
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004025777331995988,
      "loss": 2.759,
      "theoretical_loss": 3.796572247149313,
      "tokens_seen": 669646848
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040256770310932797,
      "loss": 2.9972,
      "theoretical_loss": 3.7965342235869226,
      "tokens_seen": 669712384
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040255767301905715,
      "loss": 3.1019,
      "theoretical_loss": 3.796496204786946,
      "tokens_seen": 669777920
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004025476429287864,
      "loss": 3.0526,
      "theoretical_loss": 3.796458190748322,
      "tokens_seen": 669843456
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004025376128385155,
      "loss": 2.8818,
      "theoretical_loss": 3.7964201814699874,
      "tokens_seen": 669908992
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040252758274824475,
      "loss": 3.02,
      "theoretical_loss": 3.796382176950881,
      "tokens_seen": 669974528
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040251755265797393,
      "loss": 3.0328,
      "theoretical_loss": 3.7963441771899418,
      "tokens_seen": 670040064
    },
    {
      "epoch": 0.2,
      "objective/train/advantage_avg": 0.49538204073905945,
      "objective/train/docs_used": 387914,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.532646656036377,
      "objective/train/original_loss": 2.532646894454956,
      "objective/train/theoretical_loss": 3.7963061821861084,
      "objective/train/tokens_used": 690565600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24543695151805878,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050785779953003,
      "objective/train/weighted_lm_loss": 2.661288261413574,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0437887907028198,
      "theoretical_loss": 3.7963061821861084,
      "tokens_seen": 670105600
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004025075225677031,
      "loss": 2.9229,
      "theoretical_loss": 3.7963061821861084,
      "tokens_seen": 670105600
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004024974924774323,
      "loss": 2.9767,
      "theoretical_loss": 3.79626819193832,
      "tokens_seen": 670171136
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004024874623871615,
      "loss": 2.8486,
      "theoretical_loss": 3.7962302064455162,
      "tokens_seen": 670236672
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040247743229689066,
      "loss": 3.0563,
      "theoretical_loss": 3.796192225706637,
      "tokens_seen": 670302208
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004024674022066199,
      "loss": 3.0064,
      "theoretical_loss": 3.796154249720623,
      "tokens_seen": 670367744
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.000402457372116349,
      "loss": 3.0634,
      "theoretical_loss": 3.796116278486415,
      "tokens_seen": 670433280
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040244734202607825,
      "loss": 2.7828,
      "theoretical_loss": 3.7960783120029533,
      "tokens_seen": 670498816
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004024373119358074,
      "loss": 3.0822,
      "theoretical_loss": 3.79604035026918,
      "tokens_seen": 670564352
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004024272818455366,
      "loss": 2.8498,
      "theoretical_loss": 3.7960023932840365,
      "tokens_seen": 670629888
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004024172517552658,
      "loss": 2.9932,
      "theoretical_loss": 3.795964441046465,
      "tokens_seen": 670695424
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.000402407221664995,
      "loss": 3.1904,
      "theoretical_loss": 3.795926493555407,
      "tokens_seen": 670760960
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004023971915747242,
      "loss": 2.8257,
      "theoretical_loss": 3.7958885508098064,
      "tokens_seen": 670826496
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004023871614844534,
      "loss": 2.8256,
      "theoretical_loss": 3.795850612808606,
      "tokens_seen": 670892032
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004023771313941826,
      "loss": 3.0131,
      "theoretical_loss": 3.7958126795507487,
      "tokens_seen": 670957568
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040236710130391176,
      "loss": 2.9732,
      "theoretical_loss": 3.7957747510351787,
      "tokens_seen": 671023104
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040235707121364094,
      "loss": 2.9012,
      "theoretical_loss": 3.795736827260839,
      "tokens_seen": 671088640
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004023470411233701,
      "loss": 3.0368,
      "theoretical_loss": 3.795698908226676,
      "tokens_seen": 671154176
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040233701103309935,
      "loss": 3.0129,
      "theoretical_loss": 3.795660993931633,
      "tokens_seen": 671219712
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004023269809428285,
      "loss": 3.3263,
      "theoretical_loss": 3.795623084374655,
      "tokens_seen": 671285248
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004023169508525577,
      "loss": 2.8717,
      "theoretical_loss": 3.7955851795546875,
      "tokens_seen": 671350784
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040230692076228684,
      "loss": 2.9589,
      "theoretical_loss": 3.795547279470677,
      "tokens_seen": 671416320
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004022968906720161,
      "loss": 3.1601,
      "theoretical_loss": 3.7955093841215692,
      "tokens_seen": 671481856
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040228686058174526,
      "loss": 2.983,
      "theoretical_loss": 3.79547149350631,
      "tokens_seen": 671547392
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040227683049147444,
      "loss": 3.1429,
      "theoretical_loss": 3.7954336076238473,
      "tokens_seen": 671612928
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004022668004012036,
      "loss": 3.0707,
      "theoretical_loss": 3.795395726473127,
      "tokens_seen": 671678464
    },
    {
      "epoch": 0.2,
      "objective/train/advantage_avg": 0.4928900897502899,
      "objective/train/docs_used": 389368,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.8706603050231934,
      "objective/train/original_loss": 2.8706605434417725,
      "objective/train/theoretical_loss": 3.795357850053097,
      "objective/train/tokens_used": 692204000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24302725493907928,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505244731903076,
      "objective/train/weighted_lm_loss": 3.0156095027923584,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0432155132293701,
      "theoretical_loss": 3.795357850053097,
      "tokens_seen": 671744000
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004022567703109328,
      "loss": 3.083,
      "theoretical_loss": 3.795357850053097,
      "tokens_seen": 671744000
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.000402246740220662,
      "loss": 3.1099,
      "theoretical_loss": 3.795319978362705,
      "tokens_seen": 671809536
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004022367101303912,
      "loss": 2.9803,
      "theoretical_loss": 3.7952821114008994,
      "tokens_seen": 671875072
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040222668004012035,
      "loss": 3.0478,
      "theoretical_loss": 3.795244249166629,
      "tokens_seen": 671940608
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004022166499498496,
      "loss": 2.8416,
      "theoretical_loss": 3.7952063916588417,
      "tokens_seen": 672006144
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040220661985957876,
      "loss": 2.9603,
      "theoretical_loss": 3.7951685388764864,
      "tokens_seen": 672071680
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040219658976930794,
      "loss": 2.8522,
      "theoretical_loss": 3.795130690818514,
      "tokens_seen": 672137216
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004021865596790371,
      "loss": 2.9081,
      "theoretical_loss": 3.7950928474838728,
      "tokens_seen": 672202752
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004021765295887663,
      "loss": 3.3969,
      "theoretical_loss": 3.795055008871514,
      "tokens_seen": 672268288
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004021664994984955,
      "loss": 3.1999,
      "theoretical_loss": 3.7950171749803876,
      "tokens_seen": 672333824
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004021564694082247,
      "loss": 3.1504,
      "theoretical_loss": 3.794979345809444,
      "tokens_seen": 672399360
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040214643931795385,
      "loss": 2.9366,
      "theoretical_loss": 3.794941521357635,
      "tokens_seen": 672464896
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004021364092276831,
      "loss": 3.019,
      "theoretical_loss": 3.794903701623912,
      "tokens_seen": 672530432
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004021263791374122,
      "loss": 3.0458,
      "theoretical_loss": 3.7948658866072265,
      "tokens_seen": 672595968
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040211634904714145,
      "loss": 2.9715,
      "theoretical_loss": 3.7948280763065307,
      "tokens_seen": 672661504
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040210631895687063,
      "loss": 3.1295,
      "theoretical_loss": 3.7947902707207777,
      "tokens_seen": 672727040
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004020962888665998,
      "loss": 3.0147,
      "theoretical_loss": 3.7947524698489197,
      "tokens_seen": 672792576
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.000402086258776329,
      "loss": 3.012,
      "theoretical_loss": 3.794714673689909,
      "tokens_seen": 672858112
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040207622868605817,
      "loss": 3.0269,
      "theoretical_loss": 3.794676882242701,
      "tokens_seen": 672923648
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040206619859578735,
      "loss": 3.0219,
      "theoretical_loss": 3.7946390955062483,
      "tokens_seen": 672989184
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004020561685055166,
      "loss": 3.1498,
      "theoretical_loss": 3.7946013134795047,
      "tokens_seen": 673054720
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004020461384152457,
      "loss": 2.8766,
      "theoretical_loss": 3.7945635361614256,
      "tokens_seen": 673120256
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040203610832497495,
      "loss": 3.0332,
      "theoretical_loss": 3.7945257635509657,
      "tokens_seen": 673185792
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040202607823470413,
      "loss": 3.1702,
      "theoretical_loss": 3.7944879956470796,
      "tokens_seen": 673251328
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004020160481444333,
      "loss": 3.1295,
      "theoretical_loss": 3.7944502324487233,
      "tokens_seen": 673316864
    },
    {
      "epoch": 0.2,
      "objective/train/advantage_avg": 0.4926285743713379,
      "objective/train/docs_used": 390091,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.5068840980529785,
      "objective/train/original_loss": 2.5068838596343994,
      "objective/train/theoretical_loss": 3.7944124739548526,
      "objective/train/tokens_used": 693842400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24278320372104645,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0504969358444214,
      "objective/train/weighted_lm_loss": 2.6333980560302734,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0425469875335693,
      "theoretical_loss": 3.7944124739548526,
      "tokens_seen": 673382400
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004020060180541625,
      "loss": 2.8604,
      "theoretical_loss": 3.7944124739548526,
      "tokens_seen": 673382400
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004019959879638917,
      "loss": 3.335,
      "theoretical_loss": 3.7943747201644227,
      "tokens_seen": 673447936
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040198595787362086,
      "loss": 2.8624,
      "theoretical_loss": 3.7943369710763912,
      "tokens_seen": 673513472
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004019759277833501,
      "loss": 2.9486,
      "theoretical_loss": 3.7942992266897146,
      "tokens_seen": 673579008
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004019658976930792,
      "loss": 2.7986,
      "theoretical_loss": 3.79426148700335,
      "tokens_seen": 673644544
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040195586760280845,
      "loss": 2.8639,
      "theoretical_loss": 3.794223752016255,
      "tokens_seen": 673710080
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004019458375125376,
      "loss": 2.9926,
      "theoretical_loss": 3.7941860217273877,
      "tokens_seen": 673775616
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004019358074222668,
      "loss": 3.0389,
      "theoretical_loss": 3.794148296135705,
      "tokens_seen": 673841152
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.000401925777331996,
      "loss": 3.0205,
      "theoretical_loss": 3.794110575240167,
      "tokens_seen": 673906688
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004019157472417252,
      "loss": 3.0082,
      "theoretical_loss": 3.7940728590397312,
      "tokens_seen": 673972224
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040190571715145436,
      "loss": 2.8507,
      "theoretical_loss": 3.7940351475333576,
      "tokens_seen": 674037760
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004018956870611836,
      "loss": 2.8986,
      "theoretical_loss": 3.7939974407200054,
      "tokens_seen": 674103296
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004018856569709127,
      "loss": 2.9629,
      "theoretical_loss": 3.793959738598634,
      "tokens_seen": 674168832
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040187562688064196,
      "loss": 2.9284,
      "theoretical_loss": 3.793922041168204,
      "tokens_seen": 674234368
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004018655967903711,
      "loss": 3.1698,
      "theoretical_loss": 3.793884348427676,
      "tokens_seen": 674299904
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004018555667001003,
      "loss": 2.9068,
      "theoretical_loss": 3.79384666037601,
      "tokens_seen": 674365440
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004018455366098295,
      "loss": 2.8789,
      "theoretical_loss": 3.7938089770121683,
      "tokens_seen": 674430976
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004018355065195587,
      "loss": 3.0876,
      "theoretical_loss": 3.7937712983351117,
      "tokens_seen": 674496512
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040182547642928786,
      "loss": 3.1096,
      "theoretical_loss": 3.793733624343802,
      "tokens_seen": 674562048
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040181544633901704,
      "loss": 2.6896,
      "theoretical_loss": 3.793695955037202,
      "tokens_seen": 674627584
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004018054162487462,
      "loss": 2.9682,
      "theoretical_loss": 3.793658290414273,
      "tokens_seen": 674693120
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040179538615847546,
      "loss": 3.077,
      "theoretical_loss": 3.7936206304739777,
      "tokens_seen": 674758656
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004017853560682046,
      "loss": 2.9746,
      "theoretical_loss": 3.7935829752152808,
      "tokens_seen": 674824192
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004017753259779338,
      "loss": 3.2188,
      "theoretical_loss": 3.7935453246371442,
      "tokens_seen": 674889728
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040176529588766295,
      "loss": 2.9059,
      "theoretical_loss": 3.7935076787385325,
      "tokens_seen": 674955264
    },
    {
      "epoch": 0.2,
      "objective/train/advantage_avg": 0.4919978380203247,
      "objective/train/docs_used": 391518,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.851879358291626,
      "objective/train/original_loss": 2.8518788814544678,
      "objective/train/theoretical_loss": 3.79347003751841,
      "objective/train/tokens_used": 695480800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24219174683094025,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0504308938980103,
      "objective/train/weighted_lm_loss": 2.995725393295288,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0253151655197144,
      "theoretical_loss": 3.79347003751841,
      "tokens_seen": 675020800
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004017552657973922,
      "loss": 2.8773,
      "theoretical_loss": 3.79347003751841,
      "tokens_seen": 675020800
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040174523570712137,
      "loss": 2.8672,
      "theoretical_loss": 3.79343240097574,
      "tokens_seen": 675086336
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040173520561685055,
      "loss": 2.7811,
      "theoretical_loss": 3.7933947691094883,
      "tokens_seen": 675151872
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040172517552657973,
      "loss": 3.0848,
      "theoretical_loss": 3.79335714191862,
      "tokens_seen": 675217408
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040171514543630896,
      "loss": 2.996,
      "theoretical_loss": 3.7933195194020994,
      "tokens_seen": 675282944
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004017051153460381,
      "loss": 2.7888,
      "theoretical_loss": 3.793281901558894,
      "tokens_seen": 675348480
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004016950852557673,
      "loss": 2.8099,
      "theoretical_loss": 3.7932442883879687,
      "tokens_seen": 675414016
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040168505516549645,
      "loss": 2.9516,
      "theoretical_loss": 3.79320667988829,
      "tokens_seen": 675479552
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004016750250752257,
      "loss": 3.147,
      "theoretical_loss": 3.793169076058825,
      "tokens_seen": 675545088
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040166499498495487,
      "loss": 2.9772,
      "theoretical_loss": 3.79313147689854,
      "tokens_seen": 675610624
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040165496489468405,
      "loss": 3.2039,
      "theoretical_loss": 3.7930938824064038,
      "tokens_seen": 675676160
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004016449348044133,
      "loss": 2.8317,
      "theoretical_loss": 3.793056292581383,
      "tokens_seen": 675741696
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004016349047141424,
      "loss": 2.9427,
      "theoretical_loss": 3.793018707422446,
      "tokens_seen": 675807232
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040162487462387165,
      "loss": 3.0635,
      "theoretical_loss": 3.792981126928561,
      "tokens_seen": 675872768
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040161484453360083,
      "loss": 2.8706,
      "theoretical_loss": 3.792943551098697,
      "tokens_seen": 675938304
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040160481444333,
      "loss": 2.8957,
      "theoretical_loss": 3.792905979931823,
      "tokens_seen": 676003840
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004015947843530592,
      "loss": 3.0462,
      "theoretical_loss": 3.792868413426908,
      "tokens_seen": 676069376
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040158475426278837,
      "loss": 2.7162,
      "theoretical_loss": 3.7928308515829223,
      "tokens_seen": 676134912
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040157472417251755,
      "loss": 3.1027,
      "theoretical_loss": 3.7927932943988356,
      "tokens_seen": 676200448
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004015646940822468,
      "loss": 3.1602,
      "theoretical_loss": 3.792755741873618,
      "tokens_seen": 676265984
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.0004015546639919759,
      "loss": 2.9283,
      "theoretical_loss": 3.7927181940062407,
      "tokens_seen": 676331520
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040154463390170515,
      "loss": 2.9213,
      "theoretical_loss": 3.792680650795674,
      "tokens_seen": 676397056
    },
    {
      "epoch": 0.2,
      "learning_rate": 0.00040153460381143433,
      "loss": 3.0153,
      "theoretical_loss": 3.7926431122408903,
      "tokens_seen": 676462592
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0004015245737211635,
      "loss": 2.8272,
      "theoretical_loss": 3.79260557834086,
      "tokens_seen": 676528128
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0004015145436308927,
      "loss": 2.7679,
      "theoretical_loss": 3.792568049094556,
      "tokens_seen": 676593664
    },
    {
      "epoch": 0.21,
      "objective/train/advantage_avg": 0.4929613173007965,
      "objective/train/docs_used": 392138,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.0608880519866943,
      "objective/train/original_loss": 3.0608882904052734,
      "objective/train/theoretical_loss": 3.7925305245009504,
      "objective/train/tokens_used": 697119200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24308054149150848,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505316257476807,
      "objective/train/weighted_lm_loss": 3.2156009674072266,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0363596677780151,
      "theoretical_loss": 3.7925305245009504,
      "tokens_seen": 676659200
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0004015045135406219,
      "loss": 3.1644,
      "theoretical_loss": 3.7925305245009504,
      "tokens_seen": 676659200
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00040149448345035106,
      "loss": 3.0036,
      "theoretical_loss": 3.7924930045590157,
      "tokens_seen": 676724736
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0004014844533600803,
      "loss": 3.0595,
      "theoretical_loss": 3.7924554892677254,
      "tokens_seen": 676790272
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0004014744232698094,
      "loss": 2.9651,
      "theoretical_loss": 3.7924179786260517,
      "tokens_seen": 676855808
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00040146439317953865,
      "loss": 2.9993,
      "theoretical_loss": 3.792380472632969,
      "tokens_seen": 676921344
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0004014543630892678,
      "loss": 2.9394,
      "theoretical_loss": 3.792342971287451,
      "tokens_seen": 676986880
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.000401444332998997,
      "loss": 2.6211,
      "theoretical_loss": 3.7923054745884723,
      "tokens_seen": 677052416
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0004014343029087262,
      "loss": 2.918,
      "theoretical_loss": 3.7922679825350074,
      "tokens_seen": 677117952
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0004014242728184554,
      "loss": 3.2726,
      "theoretical_loss": 3.792230495126031,
      "tokens_seen": 677183488
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00040141424272818456,
      "loss": 3.0832,
      "theoretical_loss": 3.792193012360518,
      "tokens_seen": 677249024
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0004014042126379138,
      "loss": 2.9434,
      "theoretical_loss": 3.7921555342374447,
      "tokens_seen": 677314560
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0004013941825476429,
      "loss": 3.0269,
      "theoretical_loss": 3.792118060755787,
      "tokens_seen": 677380096
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00040138415245737216,
      "loss": 3.1257,
      "theoretical_loss": 3.7920805919145204,
      "tokens_seen": 677445632
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0004013741223671013,
      "loss": 3.0963,
      "theoretical_loss": 3.792043127712622,
      "tokens_seen": 677511168
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0004013640922768305,
      "loss": 2.9537,
      "theoretical_loss": 3.792005668149068,
      "tokens_seen": 677576704
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0004013540621865597,
      "loss": 3.1455,
      "theoretical_loss": 3.791968213222837,
      "tokens_seen": 677642240
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0004013440320962889,
      "loss": 3.2466,
      "theoretical_loss": 3.791930762932905,
      "tokens_seen": 677707776
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00040133400200601806,
      "loss": 3.107,
      "theoretical_loss": 3.791893317278251,
      "tokens_seen": 677773312
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00040132397191574724,
      "loss": 2.9451,
      "theoretical_loss": 3.7918558762578525,
      "tokens_seen": 677838848
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0004013139418254764,
      "loss": 3.2651,
      "theoretical_loss": 3.791818439870688,
      "tokens_seen": 677904384
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00040130391173520566,
      "loss": 2.9675,
      "theoretical_loss": 3.791781008115737,
      "tokens_seen": 677969920
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0004012938816449348,
      "loss": 2.8882,
      "theoretical_loss": 3.7917435809919775,
      "tokens_seen": 678035456
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.000401283851554664,
      "loss": 2.9325,
      "theoretical_loss": 3.79170615849839,
      "tokens_seen": 678100992
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00040127382146439315,
      "loss": 2.9523,
      "theoretical_loss": 3.7916687406339538,
      "tokens_seen": 678166528
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0004012637913741224,
      "loss": 2.8462,
      "theoretical_loss": 3.7916313273976483,
      "tokens_seen": 678232064
    },
    {
      "epoch": 0.21,
      "objective/train/advantage_avg": 0.4931114912033081,
      "objective/train/docs_used": 393619,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.3024919033050537,
      "objective/train/original_loss": 3.3024919033050537,
      "objective/train/theoretical_loss": 3.7915939187884558,
      "objective/train/tokens_used": 698757600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2433578222990036,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505481958389282,
      "objective/train/weighted_lm_loss": 3.4696004390716553,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0363596677780151,
      "theoretical_loss": 3.7915939187884558,
      "tokens_seen": 678297600
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00040125376128385157,
      "loss": 2.9797,
      "theoretical_loss": 3.7915939187884558,
      "tokens_seen": 678297600
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00040124373119358075,
      "loss": 3.0085,
      "theoretical_loss": 3.791556514805355,
      "tokens_seen": 678363136
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00040123370110330993,
      "loss": 3.0666,
      "theoretical_loss": 3.7915191154473287,
      "tokens_seen": 678428672
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00040122367101303916,
      "loss": 3.0141,
      "theoretical_loss": 3.7914817207133575,
      "tokens_seen": 678494208
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0004012136409227683,
      "loss": 2.8132,
      "theoretical_loss": 3.791444330602423,
      "tokens_seen": 678559744
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0004012036108324975,
      "loss": 2.7876,
      "theoretical_loss": 3.7914069451135073,
      "tokens_seen": 678625280
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00040119358074222665,
      "loss": 3.01,
      "theoretical_loss": 3.791369564245593,
      "tokens_seen": 678690816
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0004011835506519559,
      "loss": 3.1744,
      "theoretical_loss": 3.7913321879976625,
      "tokens_seen": 678756352
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00040117352056168507,
      "loss": 2.8262,
      "theoretical_loss": 3.7912948163686995,
      "tokens_seen": 678821888
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00040116349047141425,
      "loss": 2.8198,
      "theoretical_loss": 3.791257449357686,
      "tokens_seen": 678887424
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00040115346038114343,
      "loss": 2.9569,
      "theoretical_loss": 3.791220086963608,
      "tokens_seen": 678952960
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0004011434302908726,
      "loss": 2.9965,
      "theoretical_loss": 3.7911827291854467,
      "tokens_seen": 679018496
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0004011334002006018,
      "loss": 2.9019,
      "theoretical_loss": 3.791145376022188,
      "tokens_seen": 679084032
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00040112337011033103,
      "loss": 2.9518,
      "theoretical_loss": 3.7911080274728164,
      "tokens_seen": 679149568
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00040111334002006016,
      "loss": 2.7066,
      "theoretical_loss": 3.791070683536316,
      "tokens_seen": 679215104
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0004011033099297894,
      "loss": 2.911,
      "theoretical_loss": 3.791033344211673,
      "tokens_seen": 679280640
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0004010932798395185,
      "loss": 3.0182,
      "theoretical_loss": 3.7909960094978734,
      "tokens_seen": 679346176
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00040108324974924775,
      "loss": 2.7033,
      "theoretical_loss": 3.790958679393902,
      "tokens_seen": 679411712
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00040107321965897693,
      "loss": 3.0447,
      "theoretical_loss": 3.790921353898745,
      "tokens_seen": 679477248
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0004010631895687061,
      "loss": 2.8774,
      "theoretical_loss": 3.79088403301139,
      "tokens_seen": 679542784
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0004010531594784353,
      "loss": 3.166,
      "theoretical_loss": 3.790846716730823,
      "tokens_seen": 679608320
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00040104312938816453,
      "loss": 2.9475,
      "theoretical_loss": 3.7908094050560317,
      "tokens_seen": 679673856
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00040103309929789366,
      "loss": 3.0824,
      "theoretical_loss": 3.790772097986003,
      "tokens_seen": 679739392
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0004010230692076229,
      "loss": 2.9097,
      "theoretical_loss": 3.7907347955197253,
      "tokens_seen": 679804928
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.000401013039117352,
      "loss": 2.8997,
      "theoretical_loss": 3.7906974976561862,
      "tokens_seen": 679870464
    },
    {
      "epoch": 0.21,
      "objective/train/advantage_avg": 0.4939137399196625,
      "objective/train/docs_used": 393966,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.9206013679504395,
      "objective/train/original_loss": 2.9206013679504395,
      "objective/train/theoretical_loss": 3.790660204394375,
      "objective/train/tokens_used": 700396000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24403724074363708,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506318807601929,
      "objective/train/weighted_lm_loss": 3.0683746337890625,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0253151655197144,
      "theoretical_loss": 3.790660204394375,
      "tokens_seen": 679936000
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00040100300902708126,
      "loss": 3.0232,
      "theoretical_loss": 3.790660204394375,
      "tokens_seen": 679936000
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00040099297893681044,
      "loss": 2.9818,
      "theoretical_loss": 3.79062291573328,
      "tokens_seen": 680001536
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0004009829488465396,
      "loss": 3.0144,
      "theoretical_loss": 3.79058563167189,
      "tokens_seen": 680067072
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0004009729187562688,
      "loss": 2.976,
      "theoretical_loss": 3.7905483522091954,
      "tokens_seen": 680132608
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.000400962888665998,
      "loss": 3.0322,
      "theoretical_loss": 3.7905110773441844,
      "tokens_seen": 680198144
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00040095285857572716,
      "loss": 2.9314,
      "theoretical_loss": 3.7904738070758484,
      "tokens_seen": 680263680
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0004009428284854564,
      "loss": 2.9689,
      "theoretical_loss": 3.790436541403177,
      "tokens_seen": 680329216
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0004009327983951855,
      "loss": 2.9933,
      "theoretical_loss": 3.790399280325162,
      "tokens_seen": 680394752
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00040092276830491476,
      "loss": 2.9285,
      "theoretical_loss": 3.7903620238407933,
      "tokens_seen": 680460288
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0004009127382146439,
      "loss": 2.8653,
      "theoretical_loss": 3.790324771949063,
      "tokens_seen": 680525824
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0004009027081243731,
      "loss": 2.8903,
      "theoretical_loss": 3.790287524648962,
      "tokens_seen": 680591360
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00040089267803410236,
      "loss": 2.9828,
      "theoretical_loss": 3.7902502819394828,
      "tokens_seen": 680656896
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0004008826479438315,
      "loss": 2.9231,
      "theoretical_loss": 3.7902130438196173,
      "tokens_seen": 680722432
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0004008726178535607,
      "loss": 2.9563,
      "theoretical_loss": 3.7901758102883587,
      "tokens_seen": 680787968
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0004008625877632899,
      "loss": 2.9811,
      "theoretical_loss": 3.7901385813446997,
      "tokens_seen": 680853504
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0004008525576730191,
      "loss": 2.691,
      "theoretical_loss": 3.790101356987633,
      "tokens_seen": 680919040
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00040084252758274826,
      "loss": 2.8161,
      "theoretical_loss": 3.790064137216153,
      "tokens_seen": 680984576
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00040083249749247744,
      "loss": 2.9335,
      "theoretical_loss": 3.7900269220292535,
      "tokens_seen": 681050112
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0004008224674022066,
      "loss": 2.9705,
      "theoretical_loss": 3.789989711425928,
      "tokens_seen": 681115648
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00040081243731193586,
      "loss": 3.092,
      "theoretical_loss": 3.7899525054051724,
      "tokens_seen": 681181184
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.000400802407221665,
      "loss": 2.8837,
      "theoretical_loss": 3.7899153039659796,
      "tokens_seen": 681246720
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0004007923771313942,
      "loss": 2.9362,
      "theoretical_loss": 3.789878107107346,
      "tokens_seen": 681312256
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00040078234704112335,
      "loss": 3.2265,
      "theoretical_loss": 3.7898409148282672,
      "tokens_seen": 681377792
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0004007723169508526,
      "loss": 3.1595,
      "theoretical_loss": 3.7898037271277385,
      "tokens_seen": 681443328
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00040076228686058177,
      "loss": 2.8637,
      "theoretical_loss": 3.789766544004756,
      "tokens_seen": 681508864
    },
    {
      "epoch": 0.21,
      "objective/train/advantage_avg": 0.4935804307460785,
      "objective/train/docs_used": 395160,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.8150172233581543,
      "objective/train/original_loss": 2.8150174617767334,
      "objective/train/theoretical_loss": 3.7897293654583164,
      "objective/train/tokens_used": 702034400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24371302127838135,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050596833229065,
      "objective/train/weighted_lm_loss": 2.9572176933288574,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0363596677780151,
      "theoretical_loss": 3.7897293654583164,
      "tokens_seen": 681574400
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00040075225677031095,
      "loss": 2.8486,
      "theoretical_loss": 3.7897293654583164,
      "tokens_seen": 681574400
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00040074222668004013,
      "loss": 2.8386,
      "theoretical_loss": 3.7896921914874167,
      "tokens_seen": 681639936
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00040073219658976936,
      "loss": 2.977,
      "theoretical_loss": 3.789655022091053,
      "tokens_seen": 681705472
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0004007221664994985,
      "loss": 3.0398,
      "theoretical_loss": 3.7896178572682238,
      "tokens_seen": 681771008
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0004007121364092277,
      "loss": 2.9944,
      "theoretical_loss": 3.789580697017926,
      "tokens_seen": 681836544
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00040070210631895685,
      "loss": 2.8305,
      "theoretical_loss": 3.7895435413391576,
      "tokens_seen": 681902080
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0004006920762286861,
      "loss": 3.1244,
      "theoretical_loss": 3.7895063902309176,
      "tokens_seen": 681967616
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00040068204613841527,
      "loss": 2.9911,
      "theoretical_loss": 3.789469243692204,
      "tokens_seen": 682033152
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00040067201604814445,
      "loss": 2.7601,
      "theoretical_loss": 3.7894321017220163,
      "tokens_seen": 682098688
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00040066198595787363,
      "loss": 2.8412,
      "theoretical_loss": 3.789394964319353,
      "tokens_seen": 682164224
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0004006519558676028,
      "loss": 3.0786,
      "theoretical_loss": 3.7893578314832146,
      "tokens_seen": 682229760
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.000400641925777332,
      "loss": 3.0512,
      "theoretical_loss": 3.7893207032126,
      "tokens_seen": 682295296
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00040063189568706123,
      "loss": 2.942,
      "theoretical_loss": 3.7892835795065105,
      "tokens_seen": 682360832
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00040062186559679036,
      "loss": 3.0864,
      "theoretical_loss": 3.7892464603639455,
      "tokens_seen": 682426368
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0004006118355065196,
      "loss": 3.0305,
      "theoretical_loss": 3.789209345783907,
      "tokens_seen": 682491904
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0004006018054162487,
      "loss": 2.9899,
      "theoretical_loss": 3.7891722357653954,
      "tokens_seen": 682557440
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00040059177532597795,
      "loss": 2.9257,
      "theoretical_loss": 3.7891351303074123,
      "tokens_seen": 682622976
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00040058174523570713,
      "loss": 2.8848,
      "theoretical_loss": 3.78909802940896,
      "tokens_seen": 682688512
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0004005717151454363,
      "loss": 3.102,
      "theoretical_loss": 3.789060933069039,
      "tokens_seen": 682754048
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0004005616850551655,
      "loss": 3.2955,
      "theoretical_loss": 3.789023841286654,
      "tokens_seen": 682819584
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00040055165496489473,
      "loss": 2.9526,
      "theoretical_loss": 3.788986754060806,
      "tokens_seen": 682885120
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00040054162487462386,
      "loss": 2.789,
      "theoretical_loss": 3.7889496713904993,
      "tokens_seen": 682950656
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0004005315947843531,
      "loss": 2.9157,
      "theoretical_loss": 3.788912593274736,
      "tokens_seen": 683016192
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0004005215646940822,
      "loss": 2.9194,
      "theoretical_loss": 3.7888755197125206,
      "tokens_seen": 683081728
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00040051153460381146,
      "loss": 3.0919,
      "theoretical_loss": 3.7888384507028565,
      "tokens_seen": 683147264
    },
    {
      "epoch": 0.21,
      "objective/train/advantage_avg": 0.4957548677921295,
      "objective/train/docs_used": 395993,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.7254929542541504,
      "objective/train/original_loss": 2.7254927158355713,
      "objective/train/theoretical_loss": 3.788801386244749,
      "objective/train/tokens_used": 703672800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2458087056875229,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0508251190185547,
      "objective/train/weighted_lm_loss": 2.8639883995056152,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0304546356201172,
      "theoretical_loss": 3.788801386244749,
      "tokens_seen": 683212800
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00040050150451354064,
      "loss": 2.7331,
      "theoretical_loss": 3.788801386244749,
      "tokens_seen": 683212800
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0004004914744232698,
      "loss": 2.892,
      "theoretical_loss": 3.7887643263372017,
      "tokens_seen": 683278336
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.000400481444332999,
      "loss": 3.3355,
      "theoretical_loss": 3.78872727097922,
      "tokens_seen": 683343872
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0004004714142427282,
      "loss": 2.8703,
      "theoretical_loss": 3.788690220169809,
      "tokens_seen": 683409408
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00040046138415245736,
      "loss": 2.996,
      "theoretical_loss": 3.7886531739079743,
      "tokens_seen": 683474944
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0004004513540621866,
      "loss": 3.0089,
      "theoretical_loss": 3.788616132192722,
      "tokens_seen": 683540480
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0004004413239719157,
      "loss": 3.1957,
      "theoretical_loss": 3.7885790950230582,
      "tokens_seen": 683606016
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00040043129388164496,
      "loss": 3.1417,
      "theoretical_loss": 3.7885420623979886,
      "tokens_seen": 683671552
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0004004212637913741,
      "loss": 2.9074,
      "theoretical_loss": 3.788505034316521,
      "tokens_seen": 683737088
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0004004112337011033,
      "loss": 2.8111,
      "theoretical_loss": 3.7884680107776623,
      "tokens_seen": 683802624
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0004004012036108325,
      "loss": 3.188,
      "theoretical_loss": 3.788430991780419,
      "tokens_seen": 683868160
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0004003911735205617,
      "loss": 2.919,
      "theoretical_loss": 3.7883939773238002,
      "tokens_seen": 683933696
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00040038114343029087,
      "loss": 2.9937,
      "theoretical_loss": 3.7883569674068136,
      "tokens_seen": 683999232
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0004003711133400201,
      "loss": 2.8987,
      "theoretical_loss": 3.788319962028467,
      "tokens_seen": 684064768
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00040036108324974923,
      "loss": 2.9304,
      "theoretical_loss": 3.788282961187769,
      "tokens_seen": 684130304
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00040035105315947846,
      "loss": 2.9993,
      "theoretical_loss": 3.78824596488373,
      "tokens_seen": 684195840
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0004003410230692076,
      "loss": 3.0272,
      "theoretical_loss": 3.7882089731153576,
      "tokens_seen": 684261376
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0004003309929789368,
      "loss": 2.8221,
      "theoretical_loss": 3.788171985881662,
      "tokens_seen": 684326912
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.000400320962888666,
      "loss": 3.082,
      "theoretical_loss": 3.7881350031816536,
      "tokens_seen": 684392448
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0004003109327983952,
      "loss": 2.86,
      "theoretical_loss": 3.788098025014342,
      "tokens_seen": 684457984
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00040030090270812437,
      "loss": 2.8603,
      "theoretical_loss": 3.788061051378738,
      "tokens_seen": 684523520
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00040029087261785355,
      "loss": 2.8174,
      "theoretical_loss": 3.7880240822738527,
      "tokens_seen": 684589056
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00040028084252758273,
      "loss": 2.7279,
      "theoretical_loss": 3.7879871176986963,
      "tokens_seen": 684654592
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00040027081243731197,
      "loss": 3.1412,
      "theoretical_loss": 3.787950157652282,
      "tokens_seen": 684720128
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0004002607823470411,
      "loss": 2.9883,
      "theoretical_loss": 3.7879132021336197,
      "tokens_seen": 684785664
    },
    {
      "epoch": 0.21,
      "objective/train/advantage_avg": 0.4943750500679016,
      "objective/train/docs_used": 397274,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.7895469665527344,
      "objective/train/original_loss": 2.7895472049713135,
      "objective/train/theoretical_loss": 3.7878762511417223,
      "objective/train/tokens_used": 705311200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24449528753757477,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506802797317505,
      "objective/train/weighted_lm_loss": 2.9306061267852783,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0396549701690674,
      "theoretical_loss": 3.7878762511417223,
      "tokens_seen": 684851200
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00040025075225677033,
      "loss": 2.8982,
      "theoretical_loss": 3.7878762511417223,
      "tokens_seen": 684851200
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00040024072216649946,
      "loss": 3.0062,
      "theoretical_loss": 3.7878393046756025,
      "tokens_seen": 684916736
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0004002306920762287,
      "loss": 3.1174,
      "theoretical_loss": 3.787802362734273,
      "tokens_seen": 684982272
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00040022066198595787,
      "loss": 2.9315,
      "theoretical_loss": 3.787765425316746,
      "tokens_seen": 685047808
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00040021063189568705,
      "loss": 2.9468,
      "theoretical_loss": 3.787728492422036,
      "tokens_seen": 685113344
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00040020060180541623,
      "loss": 2.898,
      "theoretical_loss": 3.7876915640491555,
      "tokens_seen": 685178880
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00040019057171514547,
      "loss": 3.0905,
      "theoretical_loss": 3.7876546401971196,
      "tokens_seen": 685244416
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0004001805416248746,
      "loss": 3.1022,
      "theoretical_loss": 3.787617720864942,
      "tokens_seen": 685309952
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00040017051153460383,
      "loss": 2.8745,
      "theoretical_loss": 3.7875808060516363,
      "tokens_seen": 685375488
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00040016048144433296,
      "loss": 3.0741,
      "theoretical_loss": 3.787543895756219,
      "tokens_seen": 685441024
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0004001504513540622,
      "loss": 3.0696,
      "theoretical_loss": 3.7875069899777047,
      "tokens_seen": 685506560
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00040014042126379143,
      "loss": 3.245,
      "theoretical_loss": 3.7874700887151085,
      "tokens_seen": 685572096
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00040013039117352056,
      "loss": 3.1024,
      "theoretical_loss": 3.7874331919674464,
      "tokens_seen": 685637632
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0004001203610832498,
      "loss": 3.0806,
      "theoretical_loss": 3.787396299733735,
      "tokens_seen": 685703168
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0004001103309929789,
      "loss": 2.8956,
      "theoretical_loss": 3.78735941201299,
      "tokens_seen": 685768704
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00040010030090270815,
      "loss": 3.1581,
      "theoretical_loss": 3.7873225288042285,
      "tokens_seen": 685834240
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00040009027081243733,
      "loss": 2.9793,
      "theoretical_loss": 3.787285650106468,
      "tokens_seen": 685899776
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0004000802407221665,
      "loss": 3.0807,
      "theoretical_loss": 3.7872487759187248,
      "tokens_seen": 685965312
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0004000702106318957,
      "loss": 3.197,
      "theoretical_loss": 3.7872119062400174,
      "tokens_seen": 686030848
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00040006018054162493,
      "loss": 2.9261,
      "theoretical_loss": 3.787175041069363,
      "tokens_seen": 686096384
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00040005015045135406,
      "loss": 2.8215,
      "theoretical_loss": 3.7871381804057807,
      "tokens_seen": 686161920
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0004000401203610833,
      "loss": 2.8503,
      "theoretical_loss": 3.7871013242482885,
      "tokens_seen": 686227456
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0004000300902708124,
      "loss": 2.8965,
      "theoretical_loss": 3.7870644725959055,
      "tokens_seen": 686292992
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00040002006018054166,
      "loss": 2.8233,
      "theoretical_loss": 3.787027625447651,
      "tokens_seen": 686358528
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00040001003009027084,
      "loss": 2.8515,
      "theoretical_loss": 3.7869907828025444,
      "tokens_seen": 686424064
    },
    {
      "epoch": 0.21,
      "objective/train/advantage_avg": 0.49237728118896484,
      "objective/train/docs_used": 397954,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.736746072769165,
      "objective/train/original_loss": 2.736746072769165,
      "objective/train/theoretical_loss": 3.786953944659605,
      "objective/train/tokens_used": 706949600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24250245094299316,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0504703521728516,
      "objective/train/weighted_lm_loss": 2.8747143745422363,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0461136102676392,
      "theoretical_loss": 3.786953944659605,
      "tokens_seen": 686489600
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0004,
      "loss": 2.9302,
      "theoretical_loss": 3.786953944659605,
      "tokens_seen": 686489600
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003999899699097292,
      "loss": 2.9804,
      "theoretical_loss": 3.7869171110178543,
      "tokens_seen": 686555136
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003999799398194584,
      "loss": 2.9414,
      "theoretical_loss": 3.7868802818763116,
      "tokens_seen": 686620672
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039996990972918756,
      "loss": 2.8867,
      "theoretical_loss": 3.7868434572339975,
      "tokens_seen": 686686208
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003999598796389168,
      "loss": 3.0685,
      "theoretical_loss": 3.786806637089934,
      "tokens_seen": 686751744
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003999498495486459,
      "loss": 2.9011,
      "theoretical_loss": 3.786769821443141,
      "tokens_seen": 686817280
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039993981945837516,
      "loss": 2.9758,
      "theoretical_loss": 3.7867330102926413,
      "tokens_seen": 686882816
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003999297893681043,
      "loss": 3.1025,
      "theoretical_loss": 3.7866962036374567,
      "tokens_seen": 686948352
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003999197592778335,
      "loss": 2.8154,
      "theoretical_loss": 3.7866594014766095,
      "tokens_seen": 687013888
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003999097291875627,
      "loss": 3.0783,
      "theoretical_loss": 3.7866226038091213,
      "tokens_seen": 687079424
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003998996990972919,
      "loss": 2.9682,
      "theoretical_loss": 3.786585810634016,
      "tokens_seen": 687144960
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039988966900702107,
      "loss": 3.1288,
      "theoretical_loss": 3.7865490219503166,
      "tokens_seen": 687210496
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003998796389167503,
      "loss": 3.1116,
      "theoretical_loss": 3.786512237757046,
      "tokens_seen": 687276032
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039986960882647943,
      "loss": 2.8838,
      "theoretical_loss": 3.7864754580532294,
      "tokens_seen": 687341568
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039985957873620866,
      "loss": 3.037,
      "theoretical_loss": 3.7864386828378898,
      "tokens_seen": 687407104
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003998495486459378,
      "loss": 2.7803,
      "theoretical_loss": 3.786401912110051,
      "tokens_seen": 687472640
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.000399839518555667,
      "loss": 3.1366,
      "theoretical_loss": 3.786365145868739,
      "tokens_seen": 687538176
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003998294884653962,
      "loss": 2.9988,
      "theoretical_loss": 3.786328384112978,
      "tokens_seen": 687603712
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003998194583751254,
      "loss": 2.8922,
      "theoretical_loss": 3.786291626841794,
      "tokens_seen": 687669248
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039980942828485457,
      "loss": 2.9333,
      "theoretical_loss": 3.7862548740542117,
      "tokens_seen": 687734784
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039979939819458375,
      "loss": 2.9573,
      "theoretical_loss": 3.7862181257492575,
      "tokens_seen": 687800320
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039978936810431293,
      "loss": 2.8559,
      "theoretical_loss": 3.7861813819259575,
      "tokens_seen": 687865856
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039977933801404217,
      "loss": 3.2601,
      "theoretical_loss": 3.7861446425833387,
      "tokens_seen": 687931392
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003997693079237713,
      "loss": 3.1137,
      "theoretical_loss": 3.786107907720428,
      "tokens_seen": 687996928
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039975927783350053,
      "loss": 2.723,
      "theoretical_loss": 3.786071177336251,
      "tokens_seen": 688062464
    },
    {
      "epoch": 0.21,
      "objective/train/advantage_avg": 0.49200087785720825,
      "objective/train/docs_used": 398552,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.981285810470581,
      "objective/train/original_loss": 2.981285572052002,
      "objective/train/theoretical_loss": 3.7860344514298374,
      "objective/train/tokens_used": 708588000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24214671552181244,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0504308938980103,
      "objective/train/weighted_lm_loss": 3.131465435028076,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0425469875335693,
      "theoretical_loss": 3.7860344514298374,
      "tokens_seen": 688128000
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039974924774322966,
      "loss": 2.9462,
      "theoretical_loss": 3.7860344514298374,
      "tokens_seen": 688128000
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003997392176529589,
      "loss": 3.1054,
      "theoretical_loss": 3.7859977300002132,
      "tokens_seen": 688193536
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039972918756268807,
      "loss": 2.9795,
      "theoretical_loss": 3.785961013046407,
      "tokens_seen": 688259072
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039971915747241725,
      "loss": 3.0056,
      "theoretical_loss": 3.785924300567448,
      "tokens_seen": 688324608
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039970912738214643,
      "loss": 3.0526,
      "theoretical_loss": 3.7858875925623634,
      "tokens_seen": 688390144
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039969909729187567,
      "loss": 3.1771,
      "theoretical_loss": 3.7858508890301827,
      "tokens_seen": 688455680
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003996890672016048,
      "loss": 2.9839,
      "theoretical_loss": 3.785814189969935,
      "tokens_seen": 688521216
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039967903711133403,
      "loss": 3.236,
      "theoretical_loss": 3.7857774953806507,
      "tokens_seen": 688586752
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039966900702106316,
      "loss": 3.0181,
      "theoretical_loss": 3.7857408052613595,
      "tokens_seen": 688652288
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003996589769307924,
      "loss": 3.0561,
      "theoretical_loss": 3.7857041196110908,
      "tokens_seen": 688717824
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003996489468405216,
      "loss": 3.0942,
      "theoretical_loss": 3.7856674384288755,
      "tokens_seen": 688783360
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039963891675025076,
      "loss": 2.722,
      "theoretical_loss": 3.785630761713744,
      "tokens_seen": 688848896
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039962888665997994,
      "loss": 2.9725,
      "theoretical_loss": 3.7855940894647278,
      "tokens_seen": 688914432
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003996188565697091,
      "loss": 3.0463,
      "theoretical_loss": 3.785557421680859,
      "tokens_seen": 688979968
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003996088264794383,
      "loss": 3.1939,
      "theoretical_loss": 3.7855207583611676,
      "tokens_seen": 689045504
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039959879638916754,
      "loss": 2.8392,
      "theoretical_loss": 3.7854840995046866,
      "tokens_seen": 689111040
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039958876629889666,
      "loss": 2.7374,
      "theoretical_loss": 3.7854474451104485,
      "tokens_seen": 689176576
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003995787362086259,
      "loss": 2.9473,
      "theoretical_loss": 3.7854107951774854,
      "tokens_seen": 689242112
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003995687061183551,
      "loss": 2.8469,
      "theoretical_loss": 3.7853741497048303,
      "tokens_seen": 689307648
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039955867602808426,
      "loss": 2.7997,
      "theoretical_loss": 3.7853375086915166,
      "tokens_seen": 689373184
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039954864593781344,
      "loss": 2.9865,
      "theoretical_loss": 3.785300872136577,
      "tokens_seen": 689438720
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003995386158475426,
      "loss": 2.9206,
      "theoretical_loss": 3.785264240039046,
      "tokens_seen": 689504256
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003995285857572718,
      "loss": 2.5572,
      "theoretical_loss": 3.7852276123979585,
      "tokens_seen": 689569792
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039951855566700104,
      "loss": 2.6981,
      "theoretical_loss": 3.7851909892123476,
      "tokens_seen": 689635328
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039950852557673017,
      "loss": 2.9609,
      "theoretical_loss": 3.7851543704812487,
      "tokens_seen": 689700864
    },
    {
      "epoch": 0.21,
      "objective/train/advantage_avg": 0.4946171045303345,
      "objective/train/docs_used": 399765,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.512892723083496,
      "objective/train/original_loss": 2.512892961502075,
      "objective/train/theoretical_loss": 3.7851177562036957,
      "objective/train/tokens_used": 710226400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24471257627010345,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507055521011353,
      "objective/train/weighted_lm_loss": 2.6403472423553467,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0304546356201172,
      "theoretical_loss": 3.7851177562036957,
      "tokens_seen": 689766400
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003994984954864594,
      "loss": 2.7237,
      "theoretical_loss": 3.7851177562036957,
      "tokens_seen": 689766400
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039948846539618853,
      "loss": 2.8847,
      "theoretical_loss": 3.785081146378725,
      "tokens_seen": 689831936
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039947843530591776,
      "loss": 2.8888,
      "theoretical_loss": 3.785044541005372,
      "tokens_seen": 689897472
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039946840521564694,
      "loss": 3.2223,
      "theoretical_loss": 3.785007940082673,
      "tokens_seen": 689963008
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003994583751253761,
      "loss": 3.3059,
      "theoretical_loss": 3.784971343609663,
      "tokens_seen": 690028544
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003994483450351053,
      "loss": 2.6826,
      "theoretical_loss": 3.7849347515853795,
      "tokens_seen": 690094080
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003994383149448345,
      "loss": 3.0345,
      "theoretical_loss": 3.784898164008859,
      "tokens_seen": 690159616
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039942828485456367,
      "loss": 2.954,
      "theoretical_loss": 3.784861580879139,
      "tokens_seen": 690225152
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003994182547642929,
      "loss": 3.2963,
      "theoretical_loss": 3.784825002195256,
      "tokens_seen": 690290688
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039940822467402203,
      "loss": 2.9814,
      "theoretical_loss": 3.7847884279562485,
      "tokens_seen": 690356224
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039939819458375127,
      "loss": 3.0042,
      "theoretical_loss": 3.7847518581611537,
      "tokens_seen": 690421760
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003993881644934805,
      "loss": 3.0509,
      "theoretical_loss": 3.784715292809011,
      "tokens_seen": 690487296
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039937813440320963,
      "loss": 2.6696,
      "theoretical_loss": 3.7846787318988584,
      "tokens_seen": 690552832
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039936810431293886,
      "loss": 2.9355,
      "theoretical_loss": 3.7846421754297346,
      "tokens_seen": 690618368
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.000399358074222668,
      "loss": 2.8316,
      "theoretical_loss": 3.7846056234006786,
      "tokens_seen": 690683904
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003993480441323972,
      "loss": 2.934,
      "theoretical_loss": 3.7845690758107313,
      "tokens_seen": 690749440
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003993380140421264,
      "loss": 2.9855,
      "theoretical_loss": 3.7845325326589307,
      "tokens_seen": 690814976
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003993279839518556,
      "loss": 2.9175,
      "theoretical_loss": 3.784495993944318,
      "tokens_seen": 690880512
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039931795386158477,
      "loss": 3.1346,
      "theoretical_loss": 3.784459459665934,
      "tokens_seen": 690946048
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039930792377131395,
      "loss": 3.1536,
      "theoretical_loss": 3.7844229298228176,
      "tokens_seen": 691011584
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039929789368104313,
      "loss": 2.9629,
      "theoretical_loss": 3.784386404414011,
      "tokens_seen": 691077120
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039928786359077237,
      "loss": 2.874,
      "theoretical_loss": 3.784349883438556,
      "tokens_seen": 691142656
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003992778335005015,
      "loss": 2.8068,
      "theoretical_loss": 3.784313366895493,
      "tokens_seen": 691208192
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039926780341023073,
      "loss": 2.974,
      "theoretical_loss": 3.7842768547838648,
      "tokens_seen": 691273728
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039925777331995986,
      "loss": 2.9973,
      "theoretical_loss": 3.7842403471027133,
      "tokens_seen": 691339264
    },
    {
      "epoch": 0.21,
      "objective/train/advantage_avg": 0.493630975484848,
      "objective/train/docs_used": 400468,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.872967004776001,
      "objective/train/original_loss": 2.87296724319458,
      "objective/train/theoretical_loss": 3.7842038438510803,
      "objective/train/tokens_used": 711864800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24375323951244354,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506021976470947,
      "objective/train/weighted_lm_loss": 3.018203020095825,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0396549701690674,
      "theoretical_loss": 3.7842038438510803,
      "tokens_seen": 691404800
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003992477432296891,
      "loss": 2.9465,
      "theoretical_loss": 3.7842038438510803,
      "tokens_seen": 691404800
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039923771313941827,
      "loss": 2.6796,
      "theoretical_loss": 3.7841673450280098,
      "tokens_seen": 691470336
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039922768304914745,
      "loss": 3.0072,
      "theoretical_loss": 3.784130850632544,
      "tokens_seen": 691535872
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039921765295887663,
      "loss": 3.0204,
      "theoretical_loss": 3.7840943606637265,
      "tokens_seen": 691601408
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039920762286860587,
      "loss": 3.2127,
      "theoretical_loss": 3.7840578751206015,
      "tokens_seen": 691666944
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.000399197592778335,
      "loss": 2.8414,
      "theoretical_loss": 3.784021394002212,
      "tokens_seen": 691732480
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039918756268806423,
      "loss": 2.9902,
      "theoretical_loss": 3.7839849173076026,
      "tokens_seen": 691798016
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039917753259779336,
      "loss": 3.0092,
      "theoretical_loss": 3.783948445035818,
      "tokens_seen": 691863552
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003991675025075226,
      "loss": 2.9038,
      "theoretical_loss": 3.7839119771859036,
      "tokens_seen": 691929088
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003991574724172518,
      "loss": 2.9703,
      "theoretical_loss": 3.783875513756904,
      "tokens_seen": 691994624
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039914744232698096,
      "loss": 3.0498,
      "theoretical_loss": 3.7838390547478635,
      "tokens_seen": 692060160
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039913741223671014,
      "loss": 2.9226,
      "theoretical_loss": 3.7838026001578298,
      "tokens_seen": 692125696
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003991273821464393,
      "loss": 2.8273,
      "theoretical_loss": 3.7837661499858486,
      "tokens_seen": 692191232
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003991173520561685,
      "loss": 2.9195,
      "theoretical_loss": 3.783729704230965,
      "tokens_seen": 692256768
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039910732196589774,
      "loss": 2.9334,
      "theoretical_loss": 3.783693262892227,
      "tokens_seen": 692322304
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039909729187562686,
      "loss": 3.374,
      "theoretical_loss": 3.7836568259686807,
      "tokens_seen": 692387840
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003990872617853561,
      "loss": 3.0695,
      "theoretical_loss": 3.7836203934593735,
      "tokens_seen": 692453376
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003990772316950853,
      "loss": 3.0194,
      "theoretical_loss": 3.7835839653633534,
      "tokens_seen": 692518912
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039906720160481446,
      "loss": 3.0492,
      "theoretical_loss": 3.783547541679668,
      "tokens_seen": 692584448
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039905717151454364,
      "loss": 3.0302,
      "theoretical_loss": 3.783511122407364,
      "tokens_seen": 692649984
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003990471414242728,
      "loss": 2.8783,
      "theoretical_loss": 3.7834747075454924,
      "tokens_seen": 692715520
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.000399037111334002,
      "loss": 2.8915,
      "theoretical_loss": 3.7834382970930998,
      "tokens_seen": 692781056
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039902708124373124,
      "loss": 3.0057,
      "theoretical_loss": 3.7834018910492366,
      "tokens_seen": 692846592
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039901705115346037,
      "loss": 2.8771,
      "theoretical_loss": 3.783365489412951,
      "tokens_seen": 692912128
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003990070210631896,
      "loss": 3.0391,
      "theoretical_loss": 3.783329092183293,
      "tokens_seen": 692977664
    },
    {
      "epoch": 0.21,
      "objective/train/advantage_avg": 0.4912099838256836,
      "objective/train/docs_used": 401917,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.974303960800171,
      "objective/train/original_loss": 2.974303960800171,
      "objective/train/theoretical_loss": 3.783292699359313,
      "objective/train/tokens_used": 713503200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24144220352172852,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050348162651062,
      "objective/train/weighted_lm_loss": 3.1241824626922607,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 3.783292699359313,
      "tokens_seen": 693043200
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039899699097291873,
      "loss": 2.9547,
      "theoretical_loss": 3.783292699359313,
      "tokens_seen": 693043200
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039898696088264796,
      "loss": 3.3349,
      "theoretical_loss": 3.78325631094006,
      "tokens_seen": 693108736
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039897693079237714,
      "loss": 3.027,
      "theoretical_loss": 3.7832199269245863,
      "tokens_seen": 693174272
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003989669007021063,
      "loss": 3.0825,
      "theoretical_loss": 3.783183547311941,
      "tokens_seen": 693239808
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003989568706118355,
      "loss": 3.0013,
      "theoretical_loss": 3.7831471721011756,
      "tokens_seen": 693305344
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003989468405215647,
      "loss": 2.9322,
      "theoretical_loss": 3.7831108012913415,
      "tokens_seen": 693370880
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039893681043129387,
      "loss": 2.9358,
      "theoretical_loss": 3.783074434881491,
      "tokens_seen": 693436416
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003989267803410231,
      "loss": 2.9499,
      "theoretical_loss": 3.783038072870675,
      "tokens_seen": 693501952
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039891675025075223,
      "loss": 3.0683,
      "theoretical_loss": 3.783001715257946,
      "tokens_seen": 693567488
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039890672016048147,
      "loss": 3.0149,
      "theoretical_loss": 3.7829653620423573,
      "tokens_seen": 693633024
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039889669007021065,
      "loss": 3.0237,
      "theoretical_loss": 3.782929013222961,
      "tokens_seen": 693698560
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039888665997993983,
      "loss": 2.8807,
      "theoretical_loss": 3.7828926687988105,
      "tokens_seen": 693764096
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.000398876629889669,
      "loss": 2.8604,
      "theoretical_loss": 3.7828563287689594,
      "tokens_seen": 693829632
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003988665997993982,
      "loss": 2.925,
      "theoretical_loss": 3.7828199931324606,
      "tokens_seen": 693895168
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039885656970912737,
      "loss": 2.9405,
      "theoretical_loss": 3.782783661888369,
      "tokens_seen": 693960704
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003988465396188566,
      "loss": 2.9588,
      "theoretical_loss": 3.7827473350357383,
      "tokens_seen": 694026240
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039883650952858573,
      "loss": 3.1466,
      "theoretical_loss": 3.782711012573623,
      "tokens_seen": 694091776
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039882647943831497,
      "loss": 3.1601,
      "theoretical_loss": 3.782674694501079,
      "tokens_seen": 694157312
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003988164493480441,
      "loss": 2.8207,
      "theoretical_loss": 3.7826383808171604,
      "tokens_seen": 694222848
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039880641925777333,
      "loss": 3.0496,
      "theoretical_loss": 3.782602071520923,
      "tokens_seen": 694288384
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003987963891675025,
      "loss": 3.1391,
      "theoretical_loss": 3.782565766611423,
      "tokens_seen": 694353920
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003987863590772317,
      "loss": 2.9984,
      "theoretical_loss": 3.782529466087716,
      "tokens_seen": 694419456
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003987763289869609,
      "loss": 2.9289,
      "theoretical_loss": 3.782493169948858,
      "tokens_seen": 694484992
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039876629889669006,
      "loss": 3.1624,
      "theoretical_loss": 3.782456878193906,
      "tokens_seen": 694550528
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039875626880641924,
      "loss": 3.1517,
      "theoretical_loss": 3.7824205908219173,
      "tokens_seen": 694616064
    },
    {
      "epoch": 0.21,
      "objective/train/advantage_avg": 0.49166324734687805,
      "objective/train/docs_used": 402587,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.6035444736480713,
      "objective/train/original_loss": 2.603544235229492,
      "objective/train/theoretical_loss": 3.782384307831949,
      "objective/train/tokens_used": 715141600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24180808663368225,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0503953695297241,
      "objective/train/weighted_lm_loss": 2.7346432209014893,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 3.782384307831949,
      "tokens_seen": 694681600
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039874623871614847,
      "loss": 2.9,
      "theoretical_loss": 3.782384307831949,
      "tokens_seen": 694681600
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003987362086258776,
      "loss": 3.3068,
      "theoretical_loss": 3.7823480292230576,
      "tokens_seen": 694747136
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039872617853560683,
      "loss": 2.9948,
      "theoretical_loss": 3.7823117549943017,
      "tokens_seen": 694812672
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.000398716148445336,
      "loss": 3.0056,
      "theoretical_loss": 3.7822754851447398,
      "tokens_seen": 694878208
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003987061183550652,
      "loss": 3.1407,
      "theoretical_loss": 3.78223921967343,
      "tokens_seen": 694943744
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003986960882647944,
      "loss": 2.9323,
      "theoretical_loss": 3.782202958579431,
      "tokens_seen": 695009280
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039868605817452356,
      "loss": 2.9272,
      "theoretical_loss": 3.782166701861801,
      "tokens_seen": 695074816
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039867602808425274,
      "loss": 2.8224,
      "theoretical_loss": 3.7821304495195998,
      "tokens_seen": 695140352
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.000398665997993982,
      "loss": 2.9884,
      "theoretical_loss": 3.782094201551887,
      "tokens_seen": 695205888
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003986559679037111,
      "loss": 2.9583,
      "theoretical_loss": 3.782057957957723,
      "tokens_seen": 695271424
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039864593781344034,
      "loss": 3.1134,
      "theoretical_loss": 3.782021718736167,
      "tokens_seen": 695336960
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003986359077231695,
      "loss": 3.1862,
      "theoretical_loss": 3.7819854838862796,
      "tokens_seen": 695402496
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003986258776328987,
      "loss": 2.9726,
      "theoretical_loss": 3.781949253407122,
      "tokens_seen": 695468032
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039861584754262794,
      "loss": 3.0586,
      "theoretical_loss": 3.7819130272977546,
      "tokens_seen": 695533568
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039860581745235706,
      "loss": 2.9497,
      "theoretical_loss": 3.781876805557239,
      "tokens_seen": 695599104
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003985957873620863,
      "loss": 2.8877,
      "theoretical_loss": 3.781840588184637,
      "tokens_seen": 695664640
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003985857572718155,
      "loss": 2.7726,
      "theoretical_loss": 3.7818043751790102,
      "tokens_seen": 695730176
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039857572718154466,
      "loss": 2.9458,
      "theoretical_loss": 3.781768166539421,
      "tokens_seen": 695795712
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039856569709127384,
      "loss": 2.892,
      "theoretical_loss": 3.7817319622649315,
      "tokens_seen": 695861248
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.000398555667001003,
      "loss": 2.846,
      "theoretical_loss": 3.7816957623546044,
      "tokens_seen": 695926784
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003985456369107322,
      "loss": 3.1985,
      "theoretical_loss": 3.7816595668075026,
      "tokens_seen": 695992320
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039853560682046144,
      "loss": 2.8658,
      "theoretical_loss": 3.78162337562269,
      "tokens_seen": 696057856
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039852557673019057,
      "loss": 2.9118,
      "theoretical_loss": 3.7815871887992305,
      "tokens_seen": 696123392
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003985155466399198,
      "loss": 3.0055,
      "theoretical_loss": 3.781551006336187,
      "tokens_seen": 696188928
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039850551654964893,
      "loss": 2.8184,
      "theoretical_loss": 3.7815148282326243,
      "tokens_seen": 696254464
    },
    {
      "epoch": 0.21,
      "objective/train/advantage_avg": 0.4954010546207428,
      "objective/train/docs_used": 403173,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.4904894828796387,
      "objective/train/original_loss": 2.4904897212982178,
      "objective/train/theoretical_loss": 3.7814786544876062,
      "objective/train/tokens_used": 716780000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24547475576400757,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507879257202148,
      "objective/train/weighted_lm_loss": 2.61680269241333,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0408108234405518,
      "theoretical_loss": 3.7814786544876062,
      "tokens_seen": 696320000
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039849548645937816,
      "loss": 2.677,
      "theoretical_loss": 3.7814786544876062,
      "tokens_seen": 696320000
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039848545636910734,
      "loss": 3.0033,
      "theoretical_loss": 3.7814424851001984,
      "tokens_seen": 696385536
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003984754262788365,
      "loss": 2.9913,
      "theoretical_loss": 3.7814063200694656,
      "tokens_seen": 696451072
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003984653961885657,
      "loss": 2.799,
      "theoretical_loss": 3.7813701593944726,
      "tokens_seen": 696516608
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003984553660982949,
      "loss": 2.9673,
      "theoretical_loss": 3.781334003074286,
      "tokens_seen": 696582144
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039844533600802407,
      "loss": 2.7845,
      "theoretical_loss": 3.781297851107971,
      "tokens_seen": 696647680
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003984353059177533,
      "loss": 2.921,
      "theoretical_loss": 3.781261703494594,
      "tokens_seen": 696713216
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039842527582748243,
      "loss": 2.8931,
      "theoretical_loss": 3.7812255602332216,
      "tokens_seen": 696778752
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039841524573721167,
      "loss": 3.0765,
      "theoretical_loss": 3.78118942132292,
      "tokens_seen": 696844288
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039840521564694085,
      "loss": 2.9391,
      "theoretical_loss": 3.7811532867627573,
      "tokens_seen": 696909824
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039839518555667003,
      "loss": 3.0152,
      "theoretical_loss": 3.7811171565518,
      "tokens_seen": 696975360
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003983851554663992,
      "loss": 2.7407,
      "theoretical_loss": 3.7810810306891165,
      "tokens_seen": 697040896
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003983751253761284,
      "loss": 3.0021,
      "theoretical_loss": 3.781044909173774,
      "tokens_seen": 697106432
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039836509528585757,
      "loss": 3.0666,
      "theoretical_loss": 3.7810087920048407,
      "tokens_seen": 697171968
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003983550651955868,
      "loss": 2.6797,
      "theoretical_loss": 3.780972679181386,
      "tokens_seen": 697237504
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039834503510531593,
      "loss": 2.9839,
      "theoretical_loss": 3.780936570702478,
      "tokens_seen": 697303040
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039833500501504517,
      "loss": 3.0491,
      "theoretical_loss": 3.7809004665671857,
      "tokens_seen": 697368576
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003983249749247743,
      "loss": 2.8781,
      "theoretical_loss": 3.7808643667745785,
      "tokens_seen": 697434112
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039831494483450353,
      "loss": 3.008,
      "theoretical_loss": 3.7808282713237262,
      "tokens_seen": 697499648
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003983049147442327,
      "loss": 2.9868,
      "theoretical_loss": 3.780792180213699,
      "tokens_seen": 697565184
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003982948846539619,
      "loss": 2.9325,
      "theoretical_loss": 3.780756093443567,
      "tokens_seen": 697630720
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003982848545636911,
      "loss": 2.9056,
      "theoretical_loss": 3.7807200110124004,
      "tokens_seen": 697696256
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039827482447342026,
      "loss": 2.9609,
      "theoretical_loss": 3.7806839329192705,
      "tokens_seen": 697761792
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039826479438314944,
      "loss": 3.0749,
      "theoretical_loss": 3.780647859163248,
      "tokens_seen": 697827328
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039825476429287867,
      "loss": 2.7247,
      "theoretical_loss": 3.7806117897434044,
      "tokens_seen": 697892864
    },
    {
      "epoch": 0.21,
      "objective/train/advantage_avg": 0.49372726678848267,
      "objective/train/docs_used": 404489,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.997882127761841,
      "objective/train/original_loss": 2.9978816509246826,
      "objective/train/theoretical_loss": 3.780575724658811,
      "objective/train/tokens_used": 718418400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2438434511423111,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506123304367065,
      "objective/train/weighted_lm_loss": 3.149364709854126,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0338950157165527,
      "theoretical_loss": 3.780575724658811,
      "tokens_seen": 697958400
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003982447342026078,
      "loss": 2.852,
      "theoretical_loss": 3.780575724658811,
      "tokens_seen": 697958400
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039823470411233703,
      "loss": 2.9054,
      "theoretical_loss": 3.7805396639085407,
      "tokens_seen": 698023936
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003982246740220662,
      "loss": 3.0531,
      "theoretical_loss": 3.7805036074916645,
      "tokens_seen": 698089472
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003982146439317954,
      "loss": 2.6452,
      "theoretical_loss": 3.780467555407255,
      "tokens_seen": 698155008
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003982046138415246,
      "loss": 2.9105,
      "theoretical_loss": 3.7804315076543866,
      "tokens_seen": 698220544
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039819458375125376,
      "loss": 2.811,
      "theoretical_loss": 3.7803954642321314,
      "tokens_seen": 698286080
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039818455366098294,
      "loss": 2.7307,
      "theoretical_loss": 3.780359425139562,
      "tokens_seen": 698351616
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003981745235707122,
      "loss": 2.9755,
      "theoretical_loss": 3.7803233903757527,
      "tokens_seen": 698417152
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003981644934804413,
      "loss": 2.8236,
      "theoretical_loss": 3.780287359939778,
      "tokens_seen": 698482688
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039815446339017054,
      "loss": 3.1551,
      "theoretical_loss": 3.7802513338307113,
      "tokens_seen": 698548224
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039814443329989966,
      "loss": 2.8064,
      "theoretical_loss": 3.780215312047628,
      "tokens_seen": 698613760
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003981344032096289,
      "loss": 3.0448,
      "theoretical_loss": 3.780179294589601,
      "tokens_seen": 698679296
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003981243731193581,
      "loss": 2.8214,
      "theoretical_loss": 3.7801432814557083,
      "tokens_seen": 698744832
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039811434302908726,
      "loss": 2.7705,
      "theoretical_loss": 3.780107272645023,
      "tokens_seen": 698810368
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039810431293881644,
      "loss": 2.7764,
      "theoretical_loss": 3.780071268156621,
      "tokens_seen": 698875904
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003980942828485457,
      "loss": 2.9018,
      "theoretical_loss": 3.7800352679895797,
      "tokens_seen": 698941440
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003980842527582748,
      "loss": 2.9174,
      "theoretical_loss": 3.779999272142973,
      "tokens_seen": 699006976
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039807422266800404,
      "loss": 3.0942,
      "theoretical_loss": 3.7799632806158794,
      "tokens_seen": 699072512
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039806419257773317,
      "loss": 3.003,
      "theoretical_loss": 3.779927293407375,
      "tokens_seen": 699138048
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003980541624874624,
      "loss": 3.0289,
      "theoretical_loss": 3.779891310516537,
      "tokens_seen": 699203584
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003980441323971916,
      "loss": 2.662,
      "theoretical_loss": 3.7798553319424424,
      "tokens_seen": 699269120
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039803410230692077,
      "loss": 2.6216,
      "theoretical_loss": 3.779819357684169,
      "tokens_seen": 699334656
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039802407221664995,
      "loss": 2.851,
      "theoretical_loss": 3.7797833877407947,
      "tokens_seen": 699400192
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039801404212637913,
      "loss": 2.7083,
      "theoretical_loss": 3.779747422111398,
      "tokens_seen": 699465728
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003980040120361083,
      "loss": 2.993,
      "theoretical_loss": 3.779711460795057,
      "tokens_seen": 699531264
    },
    {
      "epoch": 0.21,
      "objective/train/advantage_avg": 0.4941639006137848,
      "objective/train/docs_used": 405172,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.7227683067321777,
      "objective/train/original_loss": 2.722768545150757,
      "objective/train/theoretical_loss": 3.779675503790851,
      "objective/train/tokens_used": 720056800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24425342679023743,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050658106803894,
      "objective/train/weighted_lm_loss": 2.8605382442474365,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 3.779675503790851,
      "tokens_seen": 699596800
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039799398194583754,
      "loss": 3.048,
      "theoretical_loss": 3.779675503790851,
      "tokens_seen": 699596800
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039798395185556667,
      "loss": 3.2275,
      "theoretical_loss": 3.7796395510978584,
      "tokens_seen": 699662336
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003979739217652959,
      "loss": 3.1092,
      "theoretical_loss": 3.7796036027151585,
      "tokens_seen": 699727872
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039796389167502503,
      "loss": 3.0092,
      "theoretical_loss": 3.779567658641832,
      "tokens_seen": 699793408
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039795386158475427,
      "loss": 2.6591,
      "theoretical_loss": 3.7795317188769575,
      "tokens_seen": 699858944
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039794383149448345,
      "loss": 2.9323,
      "theoretical_loss": 3.779495783419616,
      "tokens_seen": 699924480
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039793380140421263,
      "loss": 2.892,
      "theoretical_loss": 3.779459852268888,
      "tokens_seen": 699990016
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003979237713139418,
      "loss": 2.8799,
      "theoretical_loss": 3.7794239254238535,
      "tokens_seen": 700055552
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039791374122367105,
      "loss": 2.8302,
      "theoretical_loss": 3.779388002883594,
      "tokens_seen": 700121088
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003979037111334002,
      "loss": 2.8272,
      "theoretical_loss": 3.7793520846471913,
      "tokens_seen": 700186624
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003978936810431294,
      "loss": 2.8984,
      "theoretical_loss": 3.7793161707137264,
      "tokens_seen": 700252160
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003978836509528586,
      "loss": 3.0177,
      "theoretical_loss": 3.779280261082281,
      "tokens_seen": 700317696
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039787362086258777,
      "loss": 2.9876,
      "theoretical_loss": 3.7792443557519375,
      "tokens_seen": 700383232
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.000397863590772317,
      "loss": 2.8806,
      "theoretical_loss": 3.779208454721779,
      "tokens_seen": 700448768
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039785356068204613,
      "loss": 2.8486,
      "theoretical_loss": 3.7791725579908872,
      "tokens_seen": 700514304
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039784353059177537,
      "loss": 3.088,
      "theoretical_loss": 3.7791366655583456,
      "tokens_seen": 700579840
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003978335005015045,
      "loss": 2.9548,
      "theoretical_loss": 3.7791007774232375,
      "tokens_seen": 700645376
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039782347041123373,
      "loss": 3.1788,
      "theoretical_loss": 3.7790648935846463,
      "tokens_seen": 700710912
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003978134403209629,
      "loss": 3.1297,
      "theoretical_loss": 3.7790290140416554,
      "tokens_seen": 700776448
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003978034102306921,
      "loss": 2.8892,
      "theoretical_loss": 3.7789931387933504,
      "tokens_seen": 700841984
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003977933801404213,
      "loss": 3.0043,
      "theoretical_loss": 3.7789572678388144,
      "tokens_seen": 700907520
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039778335005015046,
      "loss": 2.9293,
      "theoretical_loss": 3.778921401177133,
      "tokens_seen": 700973056
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039777331995987964,
      "loss": 2.9898,
      "theoretical_loss": 3.7788855388073896,
      "tokens_seen": 701038592
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039776328986960887,
      "loss": 2.8531,
      "theoretical_loss": 3.778849680728671,
      "tokens_seen": 701104128
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.000397753259779338,
      "loss": 3.1458,
      "theoretical_loss": 3.7788138269400626,
      "tokens_seen": 701169664
    },
    {
      "epoch": 0.21,
      "objective/train/advantage_avg": 0.4940187633037567,
      "objective/train/docs_used": 406046,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.619896411895752,
      "objective/train/original_loss": 2.6198959350585938,
      "objective/train/theoretical_loss": 3.778777977440649,
      "objective/train/tokens_used": 721695200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2441263049840927,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506428480148315,
      "objective/train/weighted_lm_loss": 2.752436399459839,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0396549701690674,
      "theoretical_loss": 3.778777977440649,
      "tokens_seen": 701235200
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039774322968906723,
      "loss": 2.795,
      "theoretical_loss": 3.778777977440649,
      "tokens_seen": 701235200
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003977331995987964,
      "loss": 3.0324,
      "theoretical_loss": 3.7787421322295183,
      "tokens_seen": 701300736
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003977231695085256,
      "loss": 2.7308,
      "theoretical_loss": 3.778706291305755,
      "tokens_seen": 701366272
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003977131394182548,
      "loss": 2.9587,
      "theoretical_loss": 3.778670454668447,
      "tokens_seen": 701431808
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039770310932798396,
      "loss": 2.6221,
      "theoretical_loss": 3.7786346223166802,
      "tokens_seen": 701497344
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039769307923771314,
      "loss": 2.6506,
      "theoretical_loss": 3.7785987942495423,
      "tokens_seen": 701562880
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003976830491474424,
      "loss": 3.2199,
      "theoretical_loss": 3.7785629704661217,
      "tokens_seen": 701628416
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003976730190571715,
      "loss": 2.9174,
      "theoretical_loss": 3.7785271509655045,
      "tokens_seen": 701693952
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039766298896690074,
      "loss": 2.9253,
      "theoretical_loss": 3.7784913357467795,
      "tokens_seen": 701759488
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039765295887662987,
      "loss": 2.8952,
      "theoretical_loss": 3.7784555248090355,
      "tokens_seen": 701825024
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003976429287863591,
      "loss": 2.8747,
      "theoretical_loss": 3.77841971815136,
      "tokens_seen": 701890560
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003976328986960883,
      "loss": 2.8349,
      "theoretical_loss": 3.778383915772843,
      "tokens_seen": 701956096
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039762286860581746,
      "loss": 2.8496,
      "theoretical_loss": 3.7783481176725737,
      "tokens_seen": 702021632
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039761283851554664,
      "loss": 2.8751,
      "theoretical_loss": 3.778312323849641,
      "tokens_seen": 702087168
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003976028084252759,
      "loss": 3.0456,
      "theoretical_loss": 3.778276534303134,
      "tokens_seen": 702152704
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.000397592778335005,
      "loss": 3.0764,
      "theoretical_loss": 3.778240749032144,
      "tokens_seen": 702218240
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039758274824473424,
      "loss": 2.7821,
      "theoretical_loss": 3.778204968035761,
      "tokens_seen": 702283776
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039757271815446337,
      "loss": 2.6909,
      "theoretical_loss": 3.7781691913130753,
      "tokens_seen": 702349312
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003975626880641926,
      "loss": 2.8734,
      "theoretical_loss": 3.7781334188631774,
      "tokens_seen": 702414848
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003975526579739218,
      "loss": 3.0193,
      "theoretical_loss": 3.778097650685159,
      "tokens_seen": 702480384
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039754262788365097,
      "loss": 2.8043,
      "theoretical_loss": 3.778061886778111,
      "tokens_seen": 702545920
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039753259779338015,
      "loss": 3.0218,
      "theoretical_loss": 3.7780261271411253,
      "tokens_seen": 702611456
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039752256770310933,
      "loss": 2.858,
      "theoretical_loss": 3.777990371773294,
      "tokens_seen": 702676992
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003975125376128385,
      "loss": 3.2169,
      "theoretical_loss": 3.777954620673709,
      "tokens_seen": 702742528
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039750250752256774,
      "loss": 2.9191,
      "theoretical_loss": 3.7779188738414637,
      "tokens_seen": 702808064
    },
    {
      "epoch": 0.21,
      "objective/train/advantage_avg": 0.49361443519592285,
      "objective/train/docs_used": 406800,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.8369140625,
      "objective/train/original_loss": 2.836914300918579,
      "objective/train/theoretical_loss": 3.77788313127565,
      "objective/train/tokens_used": 723333600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24372799694538116,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506004095077515,
      "objective/train/weighted_lm_loss": 2.9805541038513184,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0352214574813843,
      "theoretical_loss": 3.77788313127565,
      "tokens_seen": 702873600
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039749247743229687,
      "loss": 2.9326,
      "theoretical_loss": 3.77788313127565,
      "tokens_seen": 702873600
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003974824473420261,
      "loss": 3.216,
      "theoretical_loss": 3.7778473929753607,
      "tokens_seen": 702939136
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039747241725175523,
      "loss": 3.0512,
      "theoretical_loss": 3.77781165893969,
      "tokens_seen": 703004672
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039746238716148447,
      "loss": 2.8244,
      "theoretical_loss": 3.7777759291677313,
      "tokens_seen": 703070208
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039745235707121365,
      "loss": 2.9572,
      "theoretical_loss": 3.7777402036585777,
      "tokens_seen": 703135744
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039744232698094283,
      "loss": 3.004,
      "theoretical_loss": 3.7777044824113246,
      "tokens_seen": 703201280
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.000397432296890672,
      "loss": 3.0776,
      "theoretical_loss": 3.777668765425066,
      "tokens_seen": 703266816
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039742226680040125,
      "loss": 2.9468,
      "theoretical_loss": 3.7776330526988957,
      "tokens_seen": 703332352
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003974122367101304,
      "loss": 3.0638,
      "theoretical_loss": 3.7775973442319097,
      "tokens_seen": 703397888
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003974022066198596,
      "loss": 3.1873,
      "theoretical_loss": 3.7775616400232037,
      "tokens_seen": 703463424
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039739217652958874,
      "loss": 3.1271,
      "theoretical_loss": 3.777525940071872,
      "tokens_seen": 703528960
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039738214643931797,
      "loss": 2.7484,
      "theoretical_loss": 3.7774902443770113,
      "tokens_seen": 703594496
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039737211634904715,
      "loss": 2.9402,
      "theoretical_loss": 3.7774545529377175,
      "tokens_seen": 703660032
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039736208625877633,
      "loss": 2.8293,
      "theoretical_loss": 3.777418865753087,
      "tokens_seen": 703725568
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003973520561685055,
      "loss": 2.9043,
      "theoretical_loss": 3.7773831828222164,
      "tokens_seen": 703791104
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003973420260782347,
      "loss": 2.8828,
      "theoretical_loss": 3.7773475041442026,
      "tokens_seen": 703856640
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003973319959879639,
      "loss": 3.1706,
      "theoretical_loss": 3.7773118297181427,
      "tokens_seen": 703922176
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003973219658976931,
      "loss": 2.9023,
      "theoretical_loss": 3.777276159543135,
      "tokens_seen": 703987712
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039731193580742224,
      "loss": 2.8626,
      "theoretical_loss": 3.777240493618276,
      "tokens_seen": 704053248
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003973019057171515,
      "loss": 3.0366,
      "theoretical_loss": 3.7772048319426643,
      "tokens_seen": 704118784
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003972918756268806,
      "loss": 2.9017,
      "theoretical_loss": 3.7771691745153984,
      "tokens_seen": 704184320
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039728184553660984,
      "loss": 3.0198,
      "theoretical_loss": 3.7771335213355766,
      "tokens_seen": 704249856
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.000397271815446339,
      "loss": 2.8735,
      "theoretical_loss": 3.7770978724022983,
      "tokens_seen": 704315392
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003972617853560682,
      "loss": 2.9379,
      "theoretical_loss": 3.777062227714662,
      "tokens_seen": 704380928
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003972517552657974,
      "loss": 2.8786,
      "theoretical_loss": 3.7770265872717674,
      "tokens_seen": 704446464
    },
    {
      "epoch": 0.21,
      "objective/train/advantage_avg": 0.49446728825569153,
      "objective/train/docs_used": 408057,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.8180177211761475,
      "objective/train/original_loss": 2.8180174827575684,
      "objective/train/theoretical_loss": 3.7769909510727144,
      "objective/train/tokens_used": 724972000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.244566410779953,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050689935684204,
      "objective/train/weighted_lm_loss": 2.9607088565826416,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0454468727111816,
      "theoretical_loss": 3.7769909510727144,
      "tokens_seen": 704512000
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003972417251755266,
      "loss": 2.9553,
      "theoretical_loss": 3.7769909510727144,
      "tokens_seen": 704512000
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039723169508525574,
      "loss": 3.1744,
      "theoretical_loss": 3.7769553191166025,
      "tokens_seen": 704577536
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.000397221664994985,
      "loss": 3.2831,
      "theoretical_loss": 3.776919691402532,
      "tokens_seen": 704643072
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003972116349047141,
      "loss": 2.9017,
      "theoretical_loss": 3.776884067929604,
      "tokens_seen": 704708608
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039720160481444334,
      "loss": 3.1279,
      "theoretical_loss": 3.776848448696919,
      "tokens_seen": 704774144
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003971915747241725,
      "loss": 2.9494,
      "theoretical_loss": 3.776812833703578,
      "tokens_seen": 704839680
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003971815446339017,
      "loss": 3.118,
      "theoretical_loss": 3.776777222948682,
      "tokens_seen": 704905216
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003971715145436309,
      "loss": 3.1599,
      "theoretical_loss": 3.776741616431332,
      "tokens_seen": 704970752
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039716148445336007,
      "loss": 2.6417,
      "theoretical_loss": 3.776706014150632,
      "tokens_seen": 705036288
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039715145436308925,
      "loss": 3.3306,
      "theoretical_loss": 3.7766704161056825,
      "tokens_seen": 705101824
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003971414242728185,
      "loss": 3.1167,
      "theoretical_loss": 3.776634822295587,
      "tokens_seen": 705167360
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039713139418254766,
      "loss": 3.1888,
      "theoretical_loss": 3.7765992327194464,
      "tokens_seen": 705232896
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039712136409227684,
      "loss": 2.9273,
      "theoretical_loss": 3.7765636473763653,
      "tokens_seen": 705298432
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003971113340020061,
      "loss": 3.1201,
      "theoretical_loss": 3.776528066265447,
      "tokens_seen": 705363968
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003971013039117352,
      "loss": 3.0177,
      "theoretical_loss": 3.776492489385794,
      "tokens_seen": 705429504
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039709127382146444,
      "loss": 2.8953,
      "theoretical_loss": 3.7764569167365103,
      "tokens_seen": 705495040
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039708124373119357,
      "loss": 2.8664,
      "theoretical_loss": 3.776421348316701,
      "tokens_seen": 705560576
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003970712136409228,
      "loss": 3.157,
      "theoretical_loss": 3.7763857841254698,
      "tokens_seen": 705626112
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.000397061183550652,
      "loss": 2.6864,
      "theoretical_loss": 3.7763502241619205,
      "tokens_seen": 705691648
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039705115346038117,
      "loss": 3.1319,
      "theoretical_loss": 3.776314668425159,
      "tokens_seen": 705757184
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039704112337011035,
      "loss": 2.9725,
      "theoretical_loss": 3.7762791169142904,
      "tokens_seen": 705822720
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039703109327983953,
      "loss": 2.8399,
      "theoretical_loss": 3.77624356962842,
      "tokens_seen": 705888256
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003970210631895687,
      "loss": 2.9846,
      "theoretical_loss": 3.7762080265666533,
      "tokens_seen": 705953792
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039701103309929794,
      "loss": 2.8795,
      "theoretical_loss": 3.7761724877280964,
      "tokens_seen": 706019328
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039700100300902707,
      "loss": 2.9377,
      "theoretical_loss": 3.776136953111856,
      "tokens_seen": 706084864
    },
    {
      "epoch": 0.21,
      "objective/train/advantage_avg": 0.49481701850891113,
      "objective/train/docs_used": 408602,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.3117763996124268,
      "objective/train/original_loss": 3.3117761611938477,
      "objective/train/theoretical_loss": 3.7761014227170375,
      "objective/train/tokens_used": 726610400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24491290748119354,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507267713546753,
      "objective/train/weighted_lm_loss": 3.479923725128174,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0396549701690674,
      "theoretical_loss": 3.7761014227170375,
      "tokens_seen": 706150400
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003969909729187563,
      "loss": 3.1492,
      "theoretical_loss": 3.7761014227170375,
      "tokens_seen": 706150400
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039698094282848543,
      "loss": 2.9807,
      "theoretical_loss": 3.776065896542749,
      "tokens_seen": 706215936
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039697091273821467,
      "loss": 3.2569,
      "theoretical_loss": 3.7760303745880965,
      "tokens_seen": 706281472
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039696088264794385,
      "loss": 3.1486,
      "theoretical_loss": 3.775994856852188,
      "tokens_seen": 706347008
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039695085255767303,
      "loss": 3.1389,
      "theoretical_loss": 3.775959343334131,
      "tokens_seen": 706412544
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003969408224674022,
      "loss": 2.9094,
      "theoretical_loss": 3.7759238340330334,
      "tokens_seen": 706478080
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039693079237713145,
      "loss": 3.0123,
      "theoretical_loss": 3.775888328948003,
      "tokens_seen": 706543616
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003969207622868606,
      "loss": 2.7804,
      "theoretical_loss": 3.7758528280781487,
      "tokens_seen": 706609152
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003969107321965898,
      "loss": 2.9969,
      "theoretical_loss": 3.7758173314225787,
      "tokens_seen": 706674688
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039690070210631894,
      "loss": 3.0314,
      "theoretical_loss": 3.7757818389804023,
      "tokens_seen": 706740224
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039689067201604817,
      "loss": 2.8708,
      "theoretical_loss": 3.7757463507507287,
      "tokens_seen": 706805760
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039688064192577735,
      "loss": 2.7084,
      "theoretical_loss": 3.775710866732668,
      "tokens_seen": 706871296
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039687061183550653,
      "loss": 3.1004,
      "theoretical_loss": 3.775675386925329,
      "tokens_seen": 706936832
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003968605817452357,
      "loss": 3.0879,
      "theoretical_loss": 3.775639911327821,
      "tokens_seen": 707002368
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003968505516549649,
      "loss": 2.9553,
      "theoretical_loss": 3.775604439939257,
      "tokens_seen": 707067904
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003968405215646941,
      "loss": 3.065,
      "theoretical_loss": 3.7755689727587454,
      "tokens_seen": 707133440
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003968304914744233,
      "loss": 3.0789,
      "theoretical_loss": 3.775533509785398,
      "tokens_seen": 707198976
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039682046138415244,
      "loss": 2.8449,
      "theoretical_loss": 3.7754980510183254,
      "tokens_seen": 707264512
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003968104312938817,
      "loss": 2.8662,
      "theoretical_loss": 3.775462596456639,
      "tokens_seen": 707330048
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003968004012036108,
      "loss": 2.8844,
      "theoretical_loss": 3.7754271460994513,
      "tokens_seen": 707395584
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039679037111334004,
      "loss": 3.1325,
      "theoretical_loss": 3.775391699945873,
      "tokens_seen": 707461120
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003967803410230692,
      "loss": 3.0783,
      "theoretical_loss": 3.7753562579950177,
      "tokens_seen": 707526656
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003967703109327984,
      "loss": 2.9381,
      "theoretical_loss": 3.7753208202459962,
      "tokens_seen": 707592192
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003967602808425276,
      "loss": 2.737,
      "theoretical_loss": 3.7752853866979232,
      "tokens_seen": 707657728
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003967502507522568,
      "loss": 3.1227,
      "theoretical_loss": 3.77524995734991,
      "tokens_seen": 707723264
    },
    {
      "epoch": 0.21,
      "objective/train/advantage_avg": 0.49507468938827515,
      "objective/train/docs_used": 409727,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.657501697540283,
      "objective/train/original_loss": 2.657501459121704,
      "objective/train/theoretical_loss": 3.775214532201071,
      "objective/train/tokens_used": 728248800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2451799064874649,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507538318634033,
      "objective/train/weighted_lm_loss": 2.7921736240386963,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0414615869522095,
      "theoretical_loss": 3.775214532201071,
      "tokens_seen": 707788800
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039674022066198594,
      "loss": 3.0382,
      "theoretical_loss": 3.775214532201071,
      "tokens_seen": 707788800
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003967301905717152,
      "loss": 3.1342,
      "theoretical_loss": 3.7751791112505195,
      "tokens_seen": 707854336
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003967201604814443,
      "loss": 3.215,
      "theoretical_loss": 3.775143694497369,
      "tokens_seen": 707919872
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039671013039117354,
      "loss": 2.9862,
      "theoretical_loss": 3.7751082819407333,
      "tokens_seen": 707985408
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003967001003009027,
      "loss": 3.0396,
      "theoretical_loss": 3.775072873579728,
      "tokens_seen": 708050944
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003966900702106319,
      "loss": 2.9148,
      "theoretical_loss": 3.7750374694134665,
      "tokens_seen": 708116480
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003966800401203611,
      "loss": 3.069,
      "theoretical_loss": 3.7750020694410646,
      "tokens_seen": 708182016
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039667001003009027,
      "loss": 2.9938,
      "theoretical_loss": 3.774966673661637,
      "tokens_seen": 708247552
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039665997993981945,
      "loss": 2.9248,
      "theoretical_loss": 3.7749312820742995,
      "tokens_seen": 708313088
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003966499498495487,
      "loss": 3.0412,
      "theoretical_loss": 3.774895894678167,
      "tokens_seen": 708378624
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003966399197592778,
      "loss": 2.9267,
      "theoretical_loss": 3.7748605114723564,
      "tokens_seen": 708444160
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039662988966900704,
      "loss": 3.0585,
      "theoretical_loss": 3.7748251324559834,
      "tokens_seen": 708509696
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039661985957873617,
      "loss": 2.9482,
      "theoretical_loss": 3.774789757628165,
      "tokens_seen": 708575232
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003966098294884654,
      "loss": 3.0225,
      "theoretical_loss": 3.7747543869880174,
      "tokens_seen": 708640768
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003965997993981946,
      "loss": 3.0768,
      "theoretical_loss": 3.7747190205346577,
      "tokens_seen": 708706304
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039658976930792377,
      "loss": 2.9514,
      "theoretical_loss": 3.7746836582672034,
      "tokens_seen": 708771840
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039657973921765295,
      "loss": 2.8571,
      "theoretical_loss": 3.774648300184772,
      "tokens_seen": 708837376
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003965697091273822,
      "loss": 2.9091,
      "theoretical_loss": 3.7746129462864824,
      "tokens_seen": 708902912
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003965596790371113,
      "loss": 2.9561,
      "theoretical_loss": 3.7745775965714508,
      "tokens_seen": 708968448
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039654964894684055,
      "loss": 2.8175,
      "theoretical_loss": 3.774542251038797,
      "tokens_seen": 709033984
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003965396188565697,
      "loss": 2.8398,
      "theoretical_loss": 3.7745069096876396,
      "tokens_seen": 709099520
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003965295887662989,
      "loss": 3.1686,
      "theoretical_loss": 3.7744715725170965,
      "tokens_seen": 709165056
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003965195586760281,
      "loss": 2.8996,
      "theoretical_loss": 3.7744362395262883,
      "tokens_seen": 709230592
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039650952858575727,
      "loss": 2.989,
      "theoretical_loss": 3.774400910714333,
      "tokens_seen": 709296128
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039649949849548645,
      "loss": 2.9444,
      "theoretical_loss": 3.7743655860803513,
      "tokens_seen": 709361664
    },
    {
      "epoch": 0.21,
      "objective/train/advantage_avg": 0.49464160203933716,
      "objective/train/docs_used": 410190,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.7605221271514893,
      "objective/train/original_loss": 2.7605221271514893,
      "objective/train/theoretical_loss": 3.7743302656234627,
      "objective/train/tokens_used": 729887200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24473272264003754,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507081747055054,
      "objective/train/weighted_lm_loss": 2.900460720062256,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0445106029510498,
      "theoretical_loss": 3.7743302656234627,
      "tokens_seen": 709427200
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.00039648946840521563,
      "loss": 2.9697,
      "theoretical_loss": 3.7743302656234627,
      "tokens_seen": 709427200
    },
    {
      "epoch": 0.21,
      "learning_rate": 0.0003964794383149448,
      "loss": 3.1355,
      "theoretical_loss": 3.774294949342788,
      "tokens_seen": 709492736
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039646940822467405,
      "loss": 2.8568,
      "theoretical_loss": 3.7742596372374475,
      "tokens_seen": 709558272
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003964593781344032,
      "loss": 2.9635,
      "theoretical_loss": 3.774224329306562,
      "tokens_seen": 709623808
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003964493480441324,
      "loss": 3.0105,
      "theoretical_loss": 3.774189025549252,
      "tokens_seen": 709689344
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003964393179538616,
      "loss": 3.0036,
      "theoretical_loss": 3.7741537259646396,
      "tokens_seen": 709754880
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003964292878635908,
      "loss": 3.0997,
      "theoretical_loss": 3.7741184305518454,
      "tokens_seen": 709820416
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039641925777331996,
      "loss": 2.8604,
      "theoretical_loss": 3.774083139309993,
      "tokens_seen": 709885952
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039640922768304914,
      "loss": 2.8593,
      "theoretical_loss": 3.774047852238203,
      "tokens_seen": 709951488
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003963991975927783,
      "loss": 3.2988,
      "theoretical_loss": 3.774012569335598,
      "tokens_seen": 710017024
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039638916750250755,
      "loss": 2.9184,
      "theoretical_loss": 3.7739772906013007,
      "tokens_seen": 710082560
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039637913741223673,
      "loss": 2.6946,
      "theoretical_loss": 3.773942016034435,
      "tokens_seen": 710148096
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003963691073219659,
      "loss": 3.1838,
      "theoretical_loss": 3.7739067456341227,
      "tokens_seen": 710213632
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003963590772316951,
      "loss": 3.2428,
      "theoretical_loss": 3.773871479399488,
      "tokens_seen": 710279168
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003963490471414243,
      "loss": 2.8686,
      "theoretical_loss": 3.7738362173296545,
      "tokens_seen": 710344704
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003963390170511535,
      "loss": 3.1798,
      "theoretical_loss": 3.7738009594237463,
      "tokens_seen": 710410240
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039632898696088264,
      "loss": 3.0043,
      "theoretical_loss": 3.773765705680887,
      "tokens_seen": 710475776
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003963189568706119,
      "loss": 3.1843,
      "theoretical_loss": 3.7737304561002016,
      "tokens_seen": 710541312
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.000396308926780341,
      "loss": 2.862,
      "theoretical_loss": 3.7736952106808155,
      "tokens_seen": 710606848
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039629889669007024,
      "loss": 3.1384,
      "theoretical_loss": 3.7736599694218524,
      "tokens_seen": 710672384
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003962888665997994,
      "loss": 2.719,
      "theoretical_loss": 3.773624732322439,
      "tokens_seen": 710737920
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003962788365095286,
      "loss": 3.2367,
      "theoretical_loss": 3.7735894993817,
      "tokens_seen": 710803456
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003962688064192578,
      "loss": 2.9958,
      "theoretical_loss": 3.773554270598761,
      "tokens_seen": 710868992
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.000396258776328987,
      "loss": 2.9842,
      "theoretical_loss": 3.7735190459727486,
      "tokens_seen": 710934528
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039624874623871614,
      "loss": 2.8437,
      "theoretical_loss": 3.7734838255027894,
      "tokens_seen": 711000064
    },
    {
      "epoch": 0.22,
      "objective/train/advantage_avg": 0.4931311309337616,
      "objective/train/docs_used": 411414,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.7333645820617676,
      "objective/train/original_loss": 2.7333643436431885,
      "objective/train/theoretical_loss": 3.7734486091880095,
      "objective/train/tokens_used": 731525600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24331076443195343,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505499839782715,
      "objective/train/weighted_lm_loss": 2.8715388774871826,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0304546356201172,
      "theoretical_loss": 3.7734486091880095,
      "tokens_seen": 711065600
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003962387161484454,
      "loss": 3.1533,
      "theoretical_loss": 3.7734486091880095,
      "tokens_seen": 711065600
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003962286860581745,
      "loss": 2.9737,
      "theoretical_loss": 3.773413397027536,
      "tokens_seen": 711131136
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039621865596790374,
      "loss": 3.2658,
      "theoretical_loss": 3.773378189020496,
      "tokens_seen": 711196672
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003962086258776329,
      "loss": 2.918,
      "theoretical_loss": 3.7733429851660167,
      "tokens_seen": 711262208
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003961985957873621,
      "loss": 2.9397,
      "theoretical_loss": 3.773307785463226,
      "tokens_seen": 711327744
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003961885656970913,
      "loss": 2.9184,
      "theoretical_loss": 3.7732725899112527,
      "tokens_seen": 711393280
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039617853560682047,
      "loss": 2.8558,
      "theoretical_loss": 3.773237398509223,
      "tokens_seen": 711458816
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039616850551654965,
      "loss": 2.9226,
      "theoretical_loss": 3.773202211256267,
      "tokens_seen": 711524352
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003961584754262789,
      "loss": 3.015,
      "theoretical_loss": 3.773167028151514,
      "tokens_seen": 711589888
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.000396148445336008,
      "loss": 2.8359,
      "theoretical_loss": 3.773131849194091,
      "tokens_seen": 711655424
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039613841524573724,
      "loss": 2.8805,
      "theoretical_loss": 3.7730966743831282,
      "tokens_seen": 711720960
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039612838515546637,
      "loss": 3.0906,
      "theoretical_loss": 3.7730615037177557,
      "tokens_seen": 711786496
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003961183550651956,
      "loss": 3.1151,
      "theoretical_loss": 3.773026337197102,
      "tokens_seen": 711852032
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003961083249749248,
      "loss": 3.155,
      "theoretical_loss": 3.7729911748202984,
      "tokens_seen": 711917568
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039609829488465397,
      "loss": 2.909,
      "theoretical_loss": 3.7729560165864746,
      "tokens_seen": 711983104
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039608826479438315,
      "loss": 3.0667,
      "theoretical_loss": 3.7729208624947614,
      "tokens_seen": 712048640
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003960782347041124,
      "loss": 2.9626,
      "theoretical_loss": 3.7728857125442894,
      "tokens_seen": 712114176
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003960682046138415,
      "loss": 2.9199,
      "theoretical_loss": 3.77285056673419,
      "tokens_seen": 712179712
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039605817452357075,
      "loss": 2.9957,
      "theoretical_loss": 3.772815425063594,
      "tokens_seen": 712245248
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003960481444332999,
      "loss": 3.0635,
      "theoretical_loss": 3.772780287531634,
      "tokens_seen": 712310784
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003960381143430291,
      "loss": 3.0745,
      "theoretical_loss": 3.772745154137441,
      "tokens_seen": 712376320
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003960280842527583,
      "loss": 3.1785,
      "theoretical_loss": 3.7727100248801473,
      "tokens_seen": 712441856
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039601805416248747,
      "loss": 2.8782,
      "theoretical_loss": 3.7726748997588855,
      "tokens_seen": 712507392
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039600802407221665,
      "loss": 3.2718,
      "theoretical_loss": 3.7726397787727883,
      "tokens_seen": 712572928
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039599799398194583,
      "loss": 3.0842,
      "theoretical_loss": 3.772604661920989,
      "tokens_seen": 712638464
    },
    {
      "epoch": 0.22,
      "objective/train/advantage_avg": 0.4948796033859253,
      "objective/train/docs_used": 412148,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.8695976734161377,
      "objective/train/original_loss": 2.8695976734161377,
      "objective/train/theoretical_loss": 3.77256954920262,
      "objective/train/tokens_used": 733164000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24495829641819,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050733208656311,
      "objective/train/weighted_lm_loss": 3.015023708343506,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0437887907028198,
      "theoretical_loss": 3.77256954920262,
      "tokens_seen": 712704000
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.000395987963891675,
      "loss": 2.9771,
      "theoretical_loss": 3.77256954920262,
      "tokens_seen": 712704000
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039597793380140425,
      "loss": 3.0213,
      "theoretical_loss": 3.772534440616815,
      "tokens_seen": 712769536
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003959679037111334,
      "loss": 3.0517,
      "theoretical_loss": 3.7724993361627077,
      "tokens_seen": 712835072
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003959578736208626,
      "loss": 3.0847,
      "theoretical_loss": 3.7724642358394322,
      "tokens_seen": 712900608
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003959478435305918,
      "loss": 3.0682,
      "theoretical_loss": 3.772429139646123,
      "tokens_seen": 712966144
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.000395937813440321,
      "loss": 3.119,
      "theoretical_loss": 3.7723940475819147,
      "tokens_seen": 713031680
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039592778335005016,
      "loss": 3.2199,
      "theoretical_loss": 3.772358959645941,
      "tokens_seen": 713097216
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039591775325977934,
      "loss": 2.85,
      "theoretical_loss": 3.772323875837338,
      "tokens_seen": 713162752
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003959077231695085,
      "loss": 2.8837,
      "theoretical_loss": 3.7722887961552405,
      "tokens_seen": 713228288
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039589769307923775,
      "loss": 2.9183,
      "theoretical_loss": 3.7722537205987843,
      "tokens_seen": 713293824
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003958876629889669,
      "loss": 3.041,
      "theoretical_loss": 3.772218649167105,
      "tokens_seen": 713359360
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003958776328986961,
      "loss": 3.0866,
      "theoretical_loss": 3.7721835818593386,
      "tokens_seen": 713424896
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039586760280842524,
      "loss": 2.9889,
      "theoretical_loss": 3.772148518674622,
      "tokens_seen": 713490432
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003958575727181545,
      "loss": 2.9776,
      "theoretical_loss": 3.7721134596120907,
      "tokens_seen": 713555968
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039584754262788366,
      "loss": 3.0265,
      "theoretical_loss": 3.7720784046708826,
      "tokens_seen": 713621504
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039583751253761284,
      "loss": 3.0528,
      "theoretical_loss": 3.772043353850134,
      "tokens_seen": 713687040
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.000395827482447342,
      "loss": 2.9064,
      "theoretical_loss": 3.772008307148983,
      "tokens_seen": 713752576
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003958174523570712,
      "loss": 3.0233,
      "theoretical_loss": 3.771973264566567,
      "tokens_seen": 713818112
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003958074222668004,
      "loss": 2.9055,
      "theoretical_loss": 3.7719382261020233,
      "tokens_seen": 713883648
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003957973921765296,
      "loss": 3.2064,
      "theoretical_loss": 3.771903191754491,
      "tokens_seen": 713949184
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039578736208625875,
      "loss": 3.0264,
      "theoretical_loss": 3.7718681615231078,
      "tokens_seen": 714014720
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.000395777331995988,
      "loss": 2.9643,
      "theoretical_loss": 3.7718331354070127,
      "tokens_seen": 714080256
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039576730190571716,
      "loss": 3.0211,
      "theoretical_loss": 3.771798113405344,
      "tokens_seen": 714145792
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039575727181544634,
      "loss": 2.9233,
      "theoretical_loss": 3.771763095517242,
      "tokens_seen": 714211328
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003957472417251755,
      "loss": 3.0024,
      "theoretical_loss": 3.771728081741845,
      "tokens_seen": 714276864
    },
    {
      "epoch": 0.22,
      "objective/train/advantage_avg": 0.49529582262039185,
      "objective/train/docs_used": 413348,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.3452882766723633,
      "objective/train/original_loss": 2.3452885150909424,
      "objective/train/theoretical_loss": 3.7716930720782935,
      "objective/train/tokens_used": 734802400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24536541104316711,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507769584655762,
      "objective/train/weighted_lm_loss": 2.4643185138702393,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 3.7716930720782935,
      "tokens_seen": 714342400
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003957372116349047,
      "loss": 2.7786,
      "theoretical_loss": 3.7716930720782935,
      "tokens_seen": 714342400
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003957271815446339,
      "loss": 3.1619,
      "theoretical_loss": 3.7716580665257267,
      "tokens_seen": 714407936
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003957171514543631,
      "loss": 3.0389,
      "theoretical_loss": 3.7716230650832863,
      "tokens_seen": 714473472
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039570712136409225,
      "loss": 2.8312,
      "theoretical_loss": 3.771588067750111,
      "tokens_seen": 714539008
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003956970912738215,
      "loss": 2.8667,
      "theoretical_loss": 3.7715530745253423,
      "tokens_seen": 714604544
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003956870611835506,
      "loss": 2.9963,
      "theoretical_loss": 3.7715180854081214,
      "tokens_seen": 714670080
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039567703109327985,
      "loss": 3.0997,
      "theoretical_loss": 3.771483100397589,
      "tokens_seen": 714735616
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039566700100300903,
      "loss": 2.9119,
      "theoretical_loss": 3.7714481194928875,
      "tokens_seen": 714801152
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003956569709127382,
      "loss": 3.0435,
      "theoretical_loss": 3.7714131426931576,
      "tokens_seen": 714866688
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003956469408224674,
      "loss": 2.7021,
      "theoretical_loss": 3.771378169997542,
      "tokens_seen": 714932224
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039563691073219657,
      "loss": 3.1705,
      "theoretical_loss": 3.771343201405183,
      "tokens_seen": 714997760
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003956268806419258,
      "loss": 2.837,
      "theoretical_loss": 3.771308236915223,
      "tokens_seen": 715063296
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.000395616850551655,
      "loss": 3.1186,
      "theoretical_loss": 3.771273276526805,
      "tokens_seen": 715128832
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039560682046138417,
      "loss": 3.0693,
      "theoretical_loss": 3.771238320239071,
      "tokens_seen": 715194368
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039559679037111335,
      "loss": 2.9158,
      "theoretical_loss": 3.7712033680511654,
      "tokens_seen": 715259904
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003955867602808426,
      "loss": 3.0507,
      "theoretical_loss": 3.7711684199622324,
      "tokens_seen": 715325440
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003955767301905717,
      "loss": 3.0485,
      "theoretical_loss": 3.771133475971414,
      "tokens_seen": 715390976
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039556670010030095,
      "loss": 2.9319,
      "theoretical_loss": 3.7710985360778553,
      "tokens_seen": 715456512
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003955566700100301,
      "loss": 3.027,
      "theoretical_loss": 3.771063600280701,
      "tokens_seen": 715522048
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003955466399197593,
      "loss": 2.9574,
      "theoretical_loss": 3.7710286685790946,
      "tokens_seen": 715587584
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003955366098294885,
      "loss": 3.1587,
      "theoretical_loss": 3.770993740972182,
      "tokens_seen": 715653120
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039552657973921767,
      "loss": 3.1924,
      "theoretical_loss": 3.7709588174591087,
      "tokens_seen": 715718656
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039551654964894685,
      "loss": 2.7606,
      "theoretical_loss": 3.770923898039018,
      "tokens_seen": 715784192
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039550651955867603,
      "loss": 2.9907,
      "theoretical_loss": 3.7708889827110577,
      "tokens_seen": 715849728
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003954964894684052,
      "loss": 2.9125,
      "theoretical_loss": 3.770854071474372,
      "tokens_seen": 715915264
    },
    {
      "epoch": 0.22,
      "objective/train/advantage_avg": 0.4936705529689789,
      "objective/train/docs_used": 413910,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.6691460609436035,
      "objective/train/original_loss": 2.6691458225250244,
      "objective/train/theoretical_loss": 3.770819164328109,
      "objective/train/tokens_used": 736440800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24378420412540436,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050606369972229,
      "objective/train/weighted_lm_loss": 2.8042566776275635,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0338950157165527,
      "theoretical_loss": 3.770819164328109,
      "tokens_seen": 715980800
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039548645937813445,
      "loss": 3.0647,
      "theoretical_loss": 3.770819164328109,
      "tokens_seen": 715980800
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003954764292878636,
      "loss": 3.0679,
      "theoretical_loss": 3.7707842612714133,
      "tokens_seen": 716046336
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003954663991975928,
      "loss": 2.7555,
      "theoretical_loss": 3.770749362303432,
      "tokens_seen": 716111872
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.000395456369107322,
      "loss": 3.0342,
      "theoretical_loss": 3.770714467423313,
      "tokens_seen": 716177408
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003954463390170512,
      "loss": 2.8739,
      "theoretical_loss": 3.7706795766302017,
      "tokens_seen": 716242944
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039543630892678036,
      "loss": 2.8622,
      "theoretical_loss": 3.7706446899232473,
      "tokens_seen": 716308480
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039542627883650954,
      "loss": 3.3382,
      "theoretical_loss": 3.7706098073015966,
      "tokens_seen": 716374016
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003954162487462387,
      "loss": 2.774,
      "theoretical_loss": 3.7705749287643973,
      "tokens_seen": 716439552
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039540621865596795,
      "loss": 3.0487,
      "theoretical_loss": 3.770540054310798,
      "tokens_seen": 716505088
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003953961885656971,
      "loss": 3.3227,
      "theoretical_loss": 3.770505183939947,
      "tokens_seen": 716570624
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003953861584754263,
      "loss": 2.8614,
      "theoretical_loss": 3.770470317650993,
      "tokens_seen": 716636160
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039537612838515544,
      "loss": 2.7883,
      "theoretical_loss": 3.770435455443085,
      "tokens_seen": 716701696
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003953660982948847,
      "loss": 2.9947,
      "theoretical_loss": 3.7704005973153722,
      "tokens_seen": 716767232
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039535606820461386,
      "loss": 3.0938,
      "theoretical_loss": 3.7703657432670044,
      "tokens_seen": 716832768
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039534603811434304,
      "loss": 2.7527,
      "theoretical_loss": 3.770330893297131,
      "tokens_seen": 716898304
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003953360080240722,
      "loss": 2.8973,
      "theoretical_loss": 3.7702960474049014,
      "tokens_seen": 716963840
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003953259779338014,
      "loss": 3.5192,
      "theoretical_loss": 3.770261205589467,
      "tokens_seen": 717029376
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003953159478435306,
      "loss": 3.0497,
      "theoretical_loss": 3.7702263678499772,
      "tokens_seen": 717094912
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003953059177532598,
      "loss": 2.8511,
      "theoretical_loss": 3.7701915341855834,
      "tokens_seen": 717160448
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039529588766298895,
      "loss": 3.0583,
      "theoretical_loss": 3.7701567045954367,
      "tokens_seen": 717225984
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003952858575727182,
      "loss": 3.0561,
      "theoretical_loss": 3.7701218790786877,
      "tokens_seen": 717291520
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039527582748244736,
      "loss": 2.6853,
      "theoretical_loss": 3.7700870576344885,
      "tokens_seen": 717357056
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039526579739217654,
      "loss": 3.0925,
      "theoretical_loss": 3.7700522402619905,
      "tokens_seen": 717422592
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003952557673019057,
      "loss": 3.1195,
      "theoretical_loss": 3.7700174269603464,
      "tokens_seen": 717488128
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003952457372116349,
      "loss": 3.0989,
      "theoretical_loss": 3.769982617728707,
      "tokens_seen": 717553664
    },
    {
      "epoch": 0.22,
      "objective/train/advantage_avg": 0.49376043677330017,
      "objective/train/docs_used": 414484,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.5716609954833984,
      "objective/train/original_loss": 2.5716609954833984,
      "objective/train/theoretical_loss": 3.769947812566226,
      "objective/train/tokens_used": 738079200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24389223754405975,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050615906715393,
      "objective/train/weighted_lm_loss": 2.7017173767089844,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0363596677780151,
      "theoretical_loss": 3.769947812566226,
      "tokens_seen": 717619200
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003952357071213641,
      "loss": 2.7725,
      "theoretical_loss": 3.769947812566226,
      "tokens_seen": 717619200
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003952256770310933,
      "loss": 2.9475,
      "theoretical_loss": 3.769913011472056,
      "tokens_seen": 717684736
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039521564694082245,
      "loss": 2.685,
      "theoretical_loss": 3.76987821444535,
      "tokens_seen": 717750272
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003952056168505517,
      "loss": 2.9524,
      "theoretical_loss": 3.769843421485261,
      "tokens_seen": 717815808
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003951955867602808,
      "loss": 2.7426,
      "theoretical_loss": 3.7698086325909435,
      "tokens_seen": 717881344
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039518555667001005,
      "loss": 2.9948,
      "theoretical_loss": 3.7697738477615497,
      "tokens_seen": 717946880
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039517552657973923,
      "loss": 3.2029,
      "theoretical_loss": 3.7697390669962347,
      "tokens_seen": 718012416
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003951654964894684,
      "loss": 3.1092,
      "theoretical_loss": 3.769704290294152,
      "tokens_seen": 718077952
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003951554663991976,
      "loss": 2.7242,
      "theoretical_loss": 3.7696695176544575,
      "tokens_seen": 718143488
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039514543630892677,
      "loss": 2.8079,
      "theoretical_loss": 3.769634749076305,
      "tokens_seen": 718209024
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039513540621865595,
      "loss": 2.8976,
      "theoretical_loss": 3.76959998455885,
      "tokens_seen": 718274560
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003951253761283852,
      "loss": 2.7574,
      "theoretical_loss": 3.769565224101247,
      "tokens_seen": 718340096
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003951153460381143,
      "loss": 2.8282,
      "theoretical_loss": 3.769530467702652,
      "tokens_seen": 718405632
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039510531594784355,
      "loss": 3.1984,
      "theoretical_loss": 3.769495715362222,
      "tokens_seen": 718471168
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039509528585757273,
      "loss": 2.8441,
      "theoretical_loss": 3.769460967079111,
      "tokens_seen": 718536704
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003950852557673019,
      "loss": 3.1224,
      "theoretical_loss": 3.769426222852477,
      "tokens_seen": 718602240
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003950752256770311,
      "loss": 2.9186,
      "theoretical_loss": 3.769391482681475,
      "tokens_seen": 718667776
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003950651955867603,
      "loss": 2.9626,
      "theoretical_loss": 3.769356746565263,
      "tokens_seen": 718733312
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039505516549648946,
      "loss": 3.0478,
      "theoretical_loss": 3.7693220145029986,
      "tokens_seen": 718798848
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003950451354062187,
      "loss": 2.82,
      "theoretical_loss": 3.7692872864938374,
      "tokens_seen": 718864384
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003950351053159478,
      "loss": 2.8818,
      "theoretical_loss": 3.769252562536938,
      "tokens_seen": 718929920
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039502507522567705,
      "loss": 2.9978,
      "theoretical_loss": 3.769217842631458,
      "tokens_seen": 718995456
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003950150451354062,
      "loss": 3.0126,
      "theoretical_loss": 3.7691831267765563,
      "tokens_seen": 719060992
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003950050150451354,
      "loss": 3.203,
      "theoretical_loss": 3.76914841497139,
      "tokens_seen": 719126528
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003949949849548646,
      "loss": 2.8235,
      "theoretical_loss": 3.7691137072151184,
      "tokens_seen": 719192064
    },
    {
      "epoch": 0.22,
      "objective/train/advantage_avg": 0.49371081590652466,
      "objective/train/docs_used": 415542,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.9893569946289062,
      "objective/train/original_loss": 2.989356517791748,
      "objective/train/theoretical_loss": 3.7690790035069,
      "objective/train/tokens_used": 739717600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24381569027900696,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506105422973633,
      "objective/train/weighted_lm_loss": 3.1405725479125977,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0396549701690674,
      "theoretical_loss": 3.7690790035069,
      "tokens_seen": 719257600
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003949849548645938,
      "loss": 3.0886,
      "theoretical_loss": 3.7690790035069,
      "tokens_seen": 719257600
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039497492477432296,
      "loss": 3.1642,
      "theoretical_loss": 3.7690443038458943,
      "tokens_seen": 719323136
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003949648946840522,
      "loss": 2.9351,
      "theoretical_loss": 3.7690096082312596,
      "tokens_seen": 719388672
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003949548645937813,
      "loss": 3.0128,
      "theoretical_loss": 3.768974916662157,
      "tokens_seen": 719454208
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039494483450351056,
      "loss": 3.0557,
      "theoretical_loss": 3.768940229137745,
      "tokens_seen": 719519744
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003949348044132397,
      "loss": 2.9796,
      "theoretical_loss": 3.768905545657185,
      "tokens_seen": 719585280
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003949247743229689,
      "loss": 2.8506,
      "theoretical_loss": 3.768870866219636,
      "tokens_seen": 719650816
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003949147442326981,
      "loss": 2.8446,
      "theoretical_loss": 3.76883619082426,
      "tokens_seen": 719716352
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003949047141424273,
      "loss": 3.0748,
      "theoretical_loss": 3.768801519470216,
      "tokens_seen": 719781888
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003948946840521565,
      "loss": 3.1506,
      "theoretical_loss": 3.7687668521566664,
      "tokens_seen": 719847424
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039488465396188564,
      "loss": 3.0706,
      "theoretical_loss": 3.768732188882773,
      "tokens_seen": 719912960
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003948746238716149,
      "loss": 2.9639,
      "theoretical_loss": 3.768697529647696,
      "tokens_seen": 719978496
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039486459378134406,
      "loss": 2.9126,
      "theoretical_loss": 3.768662874450598,
      "tokens_seen": 720044032
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039485456369107324,
      "loss": 3.0247,
      "theoretical_loss": 3.7686282232906416,
      "tokens_seen": 720109568
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003948445336008024,
      "loss": 2.8857,
      "theoretical_loss": 3.768593576166988,
      "tokens_seen": 720175104
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003948345035105316,
      "loss": 3.0296,
      "theoretical_loss": 3.7685589330788005,
      "tokens_seen": 720240640
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003948244734202608,
      "loss": 2.8627,
      "theoretical_loss": 3.768524294025242,
      "tokens_seen": 720306176
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039481444332999,
      "loss": 3.0792,
      "theoretical_loss": 3.7684896590054757,
      "tokens_seen": 720371712
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039480441323971915,
      "loss": 2.8517,
      "theoretical_loss": 3.768455028018664,
      "tokens_seen": 720437248
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003947943831494484,
      "loss": 2.791,
      "theoretical_loss": 3.7684204010639717,
      "tokens_seen": 720502784
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039478435305917756,
      "loss": 2.9918,
      "theoretical_loss": 3.768385778140562,
      "tokens_seen": 720568320
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039477432296890674,
      "loss": 3.2451,
      "theoretical_loss": 3.768351159247599,
      "tokens_seen": 720633856
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003947642928786359,
      "loss": 3.1186,
      "theoretical_loss": 3.768316544384247,
      "tokens_seen": 720699392
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003947542627883651,
      "loss": 3.1822,
      "theoretical_loss": 3.768281933549671,
      "tokens_seen": 720764928
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003947442326980943,
      "loss": 3.0625,
      "theoretical_loss": 3.7682473267430354,
      "tokens_seen": 720830464
    },
    {
      "epoch": 0.22,
      "objective/train/advantage_avg": 0.49443137645721436,
      "objective/train/docs_used": 416165,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.906836748123169,
      "objective/train/original_loss": 2.9068360328674316,
      "objective/train/theoretical_loss": 3.7682127239635053,
      "objective/train/tokens_used": 741356000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2445361316204071,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050686240196228,
      "objective/train/weighted_lm_loss": 3.0539839267730713,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0442860126495361,
      "theoretical_loss": 3.7682127239635053,
      "tokens_seen": 720896000
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003947342026078235,
      "loss": 2.9807,
      "theoretical_loss": 3.7682127239635053,
      "tokens_seen": 720896000
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039472417251755265,
      "loss": 3.1004,
      "theoretical_loss": 3.7681781252102464,
      "tokens_seen": 720961536
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003947141424272819,
      "loss": 2.759,
      "theoretical_loss": 3.768143530482424,
      "tokens_seen": 721027072
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.000394704112337011,
      "loss": 2.9698,
      "theoretical_loss": 3.7681089397792045,
      "tokens_seen": 721092608
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039469408224674025,
      "loss": 3.387,
      "theoretical_loss": 3.7680743530997525,
      "tokens_seen": 721158144
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039468405215646943,
      "loss": 2.8283,
      "theoretical_loss": 3.7680397704432362,
      "tokens_seen": 721223680
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003946740220661986,
      "loss": 2.8576,
      "theoretical_loss": 3.768005191808821,
      "tokens_seen": 721289216
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003946639919759278,
      "loss": 3.1086,
      "theoretical_loss": 3.767970617195674,
      "tokens_seen": 721354752
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039465396188565697,
      "loss": 3.166,
      "theoretical_loss": 3.767936046602963,
      "tokens_seen": 721420288
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039464393179538615,
      "loss": 2.9466,
      "theoretical_loss": 3.7679014800298543,
      "tokens_seen": 721485824
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003946339017051154,
      "loss": 3.1736,
      "theoretical_loss": 3.7678669174755153,
      "tokens_seen": 721551360
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003946238716148445,
      "loss": 3.0309,
      "theoretical_loss": 3.767832358939115,
      "tokens_seen": 721616896
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039461384152457375,
      "loss": 3.0526,
      "theoretical_loss": 3.7677978044198204,
      "tokens_seen": 721682432
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039460381143430293,
      "loss": 3.1846,
      "theoretical_loss": 3.7677632539168004,
      "tokens_seen": 721747968
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003945937813440321,
      "loss": 3.0098,
      "theoretical_loss": 3.7677287074292236,
      "tokens_seen": 721813504
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003945837512537613,
      "loss": 3.052,
      "theoretical_loss": 3.7676941649562585,
      "tokens_seen": 721879040
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003945737211634905,
      "loss": 2.9813,
      "theoretical_loss": 3.7676596264970743,
      "tokens_seen": 721944576
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039456369107321966,
      "loss": 2.7899,
      "theoretical_loss": 3.7676250920508405,
      "tokens_seen": 722010112
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003945536609829489,
      "loss": 3.0267,
      "theoretical_loss": 3.7675905616167262,
      "tokens_seen": 722075648
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.000394543630892678,
      "loss": 2.9454,
      "theoretical_loss": 3.7675560351939015,
      "tokens_seen": 722141184
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039453360080240725,
      "loss": 3.0548,
      "theoretical_loss": 3.7675215127815367,
      "tokens_seen": 722206720
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003945235707121364,
      "loss": 2.9805,
      "theoretical_loss": 3.7674869943788014,
      "tokens_seen": 722272256
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003945135406218656,
      "loss": 2.9878,
      "theoretical_loss": 3.767452479984867,
      "tokens_seen": 722337792
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003945035105315948,
      "loss": 2.9484,
      "theoretical_loss": 3.7674179695989034,
      "tokens_seen": 722403328
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.000394493480441324,
      "loss": 2.9088,
      "theoretical_loss": 3.7673834632200824,
      "tokens_seen": 722468864
    },
    {
      "epoch": 0.22,
      "objective/train/advantage_avg": 0.4927060306072235,
      "objective/train/docs_used": 417606,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.249758243560791,
      "objective/train/original_loss": 3.249758243560791,
      "objective/train/theoretical_loss": 3.767348960847575,
      "objective/train/tokens_used": 742994400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24285829067230225,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505051612854004,
      "objective/train/weighted_lm_loss": 3.4138081073760986,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0435127019882202,
      "theoretical_loss": 3.767348960847575,
      "tokens_seen": 722534400
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039448345035105316,
      "loss": 3.0236,
      "theoretical_loss": 3.767348960847575,
      "tokens_seen": 722534400
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003944734202607824,
      "loss": 2.8221,
      "theoretical_loss": 3.7673144624805524,
      "tokens_seen": 722599936
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003944633901705115,
      "loss": 2.8168,
      "theoretical_loss": 3.767279968118187,
      "tokens_seen": 722665472
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039445336008024076,
      "loss": 2.928,
      "theoretical_loss": 3.7672454777596505,
      "tokens_seen": 722731008
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003944433299899699,
      "loss": 2.8259,
      "theoretical_loss": 3.7672109914041156,
      "tokens_seen": 722796544
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003944332998996991,
      "loss": 2.9479,
      "theoretical_loss": 3.7671765090507536,
      "tokens_seen": 722862080
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003944232698094283,
      "loss": 2.9603,
      "theoretical_loss": 3.7671420306987384,
      "tokens_seen": 722927616
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003944132397191575,
      "loss": 3.0277,
      "theoretical_loss": 3.7671075563472427,
      "tokens_seen": 722993152
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039440320962888666,
      "loss": 2.9786,
      "theoretical_loss": 3.76707308599544,
      "tokens_seen": 723058688
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039439317953861584,
      "loss": 3.0346,
      "theoretical_loss": 3.767038619642503,
      "tokens_seen": 723124224
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.000394383149448345,
      "loss": 3.3259,
      "theoretical_loss": 3.767004157287606,
      "tokens_seen": 723189760
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039437311935807426,
      "loss": 3.0296,
      "theoretical_loss": 3.7669696989299233,
      "tokens_seen": 723255296
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003943630892678034,
      "loss": 3.2074,
      "theoretical_loss": 3.7669352445686286,
      "tokens_seen": 723320832
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003943530591775326,
      "loss": 3.0498,
      "theoretical_loss": 3.7669007942028965,
      "tokens_seen": 723386368
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039434302908726175,
      "loss": 3.1294,
      "theoretical_loss": 3.766866347831902,
      "tokens_seen": 723451904
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.000394332998996991,
      "loss": 2.791,
      "theoretical_loss": 3.76683190545482,
      "tokens_seen": 723517440
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039432296890672017,
      "loss": 2.7146,
      "theoretical_loss": 3.766797467070825,
      "tokens_seen": 723582976
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039431293881644935,
      "loss": 3.0766,
      "theoretical_loss": 3.766763032679094,
      "tokens_seen": 723648512
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039430290872617853,
      "loss": 3.0565,
      "theoretical_loss": 3.7667286022788007,
      "tokens_seen": 723714048
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039429287863590776,
      "loss": 2.9314,
      "theoretical_loss": 3.766694175869123,
      "tokens_seen": 723779584
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003942828485456369,
      "loss": 2.7478,
      "theoretical_loss": 3.766659753449236,
      "tokens_seen": 723845120
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003942728184553661,
      "loss": 3.1343,
      "theoretical_loss": 3.7666253350183165,
      "tokens_seen": 723910656
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039426278836509525,
      "loss": 3.2269,
      "theoretical_loss": 3.7665909205755406,
      "tokens_seen": 723976192
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003942527582748245,
      "loss": 2.9351,
      "theoretical_loss": 3.766556510120086,
      "tokens_seen": 724041728
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039424272818455367,
      "loss": 3.1033,
      "theoretical_loss": 3.7665221036511296,
      "tokens_seen": 724107264
    },
    {
      "epoch": 0.22,
      "objective/train/advantage_avg": 0.49262040853500366,
      "objective/train/docs_used": 418215,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.928281307220459,
      "objective/train/original_loss": 2.928281545639038,
      "objective/train/theoretical_loss": 3.7664877011678484,
      "objective/train/tokens_used": 744632800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24278001487255096,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0504961013793945,
      "objective/train/weighted_lm_loss": 3.0761337280273438,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0304546356201172,
      "theoretical_loss": 3.7664877011678484,
      "tokens_seen": 724172800
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039423269809428285,
      "loss": 3.0832,
      "theoretical_loss": 3.7664877011678484,
      "tokens_seen": 724172800
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039422266800401203,
      "loss": 3.079,
      "theoretical_loss": 3.7664533026694205,
      "tokens_seen": 724238336
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003942126379137412,
      "loss": 2.9526,
      "theoretical_loss": 3.766418908155024,
      "tokens_seen": 724303872
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003942026078234704,
      "loss": 2.8717,
      "theoretical_loss": 3.7663845176238366,
      "tokens_seen": 724369408
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039419257773319963,
      "loss": 2.8744,
      "theoretical_loss": 3.766350131075037,
      "tokens_seen": 724434944
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039418254764292876,
      "loss": 2.8529,
      "theoretical_loss": 3.7663157485078047,
      "tokens_seen": 724500480
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.000394172517552658,
      "loss": 2.8326,
      "theoretical_loss": 3.766281369921316,
      "tokens_seen": 724566016
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003941624874623871,
      "loss": 2.93,
      "theoretical_loss": 3.766246995314753,
      "tokens_seen": 724631552
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039415245737211635,
      "loss": 2.8855,
      "theoretical_loss": 3.7662126246872925,
      "tokens_seen": 724697088
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003941424272818456,
      "loss": 2.9735,
      "theoretical_loss": 3.7661782580381162,
      "tokens_seen": 724762624
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003941323971915747,
      "loss": 2.4609,
      "theoretical_loss": 3.766143895366403,
      "tokens_seen": 724828160
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039412236710130395,
      "loss": 2.9404,
      "theoretical_loss": 3.7661095366713324,
      "tokens_seen": 724893696
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039411233701103313,
      "loss": 2.8119,
      "theoretical_loss": 3.766075181952086,
      "tokens_seen": 724959232
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003941023069207623,
      "loss": 3.0985,
      "theoretical_loss": 3.766040831207844,
      "tokens_seen": 725024768
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003940922768304915,
      "loss": 2.9598,
      "theoretical_loss": 3.7660064844377867,
      "tokens_seen": 725090304
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003940822467402207,
      "loss": 2.8742,
      "theoretical_loss": 3.765972141641096,
      "tokens_seen": 725155840
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039407221664994986,
      "loss": 3.0806,
      "theoretical_loss": 3.765937802816952,
      "tokens_seen": 725221376
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003940621865596791,
      "loss": 2.9629,
      "theoretical_loss": 3.7659034679645376,
      "tokens_seen": 725286912
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003940521564694082,
      "loss": 3.0153,
      "theoretical_loss": 3.7658691370830337,
      "tokens_seen": 725352448
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039404212637913745,
      "loss": 2.7319,
      "theoretical_loss": 3.7658348101716226,
      "tokens_seen": 725417984
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003940320962888666,
      "loss": 2.7567,
      "theoretical_loss": 3.7658004872294866,
      "tokens_seen": 725483520
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003940220661985958,
      "loss": 2.6467,
      "theoretical_loss": 3.7657661682558086,
      "tokens_seen": 725549056
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.000394012036108325,
      "loss": 3.011,
      "theoretical_loss": 3.765731853249771,
      "tokens_seen": 725614592
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003940020060180542,
      "loss": 2.9743,
      "theoretical_loss": 3.7656975422105563,
      "tokens_seen": 725680128
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039399197592778336,
      "loss": 2.7913,
      "theoretical_loss": 3.7656632351373487,
      "tokens_seen": 725745664
    },
    {
      "epoch": 0.22,
      "objective/train/advantage_avg": 0.49337127804756165,
      "objective/train/docs_used": 419550,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.16666316986084,
      "objective/train/original_loss": 3.1666626930236816,
      "objective/train/theoretical_loss": 3.7656289320293315,
      "objective/train/tokens_used": 746271200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24349823594093323,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050574779510498,
      "objective/train/weighted_lm_loss": 3.3268091678619385,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0425469875335693,
      "theoretical_loss": 3.7656289320293315,
      "tokens_seen": 725811200
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003939819458375126,
      "loss": 3.1815,
      "theoretical_loss": 3.7656289320293315,
      "tokens_seen": 725811200
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003939719157472417,
      "loss": 3.052,
      "theoretical_loss": 3.7655946328856875,
      "tokens_seen": 725876736
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039396188565697096,
      "loss": 2.8363,
      "theoretical_loss": 3.765560337705602,
      "tokens_seen": 725942272
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003939518555667001,
      "loss": 3.0207,
      "theoretical_loss": 3.765526046488258,
      "tokens_seen": 726007808
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003939418254764293,
      "loss": 3.0118,
      "theoretical_loss": 3.7654917592328414,
      "tokens_seen": 726073344
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003939317953861585,
      "loss": 3.199,
      "theoretical_loss": 3.7654574759385353,
      "tokens_seen": 726138880
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003939217652958877,
      "loss": 2.8363,
      "theoretical_loss": 3.7654231966045257,
      "tokens_seen": 726204416
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039391173520561686,
      "loss": 3.2152,
      "theoretical_loss": 3.765388921229998,
      "tokens_seen": 726269952
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039390170511534604,
      "loss": 2.9912,
      "theoretical_loss": 3.7653546498141366,
      "tokens_seen": 726335488
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003938916750250752,
      "loss": 2.7468,
      "theoretical_loss": 3.765320382356128,
      "tokens_seen": 726401024
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039388164493480446,
      "loss": 2.854,
      "theoretical_loss": 3.7652861188551574,
      "tokens_seen": 726466560
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003938716148445336,
      "loss": 2.8619,
      "theoretical_loss": 3.7652518593104114,
      "tokens_seen": 726532096
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003938615847542628,
      "loss": 2.9589,
      "theoretical_loss": 3.765217603721077,
      "tokens_seen": 726597632
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039385155466399195,
      "loss": 2.9754,
      "theoretical_loss": 3.7651833520863396,
      "tokens_seen": 726663168
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003938415245737212,
      "loss": 2.6623,
      "theoretical_loss": 3.7651491044053866,
      "tokens_seen": 726728704
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039383149448345037,
      "loss": 2.9424,
      "theoretical_loss": 3.7651148606774054,
      "tokens_seen": 726794240
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039382146439317955,
      "loss": 3.138,
      "theoretical_loss": 3.7650806209015837,
      "tokens_seen": 726859776
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039381143430290873,
      "loss": 2.8184,
      "theoretical_loss": 3.765046385077108,
      "tokens_seen": 726925312
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039380140421263796,
      "loss": 2.964,
      "theoretical_loss": 3.765012153203166,
      "tokens_seen": 726990848
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003937913741223671,
      "loss": 2.8015,
      "theoretical_loss": 3.764977925278947,
      "tokens_seen": 727056384
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003937813440320963,
      "loss": 2.767,
      "theoretical_loss": 3.764943701303639,
      "tokens_seen": 727121920
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039377131394182545,
      "loss": 3.0344,
      "theoretical_loss": 3.7649094812764297,
      "tokens_seen": 727187456
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003937612838515547,
      "loss": 2.7437,
      "theoretical_loss": 3.764875265196509,
      "tokens_seen": 727252992
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039375125376128387,
      "loss": 3.1144,
      "theoretical_loss": 3.7648410530630656,
      "tokens_seen": 727318528
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039374122367101305,
      "loss": 2.6827,
      "theoretical_loss": 3.764806844875288,
      "tokens_seen": 727384064
    },
    {
      "epoch": 0.22,
      "objective/train/advantage_avg": 0.4956057369709015,
      "objective/train/docs_used": 420047,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.3709402084350586,
      "objective/train/original_loss": 2.3709397315979004,
      "objective/train/theoretical_loss": 3.7647726406323665,
      "objective/train/tokens_used": 747909600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2456897348165512,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050809621810913,
      "objective/train/weighted_lm_loss": 2.491211414337158,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0408108234405518,
      "theoretical_loss": 3.7647726406323665,
      "tokens_seen": 727449600
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039373119358074223,
      "loss": 2.9549,
      "theoretical_loss": 3.7647726406323665,
      "tokens_seen": 727449600
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003937211634904714,
      "loss": 3.0184,
      "theoretical_loss": 3.764738440333491,
      "tokens_seen": 727515136
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003937111334002006,
      "loss": 2.7677,
      "theoretical_loss": 3.7647042439778504,
      "tokens_seen": 727580672
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039370110330992983,
      "loss": 2.717,
      "theoretical_loss": 3.7646700515646363,
      "tokens_seen": 727646208
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039369107321965896,
      "loss": 2.9353,
      "theoretical_loss": 3.7646358630930385,
      "tokens_seen": 727711744
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003936810431293882,
      "loss": 3.1099,
      "theoretical_loss": 3.764601678562248,
      "tokens_seen": 727777280
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003936710130391173,
      "loss": 2.7939,
      "theoretical_loss": 3.7645674979714547,
      "tokens_seen": 727842816
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039366098294884655,
      "loss": 2.9502,
      "theoretical_loss": 3.764533321319851,
      "tokens_seen": 727908352
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039365095285857573,
      "loss": 3.0922,
      "theoretical_loss": 3.764499148606628,
      "tokens_seen": 727973888
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003936409227683049,
      "loss": 2.9565,
      "theoretical_loss": 3.7644649798309775,
      "tokens_seen": 728039424
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003936308926780341,
      "loss": 3.1766,
      "theoretical_loss": 3.7644308149920906,
      "tokens_seen": 728104960
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039362086258776333,
      "loss": 3.0141,
      "theoretical_loss": 3.7643966540891602,
      "tokens_seen": 728170496
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039361083249749246,
      "loss": 3.1128,
      "theoretical_loss": 3.764362497121379,
      "tokens_seen": 728236032
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003936008024072217,
      "loss": 3.1955,
      "theoretical_loss": 3.7643283440879385,
      "tokens_seen": 728301568
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003935907723169508,
      "loss": 2.7199,
      "theoretical_loss": 3.764294194988033,
      "tokens_seen": 728367104
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039358074222668006,
      "loss": 3.0524,
      "theoretical_loss": 3.7642600498208543,
      "tokens_seen": 728432640
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039357071213640924,
      "loss": 3.053,
      "theoretical_loss": 3.764225908585596,
      "tokens_seen": 728498176
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003935606820461384,
      "loss": 3.1201,
      "theoretical_loss": 3.7641917712814523,
      "tokens_seen": 728563712
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003935506519558676,
      "loss": 3.1352,
      "theoretical_loss": 3.7641576379076165,
      "tokens_seen": 728629248
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003935406218655968,
      "loss": 2.8239,
      "theoretical_loss": 3.764123508463282,
      "tokens_seen": 728694784
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039353059177532596,
      "loss": 2.7369,
      "theoretical_loss": 3.7640893829476445,
      "tokens_seen": 728760320
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003935205616850552,
      "loss": 2.5236,
      "theoretical_loss": 3.7640552613598977,
      "tokens_seen": 728825856
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003935105315947843,
      "loss": 3.1445,
      "theoretical_loss": 3.764021143699237,
      "tokens_seen": 728891392
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039350050150451356,
      "loss": 3.2238,
      "theoretical_loss": 3.763987029964856,
      "tokens_seen": 728956928
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003934904714142427,
      "loss": 2.9903,
      "theoretical_loss": 3.763952920155951,
      "tokens_seen": 729022464
    },
    {
      "epoch": 0.22,
      "objective/train/advantage_avg": 0.4925635755062103,
      "objective/train/docs_used": 421294,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.935081958770752,
      "objective/train/original_loss": 2.935081958770752,
      "objective/train/theoretical_loss": 3.763918814271718,
      "objective/train/tokens_used": 749548000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2427338808774948,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050490140914917,
      "objective/train/weighted_lm_loss": 3.0830211639404297,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0396549701690674,
      "theoretical_loss": 3.763918814271718,
      "tokens_seen": 729088000
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003934804413239719,
      "loss": 3.0173,
      "theoretical_loss": 3.763918814271718,
      "tokens_seen": 729088000
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003934704112337011,
      "loss": 3.0682,
      "theoretical_loss": 3.763884712311352,
      "tokens_seen": 729153536
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003934603811434303,
      "loss": 3.0177,
      "theoretical_loss": 3.763850614274048,
      "tokens_seen": 729219072
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039345035105315946,
      "loss": 2.9021,
      "theoretical_loss": 3.763816520159004,
      "tokens_seen": 729284608
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003934403209628887,
      "loss": 3.1655,
      "theoretical_loss": 3.7637824299654152,
      "tokens_seen": 729350144
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039343029087261783,
      "loss": 2.6313,
      "theoretical_loss": 3.763748343692479,
      "tokens_seen": 729415680
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039342026078234706,
      "loss": 2.9752,
      "theoretical_loss": 3.763714261339392,
      "tokens_seen": 729481216
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003934102306920762,
      "loss": 3.0098,
      "theoretical_loss": 3.7636801829053503,
      "tokens_seen": 729546752
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003934002006018054,
      "loss": 2.9008,
      "theoretical_loss": 3.763646108389553,
      "tokens_seen": 729612288
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039339017051153466,
      "loss": 2.9984,
      "theoretical_loss": 3.763612037791197,
      "tokens_seen": 729677824
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003933801404212638,
      "loss": 2.8519,
      "theoretical_loss": 3.7635779711094797,
      "tokens_seen": 729743360
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.000393370110330993,
      "loss": 2.9839,
      "theoretical_loss": 3.7635439083435998,
      "tokens_seen": 729808896
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039336008024072215,
      "loss": 2.9824,
      "theoretical_loss": 3.7635098494927552,
      "tokens_seen": 729874432
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003933500501504514,
      "loss": 2.8144,
      "theoretical_loss": 3.7634757945561446,
      "tokens_seen": 729939968
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039334002006018057,
      "loss": 2.9174,
      "theoretical_loss": 3.763441743532967,
      "tokens_seen": 730005504
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039332998996990975,
      "loss": 3.1982,
      "theoretical_loss": 3.763407696422421,
      "tokens_seen": 730071040
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039331995987963893,
      "loss": 2.8857,
      "theoretical_loss": 3.763373653223706,
      "tokens_seen": 730136576
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039330992978936816,
      "loss": 2.9875,
      "theoretical_loss": 3.7633396139360213,
      "tokens_seen": 730202112
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003932998996990973,
      "loss": 3.0093,
      "theoretical_loss": 3.763305578558567,
      "tokens_seen": 730267648
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003932898696088265,
      "loss": 3.0177,
      "theoretical_loss": 3.763271547090543,
      "tokens_seen": 730333184
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039327983951855565,
      "loss": 3.082,
      "theoretical_loss": 3.76323751953115,
      "tokens_seen": 730398720
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003932698094282849,
      "loss": 3.0651,
      "theoretical_loss": 3.763203495879587,
      "tokens_seen": 730464256
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039325977933801407,
      "loss": 3.1171,
      "theoretical_loss": 3.7631694761350563,
      "tokens_seen": 730529792
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039324974924774325,
      "loss": 2.836,
      "theoretical_loss": 3.7631354602967573,
      "tokens_seen": 730595328
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039323971915747243,
      "loss": 2.8825,
      "theoretical_loss": 3.7631014483638925,
      "tokens_seen": 730660864
    },
    {
      "epoch": 0.22,
      "objective/train/advantage_avg": 0.495294451713562,
      "objective/train/docs_used": 422072,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.745354413986206,
      "objective/train/original_loss": 2.745354413986206,
      "objective/train/theoretical_loss": 3.7630674403356625,
      "objective/train/tokens_used": 751186400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24538537859916687,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507769584655762,
      "objective/train/weighted_lm_loss": 2.8847432136535645,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0417574644088745,
      "theoretical_loss": 3.7630674403356625,
      "tokens_seen": 730726400
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003932296890672016,
      "loss": 2.9179,
      "theoretical_loss": 3.7630674403356625,
      "tokens_seen": 730726400
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003932196589769308,
      "loss": 2.816,
      "theoretical_loss": 3.763033436211269,
      "tokens_seen": 730791936
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039320962888666003,
      "loss": 3.1513,
      "theoretical_loss": 3.762999435989914,
      "tokens_seen": 730857472
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039319959879638916,
      "loss": 2.9365,
      "theoretical_loss": 3.7629654396707997,
      "tokens_seen": 730923008
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003931895687061184,
      "loss": 3.141,
      "theoretical_loss": 3.7629314472531274,
      "tokens_seen": 730988544
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003931795386158475,
      "loss": 2.8405,
      "theoretical_loss": 3.7628974587361013,
      "tokens_seen": 731054080
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039316950852557675,
      "loss": 3.3702,
      "theoretical_loss": 3.762863474118923,
      "tokens_seen": 731119616
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039315947843530593,
      "loss": 2.9217,
      "theoretical_loss": 3.7628294934007958,
      "tokens_seen": 731185152
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003931494483450351,
      "loss": 3.0933,
      "theoretical_loss": 3.7627955165809226,
      "tokens_seen": 731250688
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003931394182547643,
      "loss": 3.0064,
      "theoretical_loss": 3.762761543658508,
      "tokens_seen": 731316224
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039312938816449353,
      "loss": 2.9954,
      "theoretical_loss": 3.7627275746327546,
      "tokens_seen": 731381760
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039311935807422266,
      "loss": 3.2271,
      "theoretical_loss": 3.7626936095028674,
      "tokens_seen": 731447296
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003931093279839519,
      "loss": 2.942,
      "theoretical_loss": 3.762659648268049,
      "tokens_seen": 731512832
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.000393099297893681,
      "loss": 3.0716,
      "theoretical_loss": 3.7626256909275053,
      "tokens_seen": 731578368
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039308926780341026,
      "loss": 3.0694,
      "theoretical_loss": 3.7625917374804407,
      "tokens_seen": 731643904
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039307923771313944,
      "loss": 2.805,
      "theoretical_loss": 3.762557787926059,
      "tokens_seen": 731709440
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003930692076228686,
      "loss": 3.0824,
      "theoretical_loss": 3.7625238422635667,
      "tokens_seen": 731774976
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003930591775325978,
      "loss": 3.2168,
      "theoretical_loss": 3.7624899004921684,
      "tokens_seen": 731840512
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.000393049147442327,
      "loss": 2.9991,
      "theoretical_loss": 3.76245596261107,
      "tokens_seen": 731906048
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039303911735205616,
      "loss": 2.8919,
      "theoretical_loss": 3.7624220286194774,
      "tokens_seen": 731971584
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003930290872617854,
      "loss": 2.8952,
      "theoretical_loss": 3.7623880985165963,
      "tokens_seen": 732037120
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003930190571715145,
      "loss": 2.8357,
      "theoretical_loss": 3.7623541723016327,
      "tokens_seen": 732102656
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039300902708124376,
      "loss": 2.9811,
      "theoretical_loss": 3.7623202499737944,
      "tokens_seen": 732168192
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003929989969909729,
      "loss": 3.0293,
      "theoretical_loss": 3.7622863315322865,
      "tokens_seen": 732233728
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003929889669007021,
      "loss": 2.9301,
      "theoretical_loss": 3.762252416976317,
      "tokens_seen": 732299264
    },
    {
      "epoch": 0.22,
      "objective/train/advantage_avg": 0.49536776542663574,
      "objective/train/docs_used": 423458,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.3850042819976807,
      "objective/train/original_loss": 2.3850035667419434,
      "objective/train/theoretical_loss": 3.7622185063050937,
      "objective/train/tokens_used": 752824800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24544954299926758,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507845878601074,
      "objective/train/weighted_lm_loss": 2.506054639816284,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0417574644088745,
      "theoretical_loss": 3.7622185063050937,
      "tokens_seen": 732364800
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003929789368104313,
      "loss": 2.7131,
      "theoretical_loss": 3.7622185063050937,
      "tokens_seen": 732364800
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003929689067201605,
      "loss": 2.961,
      "theoretical_loss": 3.7621845995178225,
      "tokens_seen": 732430336
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039295887662988967,
      "loss": 3.0691,
      "theoretical_loss": 3.762150696613712,
      "tokens_seen": 732495872
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003929488465396189,
      "loss": 3.0854,
      "theoretical_loss": 3.76211679759197,
      "tokens_seen": 732561408
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039293881644934803,
      "loss": 2.9669,
      "theoretical_loss": 3.7620829024518048,
      "tokens_seen": 732626944
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039292878635907726,
      "loss": 3.0436,
      "theoretical_loss": 3.7620490111924245,
      "tokens_seen": 732692480
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003929187562688064,
      "loss": 2.841,
      "theoretical_loss": 3.762015123813038,
      "tokens_seen": 732758016
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003929087261785356,
      "loss": 2.8738,
      "theoretical_loss": 3.7619812403128536,
      "tokens_seen": 732823552
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003928986960882648,
      "loss": 3.0769,
      "theoretical_loss": 3.761947360691081,
      "tokens_seen": 732889088
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.000392888665997994,
      "loss": 2.9783,
      "theoretical_loss": 3.7619134849469296,
      "tokens_seen": 732954624
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039287863590772317,
      "loss": 3.1105,
      "theoretical_loss": 3.761879613079608,
      "tokens_seen": 733020160
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039286860581745235,
      "loss": 3.2091,
      "theoretical_loss": 3.761845745088327,
      "tokens_seen": 733085696
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039285857572718153,
      "loss": 3.1012,
      "theoretical_loss": 3.761811880972296,
      "tokens_seen": 733151232
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039284854563691077,
      "loss": 3.0678,
      "theoretical_loss": 3.7617780207307256,
      "tokens_seen": 733216768
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003928385155466399,
      "loss": 2.8921,
      "theoretical_loss": 3.7617441643628258,
      "tokens_seen": 733282304
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039282848545636913,
      "loss": 2.891,
      "theoretical_loss": 3.761710311867808,
      "tokens_seen": 733347840
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039281845536609825,
      "loss": 3.0565,
      "theoretical_loss": 3.761676463244883,
      "tokens_seen": 733413376
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003928084252758275,
      "loss": 2.9015,
      "theoretical_loss": 3.7616426184932608,
      "tokens_seen": 733478912
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039279839518555667,
      "loss": 3.0658,
      "theoretical_loss": 3.761608777612154,
      "tokens_seen": 733544448
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039278836509528585,
      "loss": 3.1956,
      "theoretical_loss": 3.7615749406007746,
      "tokens_seen": 733609984
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039277833500501503,
      "loss": 3.0816,
      "theoretical_loss": 3.761541107458333,
      "tokens_seen": 733675520
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039276830491474427,
      "loss": 3.0231,
      "theoretical_loss": 3.761507278184043,
      "tokens_seen": 733741056
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003927582748244734,
      "loss": 2.926,
      "theoretical_loss": 3.761473452777115,
      "tokens_seen": 733806592
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039274824473420263,
      "loss": 2.9507,
      "theoretical_loss": 3.7614396312367626,
      "tokens_seen": 733872128
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039273821464393176,
      "loss": 2.9435,
      "theoretical_loss": 3.761405813562199,
      "tokens_seen": 733937664
    },
    {
      "epoch": 0.22,
      "objective/train/advantage_avg": 0.49377527832984924,
      "objective/train/docs_used": 424116,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.1035444736480713,
      "objective/train/original_loss": 3.1035444736480713,
      "objective/train/theoretical_loss": 3.7613719997526367,
      "objective/train/tokens_used": 754463200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24390146136283875,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506173372268677,
      "objective/train/weighted_lm_loss": 3.2604095935821533,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0408108234405518,
      "theoretical_loss": 3.7613719997526367,
      "tokens_seen": 734003200
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.000392728184553661,
      "loss": 2.9635,
      "theoretical_loss": 3.7613719997526367,
      "tokens_seen": 734003200
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003927181544633902,
      "loss": 2.8931,
      "theoretical_loss": 3.7613381898072884,
      "tokens_seen": 734068736
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039270812437311936,
      "loss": 2.8517,
      "theoretical_loss": 3.761304383725369,
      "tokens_seen": 734134272
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039269809428284854,
      "loss": 2.9752,
      "theoretical_loss": 3.7612705815060905,
      "tokens_seen": 734199808
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003926880641925777,
      "loss": 2.9632,
      "theoretical_loss": 3.761236783148668,
      "tokens_seen": 734265344
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003926780341023069,
      "loss": 2.8718,
      "theoretical_loss": 3.7612029886523155,
      "tokens_seen": 734330880
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039266800401203613,
      "loss": 2.8815,
      "theoretical_loss": 3.761169198016247,
      "tokens_seen": 734396416
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039265797392176526,
      "loss": 3.0733,
      "theoretical_loss": 3.7611354112396773,
      "tokens_seen": 734461952
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003926479438314945,
      "loss": 2.9535,
      "theoretical_loss": 3.761101628321822,
      "tokens_seen": 734527488
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039263791374122373,
      "loss": 2.9941,
      "theoretical_loss": 3.7610678492618947,
      "tokens_seen": 734593024
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039262788365095286,
      "loss": 2.7362,
      "theoretical_loss": 3.761034074059112,
      "tokens_seen": 734658560
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003926178535606821,
      "loss": 2.9397,
      "theoretical_loss": 3.761000302712689,
      "tokens_seen": 734724096
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003926078234704112,
      "loss": 3.0866,
      "theoretical_loss": 3.760966535221841,
      "tokens_seen": 734789632
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039259779338014046,
      "loss": 2.6852,
      "theoretical_loss": 3.760932771585784,
      "tokens_seen": 734855168
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039258776328986964,
      "loss": 2.8705,
      "theoretical_loss": 3.760899011803735,
      "tokens_seen": 734920704
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003925777331995988,
      "loss": 2.8984,
      "theoretical_loss": 3.7608652558749105,
      "tokens_seen": 734986240
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.000392567703109328,
      "loss": 3.1519,
      "theoretical_loss": 3.760831503798527,
      "tokens_seen": 735051776
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003925576730190572,
      "loss": 2.9875,
      "theoretical_loss": 3.7607977555738006,
      "tokens_seen": 735117312
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039254764292878636,
      "loss": 3.0466,
      "theoretical_loss": 3.7607640111999494,
      "tokens_seen": 735182848
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003925376128385156,
      "loss": 2.8806,
      "theoretical_loss": 3.7607302706761896,
      "tokens_seen": 735248384
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003925275827482447,
      "loss": 2.7533,
      "theoretical_loss": 3.760696534001741,
      "tokens_seen": 735313920
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039251755265797396,
      "loss": 3.0923,
      "theoretical_loss": 3.760662801175819,
      "tokens_seen": 735379456
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003925075225677031,
      "loss": 3.0032,
      "theoretical_loss": 3.760629072197643,
      "tokens_seen": 735444992
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003924974924774323,
      "loss": 2.9853,
      "theoretical_loss": 3.760595347066431,
      "tokens_seen": 735510528
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003924874623871615,
      "loss": 2.9189,
      "theoretical_loss": 3.760561625781401,
      "tokens_seen": 735576064
    },
    {
      "epoch": 0.22,
      "objective/train/advantage_avg": 0.4939987361431122,
      "objective/train/docs_used": 424834,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.5579793453216553,
      "objective/train/original_loss": 2.5579795837402344,
      "objective/train/theoretical_loss": 3.760527908341773,
      "objective/train/tokens_used": 756101600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24409978091716766,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506408214569092,
      "objective/train/weighted_lm_loss": 2.6874098777770996,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0417574644088745,
      "theoretical_loss": 3.760527908341773,
      "tokens_seen": 735641600
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003924774322968907,
      "loss": 2.8173,
      "theoretical_loss": 3.760527908341773,
      "tokens_seen": 735641600
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039246740220661987,
      "loss": 2.8507,
      "theoretical_loss": 3.760494194746765,
      "tokens_seen": 735707136
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003924573721163491,
      "loss": 2.959,
      "theoretical_loss": 3.7604604849955967,
      "tokens_seen": 735772672
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039244734202607823,
      "loss": 2.8308,
      "theoretical_loss": 3.760426779087487,
      "tokens_seen": 735838208
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039243731193580746,
      "loss": 3.1457,
      "theoretical_loss": 3.760393077021656,
      "tokens_seen": 735903744
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003924272818455366,
      "loss": 2.9389,
      "theoretical_loss": 3.760359378797323,
      "tokens_seen": 735969280
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003924172517552658,
      "loss": 3.0297,
      "theoretical_loss": 3.7603256844137087,
      "tokens_seen": 736034816
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.000392407221664995,
      "loss": 3.0778,
      "theoretical_loss": 3.760291993870034,
      "tokens_seen": 736100352
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003923971915747242,
      "loss": 3.1042,
      "theoretical_loss": 3.760258307165518,
      "tokens_seen": 736165888
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039238716148445337,
      "loss": 3.1511,
      "theoretical_loss": 3.7602246242993824,
      "tokens_seen": 736231424
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039237713139418255,
      "loss": 3.1033,
      "theoretical_loss": 3.760190945270848,
      "tokens_seen": 736296960
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039236710130391173,
      "loss": 3.2024,
      "theoretical_loss": 3.7601572700791364,
      "tokens_seen": 736362496
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039235707121364097,
      "loss": 3.0985,
      "theoretical_loss": 3.7601235987234682,
      "tokens_seen": 736428032
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003923470411233701,
      "loss": 2.8538,
      "theoretical_loss": 3.7600899312030664,
      "tokens_seen": 736493568
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039233701103309933,
      "loss": 2.8167,
      "theoretical_loss": 3.760056267517152,
      "tokens_seen": 736559104
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039232698094282846,
      "loss": 2.8447,
      "theoretical_loss": 3.760022607664947,
      "tokens_seen": 736624640
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003923169508525577,
      "loss": 2.8929,
      "theoretical_loss": 3.7599889516456746,
      "tokens_seen": 736690176
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039230692076228687,
      "loss": 2.8076,
      "theoretical_loss": 3.759955299458557,
      "tokens_seen": 736755712
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039229689067201605,
      "loss": 2.988,
      "theoretical_loss": 3.7599216511028173,
      "tokens_seen": 736821248
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039228686058174523,
      "loss": 3.0884,
      "theoretical_loss": 3.7598880065776776,
      "tokens_seen": 736886784
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039227683049147447,
      "loss": 3.1579,
      "theoretical_loss": 3.7598543658823624,
      "tokens_seen": 736952320
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003922668004012036,
      "loss": 2.8706,
      "theoretical_loss": 3.7598207290160945,
      "tokens_seen": 737017856
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039225677031093283,
      "loss": 2.991,
      "theoretical_loss": 3.7597870959780977,
      "tokens_seen": 737083392
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039224674022066196,
      "loss": 3.0623,
      "theoretical_loss": 3.759753466767597,
      "tokens_seen": 737148928
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003922367101303912,
      "loss": 2.9826,
      "theoretical_loss": 3.759719841383815,
      "tokens_seen": 737214464
    },
    {
      "epoch": 0.22,
      "objective/train/advantage_avg": 0.4938200116157532,
      "objective/train/docs_used": 426204,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.3953068256378174,
      "objective/train/original_loss": 3.3953070640563965,
      "objective/train/theoretical_loss": 3.7596862198259773,
      "objective/train/tokens_used": 757740000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2439526617527008,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506221055984497,
      "objective/train/weighted_lm_loss": 3.5671422481536865,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0417574644088745,
      "theoretical_loss": 3.7596862198259773,
      "tokens_seen": 737280000
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003922266800401204,
      "loss": 3.2007,
      "theoretical_loss": 3.7596862198259773,
      "tokens_seen": 737280000
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039221664994984956,
      "loss": 2.926,
      "theoretical_loss": 3.7596526020933076,
      "tokens_seen": 737345536
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039220661985957874,
      "loss": 2.9908,
      "theoretical_loss": 3.7596189881850313,
      "tokens_seen": 737411072
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003921965897693079,
      "loss": 3.1251,
      "theoretical_loss": 3.7595853781003745,
      "tokens_seen": 737476608
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003921865596790371,
      "loss": 2.9955,
      "theoretical_loss": 3.7595517718385603,
      "tokens_seen": 737542144
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039217652958876633,
      "loss": 2.7761,
      "theoretical_loss": 3.7595181693988162,
      "tokens_seen": 737607680
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039216649949849546,
      "loss": 2.9485,
      "theoretical_loss": 3.7594845707803675,
      "tokens_seen": 737673216
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003921564694082247,
      "loss": 3.2207,
      "theoretical_loss": 3.7594509759824395,
      "tokens_seen": 737738752
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003921464393179539,
      "loss": 2.9291,
      "theoretical_loss": 3.7594173850042587,
      "tokens_seen": 737804288
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039213640922768306,
      "loss": 3.0312,
      "theoretical_loss": 3.7593837978450524,
      "tokens_seen": 737869824
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039212637913741224,
      "loss": 2.641,
      "theoretical_loss": 3.7593502145040465,
      "tokens_seen": 737935360
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003921163490471414,
      "loss": 2.9452,
      "theoretical_loss": 3.7593166349804683,
      "tokens_seen": 738000896
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003921063189568706,
      "loss": 2.9464,
      "theoretical_loss": 3.7592830592735442,
      "tokens_seen": 738066432
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039209628886659984,
      "loss": 3.2274,
      "theoretical_loss": 3.759249487382502,
      "tokens_seen": 738131968
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039208625877632896,
      "loss": 2.9339,
      "theoretical_loss": 3.7592159193065697,
      "tokens_seen": 738197504
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003920762286860582,
      "loss": 3.1189,
      "theoretical_loss": 3.7591823550449748,
      "tokens_seen": 738263040
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003920661985957873,
      "loss": 2.9585,
      "theoretical_loss": 3.7591487945969453,
      "tokens_seen": 738328576
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039205616850551656,
      "loss": 3.0182,
      "theoretical_loss": 3.759115237961709,
      "tokens_seen": 738394112
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039204613841524574,
      "loss": 2.8475,
      "theoretical_loss": 3.7590816851384954,
      "tokens_seen": 738459648
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003920361083249749,
      "loss": 2.9391,
      "theoretical_loss": 3.759048136126532,
      "tokens_seen": 738525184
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003920260782347041,
      "loss": 2.8896,
      "theoretical_loss": 3.7590145909250494,
      "tokens_seen": 738590720
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003920160481444333,
      "loss": 2.75,
      "theoretical_loss": 3.7589810495332747,
      "tokens_seen": 738656256
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039200601805416247,
      "loss": 2.8016,
      "theoretical_loss": 3.7589475119504394,
      "tokens_seen": 738721792
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003919959879638917,
      "loss": 3.0801,
      "theoretical_loss": 3.758913978175771,
      "tokens_seen": 738787328
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039198595787362083,
      "loss": 3.0015,
      "theoretical_loss": 3.758880448208501,
      "tokens_seen": 738852864
    },
    {
      "epoch": 0.22,
      "objective/train/advantage_avg": 0.4940301477909088,
      "objective/train/docs_used": 426747,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.7184813022613525,
      "objective/train/original_loss": 2.71848201751709,
      "objective/train/theoretical_loss": 3.7588469220478586,
      "objective/train/tokens_used": 759378400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24413254857063293,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050644040107727,
      "objective/train/weighted_lm_loss": 2.8560712337493896,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0304546356201172,
      "theoretical_loss": 3.7588469220478586,
      "tokens_seen": 738918400
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039197592778335007,
      "loss": 3.0987,
      "theoretical_loss": 3.7588469220478586,
      "tokens_seen": 738918400
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039196589769307925,
      "loss": 2.9003,
      "theoretical_loss": 3.758813399693074,
      "tokens_seen": 738983936
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039195586760280843,
      "loss": 2.6596,
      "theoretical_loss": 3.7587798811433784,
      "tokens_seen": 739049472
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003919458375125376,
      "loss": 2.822,
      "theoretical_loss": 3.758746366398002,
      "tokens_seen": 739115008
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003919358074222668,
      "loss": 2.9661,
      "theoretical_loss": 3.7587128554561753,
      "tokens_seen": 739180544
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039192577733199597,
      "loss": 2.718,
      "theoretical_loss": 3.758679348317131,
      "tokens_seen": 739246080
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003919157472417252,
      "loss": 3.0057,
      "theoretical_loss": 3.758645844980099,
      "tokens_seen": 739311616
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039190571715145433,
      "loss": 3.0321,
      "theoretical_loss": 3.7586123454443117,
      "tokens_seen": 739377152
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039189568706118357,
      "loss": 3.0404,
      "theoretical_loss": 3.758578849709,
      "tokens_seen": 739442688
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039188565697091275,
      "loss": 2.9591,
      "theoretical_loss": 3.7585453577733974,
      "tokens_seen": 739508224
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039187562688064193,
      "loss": 2.8998,
      "theoretical_loss": 3.7585118696367354,
      "tokens_seen": 739573760
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039186559679037117,
      "loss": 3.0995,
      "theoretical_loss": 3.758478385298246,
      "tokens_seen": 739639296
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003918555667001003,
      "loss": 2.7898,
      "theoretical_loss": 3.7584449047571633,
      "tokens_seen": 739704832
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039184553660982953,
      "loss": 3.0078,
      "theoretical_loss": 3.758411428012719,
      "tokens_seen": 739770368
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039183550651955866,
      "loss": 2.9239,
      "theoretical_loss": 3.7583779550641467,
      "tokens_seen": 739835904
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003918254764292879,
      "loss": 2.9282,
      "theoretical_loss": 3.7583444859106803,
      "tokens_seen": 739901440
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039181544633901707,
      "loss": 2.8704,
      "theoretical_loss": 3.7583110205515524,
      "tokens_seen": 739966976
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039180541624874625,
      "loss": 3.1633,
      "theoretical_loss": 3.7582775589859976,
      "tokens_seen": 740032512
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039179538615847543,
      "loss": 3.0069,
      "theoretical_loss": 3.7582441012132497,
      "tokens_seen": 740098048
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039178535606820467,
      "loss": 3.0549,
      "theoretical_loss": 3.758210647232543,
      "tokens_seen": 740163584
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003917753259779338,
      "loss": 3.2268,
      "theoretical_loss": 3.7581771970431124,
      "tokens_seen": 740229120
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039176529588766303,
      "loss": 2.9716,
      "theoretical_loss": 3.7581437506441926,
      "tokens_seen": 740294656
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039175526579739216,
      "loss": 2.9307,
      "theoretical_loss": 3.758110308035018,
      "tokens_seen": 740360192
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003917452357071214,
      "loss": 2.7297,
      "theoretical_loss": 3.7580768692148236,
      "tokens_seen": 740425728
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003917352056168506,
      "loss": 2.9454,
      "theoretical_loss": 3.758043434182846,
      "tokens_seen": 740491264
    },
    {
      "epoch": 0.22,
      "objective/train/advantage_avg": 0.49524280428886414,
      "objective/train/docs_used": 427723,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.6957778930664062,
      "objective/train/original_loss": 2.6957778930664062,
      "objective/train/theoretical_loss": 3.75801000293832,
      "objective/train/tokens_used": 761016800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24532143771648407,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507713556289673,
      "objective/train/weighted_lm_loss": 2.8326942920684814,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0454468727111816,
      "theoretical_loss": 3.75801000293832,
      "tokens_seen": 740556800
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039172517552657976,
      "loss": 3.0565,
      "theoretical_loss": 3.75801000293832,
      "tokens_seen": 740556800
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039171514543630894,
      "loss": 2.782,
      "theoretical_loss": 3.7579765754804813,
      "tokens_seen": 740622336
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003917051153460381,
      "loss": 3.091,
      "theoretical_loss": 3.7579431518085666,
      "tokens_seen": 740687872
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003916950852557673,
      "loss": 3.0416,
      "theoretical_loss": 3.7579097319218113,
      "tokens_seen": 740753408
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039168505516549653,
      "loss": 3.2872,
      "theoretical_loss": 3.757876315819453,
      "tokens_seen": 740818944
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039167502507522566,
      "loss": 2.6753,
      "theoretical_loss": 3.7578429035007277,
      "tokens_seen": 740884480
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003916649949849549,
      "loss": 3.0454,
      "theoretical_loss": 3.7578094949648726,
      "tokens_seen": 740950016
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003916549648946841,
      "loss": 3.1079,
      "theoretical_loss": 3.7577760902111246,
      "tokens_seen": 741015552
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039164493480441326,
      "loss": 2.8268,
      "theoretical_loss": 3.7577426892387216,
      "tokens_seen": 741081088
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039163490471414244,
      "loss": 3.1562,
      "theoretical_loss": 3.7577092920469006,
      "tokens_seen": 741146624
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003916248746238716,
      "loss": 2.7975,
      "theoretical_loss": 3.7576758986348997,
      "tokens_seen": 741212160
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003916148445336008,
      "loss": 2.8016,
      "theoretical_loss": 3.7576425090019576,
      "tokens_seen": 741277696
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039160481444333004,
      "loss": 3.0441,
      "theoretical_loss": 3.7576091231473114,
      "tokens_seen": 741343232
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039159478435305916,
      "loss": 3.0708,
      "theoretical_loss": 3.757575741070201,
      "tokens_seen": 741408768
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003915847542627884,
      "loss": 3.005,
      "theoretical_loss": 3.757542362769864,
      "tokens_seen": 741474304
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003915747241725175,
      "loss": 3.1012,
      "theoretical_loss": 3.7575089882455392,
      "tokens_seen": 741539840
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039156469408224676,
      "loss": 3.0515,
      "theoretical_loss": 3.7574756174964667,
      "tokens_seen": 741605376
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039155466399197594,
      "loss": 3.1875,
      "theoretical_loss": 3.757442250521886,
      "tokens_seen": 741670912
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003915446339017051,
      "loss": 2.8531,
      "theoretical_loss": 3.7574088873210356,
      "tokens_seen": 741736448
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003915346038114343,
      "loss": 3.2103,
      "theoretical_loss": 3.757375527893156,
      "tokens_seen": 741801984
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003915245737211635,
      "loss": 3.2555,
      "theoretical_loss": 3.757342172237487,
      "tokens_seen": 741867520
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039151454363089267,
      "loss": 3.0086,
      "theoretical_loss": 3.7573088203532694,
      "tokens_seen": 741933056
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003915045135406219,
      "loss": 2.8413,
      "theoretical_loss": 3.7572754722397432,
      "tokens_seen": 741998592
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039149448345035103,
      "loss": 3.0338,
      "theoretical_loss": 3.757242127896149,
      "tokens_seen": 742064128
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039148445336008027,
      "loss": 2.8362,
      "theoretical_loss": 3.757208787321728,
      "tokens_seen": 742129664
    },
    {
      "epoch": 0.22,
      "objective/train/advantage_avg": 0.4959867596626282,
      "objective/train/docs_used": 429091,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.8136589527130127,
      "objective/train/original_loss": 2.813659429550171,
      "objective/train/theoretical_loss": 3.7571754505157218,
      "objective/train/tokens_used": 762655200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24603882431983948,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0508495569229126,
      "objective/train/weighted_lm_loss": 2.95666241645813,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0458937883377075,
      "theoretical_loss": 3.7571754505157218,
      "tokens_seen": 742195200
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039147442326980945,
      "loss": 2.8824,
      "theoretical_loss": 3.7571754505157218,
      "tokens_seen": 742195200
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039146439317953863,
      "loss": 2.8046,
      "theoretical_loss": 3.7571421174773705,
      "tokens_seen": 742260736
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.0003914543630892678,
      "loss": 2.8966,
      "theoretical_loss": 3.7571087882059166,
      "tokens_seen": 742326272
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.000391444332998997,
      "loss": 2.908,
      "theoretical_loss": 3.7570754627006018,
      "tokens_seen": 742391808
    },
    {
      "epoch": 0.22,
      "learning_rate": 0.00039143430290872617,
      "loss": 3.0931,
      "theoretical_loss": 3.7570421409606682,
      "tokens_seen": 742457344
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003914242728184554,
      "loss": 2.8048,
      "theoretical_loss": 3.757008822985358,
      "tokens_seen": 742522880
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00039141424272818453,
      "loss": 2.9987,
      "theoretical_loss": 3.756975508773914,
      "tokens_seen": 742588416
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00039140421263791377,
      "loss": 3.1974,
      "theoretical_loss": 3.7569421983255777,
      "tokens_seen": 742653952
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003913941825476429,
      "loss": 3.0211,
      "theoretical_loss": 3.7569088916395934,
      "tokens_seen": 742719488
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00039138415245737213,
      "loss": 2.8964,
      "theoretical_loss": 3.756875588715203,
      "tokens_seen": 742785024
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003913741223671013,
      "loss": 3.0367,
      "theoretical_loss": 3.756842289551651,
      "tokens_seen": 742850560
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003913640922768305,
      "loss": 2.6271,
      "theoretical_loss": 3.7568089941481806,
      "tokens_seen": 742916096
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003913540621865597,
      "loss": 3.1888,
      "theoretical_loss": 3.7567757025040347,
      "tokens_seen": 742981632
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00039134403209628886,
      "loss": 2.8426,
      "theoretical_loss": 3.7567424146184587,
      "tokens_seen": 743047168
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00039133400200601804,
      "loss": 3.0344,
      "theoretical_loss": 3.7567091304906954,
      "tokens_seen": 743112704
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00039132397191574727,
      "loss": 3.0788,
      "theoretical_loss": 3.756675850119991,
      "tokens_seen": 743178240
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003913139418254764,
      "loss": 3.0321,
      "theoretical_loss": 3.756642573505588,
      "tokens_seen": 743243776
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00039130391173520563,
      "loss": 2.9319,
      "theoretical_loss": 3.7566093006467334,
      "tokens_seen": 743309312
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003912938816449348,
      "loss": 2.8537,
      "theoretical_loss": 3.7565760315426706,
      "tokens_seen": 743374848
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.000391283851554664,
      "loss": 3.2122,
      "theoretical_loss": 3.756542766192646,
      "tokens_seen": 743440384
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003912738214643932,
      "loss": 2.7898,
      "theoretical_loss": 3.756509504595904,
      "tokens_seen": 743505920
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00039126379137412236,
      "loss": 3.0594,
      "theoretical_loss": 3.7564762467516917,
      "tokens_seen": 743571456
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00039125376128385154,
      "loss": 2.97,
      "theoretical_loss": 3.756442992659254,
      "tokens_seen": 743636992
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003912437311935808,
      "loss": 2.8366,
      "theoretical_loss": 3.756409742317838,
      "tokens_seen": 743702528
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003912337011033099,
      "loss": 3.2802,
      "theoretical_loss": 3.7563764957266894,
      "tokens_seen": 743768064
    },
    {
      "epoch": 0.23,
      "objective/train/advantage_avg": 0.4948459267616272,
      "objective/train/docs_used": 429832,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.958930730819702,
      "objective/train/original_loss": 2.958930730819702,
      "objective/train/theoretical_loss": 3.756343252885055,
      "objective/train/tokens_used": 764293600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2449265569448471,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050729751586914,
      "objective/train/weighted_lm_loss": 3.109095335006714,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0442860126495361,
      "theoretical_loss": 3.756343252885055,
      "tokens_seen": 743833600
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00039122367101303914,
      "loss": 3.0442,
      "theoretical_loss": 3.756343252885055,
      "tokens_seen": 743833600
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00039121364092276826,
      "loss": 2.9846,
      "theoretical_loss": 3.7563100137921817,
      "tokens_seen": 743899136
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003912036108324975,
      "loss": 3.0368,
      "theoretical_loss": 3.756276778447316,
      "tokens_seen": 743964672
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003911935807422267,
      "loss": 2.9157,
      "theoretical_loss": 3.7562435468497064,
      "tokens_seen": 744030208
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00039118355065195586,
      "loss": 3.1683,
      "theoretical_loss": 3.7562103189985994,
      "tokens_seen": 744095744
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00039117352056168504,
      "loss": 3.0184,
      "theoretical_loss": 3.756177094893243,
      "tokens_seen": 744161280
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003911634904714143,
      "loss": 3.0701,
      "theoretical_loss": 3.7561438745328855,
      "tokens_seen": 744226816
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003911534603811434,
      "loss": 3.4026,
      "theoretical_loss": 3.756110657916774,
      "tokens_seen": 744292352
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00039114343029087264,
      "loss": 3.0336,
      "theoretical_loss": 3.7560774450441583,
      "tokens_seen": 744357888
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003911334002006018,
      "loss": 2.8821,
      "theoretical_loss": 3.7560442359142856,
      "tokens_seen": 744423424
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.000391123370110331,
      "loss": 3.0806,
      "theoretical_loss": 3.7560110305264054,
      "tokens_seen": 744488960
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00039111334002006024,
      "loss": 3.0858,
      "theoretical_loss": 3.7559778288797663,
      "tokens_seen": 744554496
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00039110330992978936,
      "loss": 2.8402,
      "theoretical_loss": 3.755944630973618,
      "tokens_seen": 744620032
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003910932798395186,
      "loss": 3.0273,
      "theoretical_loss": 3.75591143680721,
      "tokens_seen": 744685568
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00039108324974924773,
      "loss": 2.902,
      "theoretical_loss": 3.7558782463797917,
      "tokens_seen": 744751104
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00039107321965897696,
      "loss": 2.9879,
      "theoretical_loss": 3.755845059690613,
      "tokens_seen": 744816640
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00039106318956870614,
      "loss": 3.0501,
      "theoretical_loss": 3.7558118767389237,
      "tokens_seen": 744882176
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003910531594784353,
      "loss": 3.0581,
      "theoretical_loss": 3.755778697523975,
      "tokens_seen": 744947712
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003910431293881645,
      "loss": 3.117,
      "theoretical_loss": 3.755745522045016,
      "tokens_seen": 745013248
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003910330992978937,
      "loss": 2.7658,
      "theoretical_loss": 3.7557123503012986,
      "tokens_seen": 745078784
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00039102306920762287,
      "loss": 3.0266,
      "theoretical_loss": 3.755679182292073,
      "tokens_seen": 745144320
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003910130391173521,
      "loss": 2.8218,
      "theoretical_loss": 3.7556460180165914,
      "tokens_seen": 745209856
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00039100300902708123,
      "loss": 2.994,
      "theoretical_loss": 3.7556128574741043,
      "tokens_seen": 745275392
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00039099297893681047,
      "loss": 3.0318,
      "theoretical_loss": 3.755579700663864,
      "tokens_seen": 745340928
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00039098294884653965,
      "loss": 3.0949,
      "theoretical_loss": 3.755546547585121,
      "tokens_seen": 745406464
    },
    {
      "epoch": 0.23,
      "objective/train/advantage_avg": 0.4961787760257721,
      "objective/train/docs_used": 430489,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.5371859073638916,
      "objective/train/original_loss": 2.5371859073638916,
      "objective/train/theoretical_loss": 3.755513398237129,
      "objective/train/tokens_used": 765932000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2462252527475357,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0508695840835571,
      "objective/train/weighted_lm_loss": 2.666146993637085,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.044721245765686,
      "theoretical_loss": 3.755513398237129,
      "tokens_seen": 745472000
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00039097291875626883,
      "loss": 2.8302,
      "theoretical_loss": 3.755513398237129,
      "tokens_seen": 745472000
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.000390962888665998,
      "loss": 3.19,
      "theoretical_loss": 3.7554802526191393,
      "tokens_seen": 745537536
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003909528585757272,
      "loss": 3.0023,
      "theoretical_loss": 3.7554471107304046,
      "tokens_seen": 745603072
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00039094282848545637,
      "loss": 2.951,
      "theoretical_loss": 3.755413972570177,
      "tokens_seen": 745668608
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003909327983951856,
      "loss": 3.0824,
      "theoretical_loss": 3.7553808381377105,
      "tokens_seen": 745734144
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00039092276830491473,
      "loss": 2.5658,
      "theoretical_loss": 3.7553477074322577,
      "tokens_seen": 745799680
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00039091273821464397,
      "loss": 2.9311,
      "theoretical_loss": 3.7553145804530716,
      "tokens_seen": 745865216
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003909027081243731,
      "loss": 3.0438,
      "theoretical_loss": 3.7552814571994064,
      "tokens_seen": 745930752
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00039089267803410233,
      "loss": 3.0381,
      "theoretical_loss": 3.755248337670515,
      "tokens_seen": 745996288
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003908826479438315,
      "loss": 3.0449,
      "theoretical_loss": 3.755215221865652,
      "tokens_seen": 746061824
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003908726178535607,
      "loss": 2.908,
      "theoretical_loss": 3.755182109784071,
      "tokens_seen": 746127360
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003908625877632899,
      "loss": 2.9289,
      "theoretical_loss": 3.755149001425028,
      "tokens_seen": 746192896
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00039085255767301906,
      "loss": 3.0624,
      "theoretical_loss": 3.7551158967877756,
      "tokens_seen": 746258432
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00039084252758274824,
      "loss": 2.7156,
      "theoretical_loss": 3.75508279587157,
      "tokens_seen": 746323968
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00039083249749247747,
      "loss": 2.9803,
      "theoretical_loss": 3.7550496986756654,
      "tokens_seen": 746389504
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003908224674022066,
      "loss": 3.1972,
      "theoretical_loss": 3.7550166051993172,
      "tokens_seen": 746455040
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00039081243731193583,
      "loss": 2.7868,
      "theoretical_loss": 3.7549835154417814,
      "tokens_seen": 746520576
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.000390802407221665,
      "loss": 2.9653,
      "theoretical_loss": 3.7549504294023137,
      "tokens_seen": 746586112
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003907923771313942,
      "loss": 3.2018,
      "theoretical_loss": 3.7549173470801693,
      "tokens_seen": 746651648
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003907823470411234,
      "loss": 3.1043,
      "theoretical_loss": 3.754884268474605,
      "tokens_seen": 746717184
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00039077231695085256,
      "loss": 3.1177,
      "theoretical_loss": 3.7548511935848765,
      "tokens_seen": 746782720
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00039076228686058174,
      "loss": 3.134,
      "theoretical_loss": 3.7548181224102404,
      "tokens_seen": 746848256
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.000390752256770311,
      "loss": 3.1346,
      "theoretical_loss": 3.754785054949954,
      "tokens_seen": 746913792
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003907422266800401,
      "loss": 2.983,
      "theoretical_loss": 3.7547519912032747,
      "tokens_seen": 746979328
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00039073219658976934,
      "loss": 3.0622,
      "theoretical_loss": 3.7547189311694584,
      "tokens_seen": 747044864
    },
    {
      "epoch": 0.23,
      "objective/train/advantage_avg": 0.4928646981716156,
      "objective/train/docs_used": 431396,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.70693039894104,
      "objective/train/original_loss": 2.70693039894104,
      "objective/train/theoretical_loss": 3.7546858748477634,
      "objective/train/tokens_used": 767570400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2430116981267929,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505218505859375,
      "objective/train/weighted_lm_loss": 2.843421220779419,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0425469875335693,
      "theoretical_loss": 3.7546858748477634,
      "tokens_seen": 747110400
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00039072216649949846,
      "loss": 2.9338,
      "theoretical_loss": 3.7546858748477634,
      "tokens_seen": 747110400
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003907121364092277,
      "loss": 3.0015,
      "theoretical_loss": 3.7546528222374467,
      "tokens_seen": 747175936
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003907021063189569,
      "loss": 3.1938,
      "theoretical_loss": 3.7546197733377666,
      "tokens_seen": 747241472
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00039069207622868606,
      "loss": 3.0696,
      "theoretical_loss": 3.7545867281479812,
      "tokens_seen": 747307008
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00039068204613841524,
      "loss": 3.0753,
      "theoretical_loss": 3.7545536866673483,
      "tokens_seen": 747372544
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003906720160481445,
      "loss": 3.1775,
      "theoretical_loss": 3.754520648895127,
      "tokens_seen": 747438080
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003906619859578736,
      "loss": 3.0792,
      "theoretical_loss": 3.754487614830575,
      "tokens_seen": 747503616
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00039065195586760284,
      "loss": 2.9447,
      "theoretical_loss": 3.7544545844729527,
      "tokens_seen": 747569152
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00039064192577733197,
      "loss": 2.9214,
      "theoretical_loss": 3.7544215578215177,
      "tokens_seen": 747634688
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003906318956870612,
      "loss": 3.0525,
      "theoretical_loss": 3.75438853487553,
      "tokens_seen": 747700224
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003906218655967904,
      "loss": 3.2306,
      "theoretical_loss": 3.7543555156342494,
      "tokens_seen": 747765760
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00039061183550651957,
      "loss": 3.2033,
      "theoretical_loss": 3.7543225000969356,
      "tokens_seen": 747831296
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00039060180541624875,
      "loss": 3.0245,
      "theoretical_loss": 3.7542894882628475,
      "tokens_seen": 747896832
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00039059177532597793,
      "loss": 3.0216,
      "theoretical_loss": 3.7542564801312466,
      "tokens_seen": 747962368
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003905817452357071,
      "loss": 2.968,
      "theoretical_loss": 3.7542234757013926,
      "tokens_seen": 748027904
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00039057171514543634,
      "loss": 2.9323,
      "theoretical_loss": 3.7541904749725465,
      "tokens_seen": 748093440
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00039056168505516547,
      "loss": 3.0131,
      "theoretical_loss": 3.7541574779439686,
      "tokens_seen": 748158976
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003905516549648947,
      "loss": 2.9269,
      "theoretical_loss": 3.7541244846149207,
      "tokens_seen": 748224512
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00039054162487462383,
      "loss": 3.0447,
      "theoretical_loss": 3.754091494984663,
      "tokens_seen": 748290048
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00039053159478435307,
      "loss": 2.9117,
      "theoretical_loss": 3.7540585090524576,
      "tokens_seen": 748355584
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00039052156469408225,
      "loss": 2.8914,
      "theoretical_loss": 3.7540255268175664,
      "tokens_seen": 748421120
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00039051153460381143,
      "loss": 3.0194,
      "theoretical_loss": 3.7539925482792507,
      "tokens_seen": 748486656
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003905015045135406,
      "loss": 3.0254,
      "theoretical_loss": 3.753959573436773,
      "tokens_seen": 748552192
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00039049147442326985,
      "loss": 2.9089,
      "theoretical_loss": 3.753926602289395,
      "tokens_seen": 748617728
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.000390481444332999,
      "loss": 3.0923,
      "theoretical_loss": 3.75389363483638,
      "tokens_seen": 748683264
    },
    {
      "epoch": 0.23,
      "objective/train/advantage_avg": 0.4951666295528412,
      "objective/train/docs_used": 432599,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.95467472076416,
      "objective/train/original_loss": 2.954674482345581,
      "objective/train/theoretical_loss": 3.7538606710769904,
      "objective/train/tokens_used": 769208800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24526771903038025,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050763487815857,
      "objective/train/weighted_lm_loss": 3.104649782180786,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0363596677780151,
      "theoretical_loss": 3.7538606710769904,
      "tokens_seen": 748748800
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003904714142427282,
      "loss": 2.9136,
      "theoretical_loss": 3.7538606710769904,
      "tokens_seen": 748748800
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00039046138415245734,
      "loss": 3.2267,
      "theoretical_loss": 3.753827711010489,
      "tokens_seen": 748814336
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00039045135406218657,
      "loss": 3.2386,
      "theoretical_loss": 3.7537947546361385,
      "tokens_seen": 748879872
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00039044132397191575,
      "loss": 3.1505,
      "theoretical_loss": 3.7537618019532033,
      "tokens_seen": 748945408
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00039043129388164493,
      "loss": 3.226,
      "theoretical_loss": 3.753728852960947,
      "tokens_seen": 749010944
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003904212637913741,
      "loss": 2.8737,
      "theoretical_loss": 3.7536959076586314,
      "tokens_seen": 749076480
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003904112337011033,
      "loss": 3.1572,
      "theoretical_loss": 3.753662966045523,
      "tokens_seen": 749142016
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003904012036108325,
      "loss": 3.3752,
      "theoretical_loss": 3.7536300281208845,
      "tokens_seen": 749207552
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003903911735205617,
      "loss": 3.2242,
      "theoretical_loss": 3.753597093883981,
      "tokens_seen": 749273088
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003903811434302909,
      "loss": 2.9871,
      "theoretical_loss": 3.753564163334077,
      "tokens_seen": 749338624
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003903711133400201,
      "loss": 2.9524,
      "theoretical_loss": 3.7535312364704367,
      "tokens_seen": 749404160
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00039036108324974926,
      "loss": 2.9227,
      "theoretical_loss": 3.753498313292326,
      "tokens_seen": 749469696
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00039035105315947844,
      "loss": 3.1917,
      "theoretical_loss": 3.7534653937990097,
      "tokens_seen": 749535232
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00039034102306920767,
      "loss": 3.1795,
      "theoretical_loss": 3.7534324779897528,
      "tokens_seen": 749600768
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003903309929789368,
      "loss": 3.2567,
      "theoretical_loss": 3.7533995658638224,
      "tokens_seen": 749666304
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00039032096288866603,
      "loss": 3.1004,
      "theoretical_loss": 3.753366657420483,
      "tokens_seen": 749731840
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003903109327983952,
      "loss": 2.9543,
      "theoretical_loss": 3.7533337526590014,
      "tokens_seen": 749797376
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003903009027081244,
      "loss": 3.0018,
      "theoretical_loss": 3.7533008515786435,
      "tokens_seen": 749862912
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003902908726178536,
      "loss": 3.1346,
      "theoretical_loss": 3.753267954178676,
      "tokens_seen": 749928448
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00039028084252758276,
      "loss": 3.1695,
      "theoretical_loss": 3.7532350604583664,
      "tokens_seen": 749993984
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00039027081243731194,
      "loss": 3.0908,
      "theoretical_loss": 3.75320217041698,
      "tokens_seen": 750059520
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003902607823470412,
      "loss": 3.005,
      "theoretical_loss": 3.7531692840537856,
      "tokens_seen": 750125056
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003902507522567703,
      "loss": 2.8017,
      "theoretical_loss": 3.7531364013680495,
      "tokens_seen": 750190592
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00039024072216649954,
      "loss": 3.0916,
      "theoretical_loss": 3.7531035223590394,
      "tokens_seen": 750256128
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00039023069207622866,
      "loss": 2.9084,
      "theoretical_loss": 3.7530706470260236,
      "tokens_seen": 750321664
    },
    {
      "epoch": 0.23,
      "objective/train/advantage_avg": 0.49422022700309753,
      "objective/train/docs_used": 433353,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.9255757331848145,
      "objective/train/original_loss": 2.9255762100219727,
      "objective/train/theoretical_loss": 3.7530377753682695,
      "objective/train/tokens_used": 770847200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24431541562080383,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050663948059082,
      "objective/train/weighted_lm_loss": 3.073798179626465,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0338950157165527,
      "theoretical_loss": 3.7530377753682695,
      "tokens_seen": 750387200
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003902206619859579,
      "loss": 3.1859,
      "theoretical_loss": 3.7530377753682695,
      "tokens_seen": 750387200
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003902106318956871,
      "loss": 3.0493,
      "theoretical_loss": 3.7530049073850456,
      "tokens_seen": 750452736
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00039020060180541626,
      "loss": 2.9112,
      "theoretical_loss": 3.75297204307562,
      "tokens_seen": 750518272
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00039019057171514544,
      "loss": 2.9317,
      "theoretical_loss": 3.7529391824392624,
      "tokens_seen": 750583808
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003901805416248747,
      "loss": 3.0049,
      "theoretical_loss": 3.75290632547524,
      "tokens_seen": 750649344
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003901705115346038,
      "loss": 3.1669,
      "theoretical_loss": 3.7528734721828227,
      "tokens_seen": 750714880
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00039016048144433304,
      "loss": 3.0932,
      "theoretical_loss": 3.75284062256128,
      "tokens_seen": 750780416
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00039015045135406217,
      "loss": 3.0777,
      "theoretical_loss": 3.752807776609881,
      "tokens_seen": 750845952
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003901404212637914,
      "loss": 2.9901,
      "theoretical_loss": 3.7527749343278956,
      "tokens_seen": 750911488
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003901303911735206,
      "loss": 3.018,
      "theoretical_loss": 3.752742095714593,
      "tokens_seen": 750977024
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00039012036108324977,
      "loss": 3.0716,
      "theoretical_loss": 3.7527092607692443,
      "tokens_seen": 751042560
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00039011033099297895,
      "loss": 3.0076,
      "theoretical_loss": 3.752676429491119,
      "tokens_seen": 751108096
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00039010030090270813,
      "loss": 2.9663,
      "theoretical_loss": 3.7526436018794875,
      "tokens_seen": 751173632
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003900902708124373,
      "loss": 3.0223,
      "theoretical_loss": 3.7526107779336213,
      "tokens_seen": 751239168
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00039008024072216654,
      "loss": 3.0084,
      "theoretical_loss": 3.752577957652791,
      "tokens_seen": 751304704
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00039007021063189567,
      "loss": 3.0432,
      "theoretical_loss": 3.752545141036267,
      "tokens_seen": 751370240
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003900601805416249,
      "loss": 3.2705,
      "theoretical_loss": 3.7525123280833217,
      "tokens_seen": 751435776
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00039005015045135403,
      "loss": 3.2071,
      "theoretical_loss": 3.7524795187932263,
      "tokens_seen": 751501312
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00039004012036108327,
      "loss": 2.9323,
      "theoretical_loss": 3.752446713165252,
      "tokens_seen": 751566848
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00039003009027081245,
      "loss": 2.9517,
      "theoretical_loss": 3.7524139111986714,
      "tokens_seen": 751632384
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00039002006018054163,
      "loss": 2.9962,
      "theoretical_loss": 3.7523811128927567,
      "tokens_seen": 751697920
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003900100300902708,
      "loss": 2.9109,
      "theoretical_loss": 3.7523483182467796,
      "tokens_seen": 751763456
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00039000000000000005,
      "loss": 2.78,
      "theoretical_loss": 3.7523155272600137,
      "tokens_seen": 751828992
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003899899699097292,
      "loss": 2.9541,
      "theoretical_loss": 3.7522827399317307,
      "tokens_seen": 751894528
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003899799398194584,
      "loss": 3.1659,
      "theoretical_loss": 3.752249956261204,
      "tokens_seen": 751960064
    },
    {
      "epoch": 0.23,
      "objective/train/advantage_avg": 0.4936263859272003,
      "objective/train/docs_used": 434593,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.6500208377838135,
      "objective/train/original_loss": 2.6500208377838135,
      "objective/train/theoretical_loss": 3.752217176247707,
      "objective/train/tokens_used": 772485600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2437974512577057,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506020784378052,
      "objective/train/weighted_lm_loss": 2.783993721008301,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0338950157165527,
      "theoretical_loss": 3.752217176247707,
      "tokens_seen": 752025600
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038996990972918754,
      "loss": 3.15,
      "theoretical_loss": 3.752217176247707,
      "tokens_seen": 752025600
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038995987963891677,
      "loss": 2.966,
      "theoretical_loss": 3.7521843998905133,
      "tokens_seen": 752091136
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038994984954864595,
      "loss": 2.9719,
      "theoretical_loss": 3.7521516271888964,
      "tokens_seen": 752156672
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038993981945837513,
      "loss": 2.9305,
      "theoretical_loss": 3.752118858142129,
      "tokens_seen": 752222208
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003899297893681043,
      "loss": 3.0231,
      "theoretical_loss": 3.7520860927494875,
      "tokens_seen": 752287744
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003899197592778335,
      "loss": 2.7431,
      "theoretical_loss": 3.752053331010244,
      "tokens_seen": 752353280
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003899097291875627,
      "loss": 3.127,
      "theoretical_loss": 3.7520205729236737,
      "tokens_seen": 752418816
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003898996990972919,
      "loss": 2.7651,
      "theoretical_loss": 3.7519878184890514,
      "tokens_seen": 752484352
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038988966900702104,
      "loss": 2.9967,
      "theoretical_loss": 3.7519550677056523,
      "tokens_seen": 752549888
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003898796389167503,
      "loss": 3.1228,
      "theoretical_loss": 3.75192232057275,
      "tokens_seen": 752615424
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003898696088264794,
      "loss": 2.9607,
      "theoretical_loss": 3.7518895770896217,
      "tokens_seen": 752680960
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038985957873620864,
      "loss": 2.9733,
      "theoretical_loss": 3.7518568372555414,
      "tokens_seen": 752746496
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003898495486459378,
      "loss": 3.1252,
      "theoretical_loss": 3.7518241010697864,
      "tokens_seen": 752812032
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.000389839518555667,
      "loss": 2.8801,
      "theoretical_loss": 3.751791368531631,
      "tokens_seen": 752877568
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003898294884653962,
      "loss": 2.9956,
      "theoretical_loss": 3.751758639640351,
      "tokens_seen": 752943104
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003898194583751254,
      "loss": 3.2665,
      "theoretical_loss": 3.751725914395225,
      "tokens_seen": 753008640
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038980942828485454,
      "loss": 2.983,
      "theoretical_loss": 3.751693192795527,
      "tokens_seen": 753074176
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003897993981945838,
      "loss": 3.2748,
      "theoretical_loss": 3.7516604748405356,
      "tokens_seen": 753139712
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003897893681043129,
      "loss": 3.0414,
      "theoretical_loss": 3.7516277605295265,
      "tokens_seen": 753205248
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038977933801404214,
      "loss": 3.1573,
      "theoretical_loss": 3.751595049861778,
      "tokens_seen": 753270784
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003897693079237713,
      "loss": 2.9742,
      "theoretical_loss": 3.751562342836566,
      "tokens_seen": 753336320
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003897592778335005,
      "loss": 3.021,
      "theoretical_loss": 3.751529639453169,
      "tokens_seen": 753401856
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003897492477432297,
      "loss": 3.004,
      "theoretical_loss": 3.751496939710865,
      "tokens_seen": 753467392
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038973921765295886,
      "loss": 3.2756,
      "theoretical_loss": 3.7514642436089316,
      "tokens_seen": 753532928
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038972918756268805,
      "loss": 2.8773,
      "theoretical_loss": 3.7514315511466467,
      "tokens_seen": 753598464
    },
    {
      "epoch": 0.23,
      "objective/train/advantage_avg": 0.4960218667984009,
      "objective/train/docs_used": 435245,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.7502408027648926,
      "objective/train/original_loss": 2.750241279602051,
      "objective/train/theoretical_loss": 3.7513988623232883,
      "objective/train/tokens_used": 774124000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2460862547159195,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0508532524108887,
      "objective/train/weighted_lm_loss": 2.8899593353271484,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 3.7513988623232883,
      "tokens_seen": 753664000
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003897191574724173,
      "loss": 2.8516,
      "theoretical_loss": 3.7513988623232883,
      "tokens_seen": 753664000
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003897091273821464,
      "loss": 3.2981,
      "theoretical_loss": 3.7513661771381366,
      "tokens_seen": 753729536
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038969909729187564,
      "loss": 2.8206,
      "theoretical_loss": 3.751333495590469,
      "tokens_seen": 753795072
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038968906720160477,
      "loss": 2.9286,
      "theoretical_loss": 3.7513008176795646,
      "tokens_seen": 753860608
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.000389679037111334,
      "loss": 3.0939,
      "theoretical_loss": 3.7512681434047033,
      "tokens_seen": 753926144
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003896690070210632,
      "loss": 2.9958,
      "theoretical_loss": 3.7512354727651642,
      "tokens_seen": 753991680
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038965897693079237,
      "loss": 3.1131,
      "theoretical_loss": 3.7512028057602267,
      "tokens_seen": 754057216
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038964894684052155,
      "loss": 2.9906,
      "theoretical_loss": 3.7511701423891703,
      "tokens_seen": 754122752
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003896389167502508,
      "loss": 2.7691,
      "theoretical_loss": 3.7511374826512753,
      "tokens_seen": 754188288
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038962888665997997,
      "loss": 2.7608,
      "theoretical_loss": 3.7511048265458227,
      "tokens_seen": 754253824
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038961885656970915,
      "loss": 2.9936,
      "theoretical_loss": 3.751072174072092,
      "tokens_seen": 754319360
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038960882647943833,
      "loss": 3.045,
      "theoretical_loss": 3.751039525229364,
      "tokens_seen": 754384896
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003895987963891675,
      "loss": 2.9702,
      "theoretical_loss": 3.75100688001692,
      "tokens_seen": 754450432
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038958876629889674,
      "loss": 3.1217,
      "theoretical_loss": 3.7509742384340403,
      "tokens_seen": 754515968
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038957873620862587,
      "loss": 3.0451,
      "theoretical_loss": 3.750941600480007,
      "tokens_seen": 754581504
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003895687061183551,
      "loss": 2.9492,
      "theoretical_loss": 3.7509089661541006,
      "tokens_seen": 754647040
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038955867602808423,
      "loss": 3.0779,
      "theoretical_loss": 3.750876335455604,
      "tokens_seen": 754712576
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038954864593781347,
      "loss": 3.0891,
      "theoretical_loss": 3.750843708383798,
      "tokens_seen": 754778112
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038953861584754265,
      "loss": 3.0514,
      "theoretical_loss": 3.750811084937965,
      "tokens_seen": 754843648
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038952858575727183,
      "loss": 2.9281,
      "theoretical_loss": 3.750778465117387,
      "tokens_seen": 754909184
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.000389518555667001,
      "loss": 3.0191,
      "theoretical_loss": 3.7507458489213477,
      "tokens_seen": 754974720
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038950852557673025,
      "loss": 2.8219,
      "theoretical_loss": 3.750713236349128,
      "tokens_seen": 755040256
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003894984954864594,
      "loss": 3.0941,
      "theoretical_loss": 3.750680627400012,
      "tokens_seen": 755105792
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003894884653961886,
      "loss": 2.8996,
      "theoretical_loss": 3.7506480220732823,
      "tokens_seen": 755171328
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038947843530591774,
      "loss": 2.9595,
      "theoretical_loss": 3.7506154203682223,
      "tokens_seen": 755236864
    },
    {
      "epoch": 0.23,
      "objective/train/advantage_avg": 0.4921579658985138,
      "objective/train/docs_used": 436337,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.6170692443847656,
      "objective/train/original_loss": 2.6170692443847656,
      "objective/train/theoretical_loss": 3.7505828222841155,
      "objective/train/tokens_used": 775762400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2423032969236374,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0504473447799683,
      "objective/train/weighted_lm_loss": 2.749159097671509,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0408108234405518,
      "theoretical_loss": 3.7505828222841155,
      "tokens_seen": 755302400
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038946840521564697,
      "loss": 2.825,
      "theoretical_loss": 3.7505828222841155,
      "tokens_seen": 755302400
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038945837512537615,
      "loss": 3.0148,
      "theoretical_loss": 3.750550227820246,
      "tokens_seen": 755367936
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038944834503510533,
      "loss": 3.1407,
      "theoretical_loss": 3.7505176369758972,
      "tokens_seen": 755433472
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003894383149448345,
      "loss": 2.733,
      "theoretical_loss": 3.750485049750353,
      "tokens_seen": 755499008
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003894282848545637,
      "loss": 2.8513,
      "theoretical_loss": 3.7504524661428986,
      "tokens_seen": 755564544
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003894182547642929,
      "loss": 3.2044,
      "theoretical_loss": 3.7504198861528177,
      "tokens_seen": 755630080
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003894082246740221,
      "loss": 3.0941,
      "theoretical_loss": 3.750387309779395,
      "tokens_seen": 755695616
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038939819458375124,
      "loss": 3.1126,
      "theoretical_loss": 3.750354737021916,
      "tokens_seen": 755761152
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003893881644934805,
      "loss": 3.0228,
      "theoretical_loss": 3.750322167879666,
      "tokens_seen": 755826688
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003893781344032096,
      "loss": 3.0766,
      "theoretical_loss": 3.7502896023519297,
      "tokens_seen": 755892224
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038936810431293884,
      "loss": 2.9126,
      "theoretical_loss": 3.7502570404379925,
      "tokens_seen": 755957760
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.000389358074222668,
      "loss": 3.1369,
      "theoretical_loss": 3.7502244821371407,
      "tokens_seen": 756023296
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003893480441323972,
      "loss": 3.0651,
      "theoretical_loss": 3.75019192744866,
      "tokens_seen": 756088832
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003893380140421264,
      "loss": 2.8314,
      "theoretical_loss": 3.750159376371837,
      "tokens_seen": 756154368
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003893279839518556,
      "loss": 3.1123,
      "theoretical_loss": 3.7501268289059566,
      "tokens_seen": 756219904
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038931795386158474,
      "loss": 3.2004,
      "theoretical_loss": 3.7500942850503067,
      "tokens_seen": 756285440
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.000389307923771314,
      "loss": 2.8577,
      "theoretical_loss": 3.7500617448041744,
      "tokens_seen": 756350976
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003892978936810431,
      "loss": 2.9204,
      "theoretical_loss": 3.750029208166845,
      "tokens_seen": 756416512
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038928786359077234,
      "loss": 2.899,
      "theoretical_loss": 3.7499966751376075,
      "tokens_seen": 756482048
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003892778335005015,
      "loss": 3.0984,
      "theoretical_loss": 3.749964145715748,
      "tokens_seen": 756547584
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003892678034102307,
      "loss": 3.2425,
      "theoretical_loss": 3.7499316199005546,
      "tokens_seen": 756613120
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003892577733199599,
      "loss": 3.1722,
      "theoretical_loss": 3.7498990976913147,
      "tokens_seen": 756678656
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038924774322968906,
      "loss": 3.0049,
      "theoretical_loss": 3.7498665790873167,
      "tokens_seen": 756744192
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038923771313941825,
      "loss": 3.0052,
      "theoretical_loss": 3.7498340640878487,
      "tokens_seen": 756809728
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003892276830491475,
      "loss": 3.0626,
      "theoretical_loss": 3.7498015526921984,
      "tokens_seen": 756875264
    },
    {
      "epoch": 0.23,
      "objective/train/advantage_avg": 0.49288246035575867,
      "objective/train/docs_used": 436904,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.1557726860046387,
      "objective/train/original_loss": 3.1557724475860596,
      "objective/train/theoretical_loss": 3.7497690448996552,
      "objective/train/tokens_used": 777400800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24302218854427338,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505235195159912,
      "objective/train/weighted_lm_loss": 3.3151166439056396,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0338950157165527,
      "theoretical_loss": 3.7497690448996552,
      "tokens_seen": 756940800
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003892176529588766,
      "loss": 3.0141,
      "theoretical_loss": 3.7497690448996552,
      "tokens_seen": 756940800
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038920762286860584,
      "loss": 2.9457,
      "theoretical_loss": 3.749736540709508,
      "tokens_seen": 757006336
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038919759277833497,
      "loss": 2.9896,
      "theoretical_loss": 3.7497040401210446,
      "tokens_seen": 757071872
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003891875626880642,
      "loss": 2.9534,
      "theoretical_loss": 3.749671543133555,
      "tokens_seen": 757137408
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003891775325977934,
      "loss": 3.1906,
      "theoretical_loss": 3.749639049746329,
      "tokens_seen": 757202944
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038916750250752257,
      "loss": 2.8945,
      "theoretical_loss": 3.7496065599586554,
      "tokens_seen": 757268480
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038915747241725175,
      "loss": 2.8793,
      "theoretical_loss": 3.7495740737698244,
      "tokens_seen": 757334016
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.000389147442326981,
      "loss": 2.9302,
      "theoretical_loss": 3.749541591179126,
      "tokens_seen": 757399552
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003891374122367101,
      "loss": 2.9158,
      "theoretical_loss": 3.7495091121858497,
      "tokens_seen": 757465088
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038912738214643935,
      "loss": 3.3718,
      "theoretical_loss": 3.7494766367892876,
      "tokens_seen": 757530624
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003891173520561685,
      "loss": 2.9627,
      "theoretical_loss": 3.749444164988728,
      "tokens_seen": 757596160
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003891073219658977,
      "loss": 2.7693,
      "theoretical_loss": 3.749411696783463,
      "tokens_seen": 757661696
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003890972918756269,
      "loss": 3.5,
      "theoretical_loss": 3.7493792321727843,
      "tokens_seen": 757727232
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038908726178535607,
      "loss": 3.086,
      "theoretical_loss": 3.749346771155982,
      "tokens_seen": 757792768
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038907723169508525,
      "loss": 2.8306,
      "theoretical_loss": 3.749314313732347,
      "tokens_seen": 757858304
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038906720160481443,
      "loss": 3.0326,
      "theoretical_loss": 3.749281859901172,
      "tokens_seen": 757923840
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003890571715145436,
      "loss": 2.8948,
      "theoretical_loss": 3.749249409661748,
      "tokens_seen": 757989376
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038904714142427285,
      "loss": 3.1135,
      "theoretical_loss": 3.7492169630133683,
      "tokens_seen": 758054912
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.000389037111334002,
      "loss": 2.8483,
      "theoretical_loss": 3.7491845199553238,
      "tokens_seen": 758120448
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003890270812437312,
      "loss": 3.0363,
      "theoretical_loss": 3.7491520804869074,
      "tokens_seen": 758185984
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003890170511534604,
      "loss": 3.1068,
      "theoretical_loss": 3.7491196446074113,
      "tokens_seen": 758251520
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003890070210631896,
      "loss": 3.0346,
      "theoretical_loss": 3.7490872123161285,
      "tokens_seen": 758317056
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038899699097291876,
      "loss": 2.9924,
      "theoretical_loss": 3.749054783612353,
      "tokens_seen": 758382592
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038898696088264794,
      "loss": 2.9813,
      "theoretical_loss": 3.7490223584953766,
      "tokens_seen": 758448128
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003889769307923771,
      "loss": 3.0196,
      "theoretical_loss": 3.748989936964493,
      "tokens_seen": 758513664
    },
    {
      "epoch": 0.23,
      "objective/train/advantage_avg": 0.49338769912719727,
      "objective/train/docs_used": 438157,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.7842562198638916,
      "objective/train/original_loss": 2.7842562198638916,
      "objective/train/theoretical_loss": 3.7489575190189965,
      "objective/train/tokens_used": 779039200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24350613355636597,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505764484405518,
      "objective/train/weighted_lm_loss": 2.9249866008758545,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0417574644088745,
      "theoretical_loss": 3.7489575190189965,
      "tokens_seen": 758579200
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038896690070210635,
      "loss": 2.9277,
      "theoretical_loss": 3.7489575190189965,
      "tokens_seen": 758579200
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003889568706118355,
      "loss": 2.8954,
      "theoretical_loss": 3.7489251046581797,
      "tokens_seen": 758644736
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003889468405215647,
      "loss": 2.9883,
      "theoretical_loss": 3.7488926938813383,
      "tokens_seen": 758710272
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038893681043129384,
      "loss": 3.023,
      "theoretical_loss": 3.748860286687765,
      "tokens_seen": 758775808
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003889267803410231,
      "loss": 2.9434,
      "theoretical_loss": 3.748827883076755,
      "tokens_seen": 758841344
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038891675025075226,
      "loss": 2.8453,
      "theoretical_loss": 3.7487954830476022,
      "tokens_seen": 758906880
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038890672016048144,
      "loss": 2.7873,
      "theoretical_loss": 3.7487630865996024,
      "tokens_seen": 758972416
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003888966900702106,
      "loss": 2.8666,
      "theoretical_loss": 3.7487306937320497,
      "tokens_seen": 759037952
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003888866599799398,
      "loss": 3.0614,
      "theoretical_loss": 3.7486983044442397,
      "tokens_seen": 759103488
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038887662988966904,
      "loss": 3.0814,
      "theoretical_loss": 3.748665918735468,
      "tokens_seen": 759169024
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003888665997993982,
      "loss": 2.965,
      "theoretical_loss": 3.74863353660503,
      "tokens_seen": 759234560
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003888565697091274,
      "loss": 3.0189,
      "theoretical_loss": 3.7486011580522214,
      "tokens_seen": 759300096
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003888465396188566,
      "loss": 3.2492,
      "theoretical_loss": 3.7485687830763377,
      "tokens_seen": 759365632
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003888365095285858,
      "loss": 2.9428,
      "theoretical_loss": 3.7485364116766764,
      "tokens_seen": 759431168
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038882647943831494,
      "loss": 3.0017,
      "theoretical_loss": 3.7485040438525328,
      "tokens_seen": 759496704
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003888164493480442,
      "loss": 3.0066,
      "theoretical_loss": 3.748471679603204,
      "tokens_seen": 759562240
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003888064192577733,
      "loss": 3.0332,
      "theoretical_loss": 3.7484393189279865,
      "tokens_seen": 759627776
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038879638916750254,
      "loss": 3.0351,
      "theoretical_loss": 3.7484069618261775,
      "tokens_seen": 759693312
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003887863590772317,
      "loss": 3.0515,
      "theoretical_loss": 3.748374608297074,
      "tokens_seen": 759758848
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003887763289869609,
      "loss": 2.9481,
      "theoretical_loss": 3.7483422583399735,
      "tokens_seen": 759824384
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003887662988966901,
      "loss": 3.0044,
      "theoretical_loss": 3.7483099119541743,
      "tokens_seen": 759889920
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038875626880641927,
      "loss": 2.8909,
      "theoretical_loss": 3.748277569138973,
      "tokens_seen": 759955456
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038874623871614845,
      "loss": 2.9577,
      "theoretical_loss": 3.7482452298936684,
      "tokens_seen": 760020992
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003887362086258777,
      "loss": 3.0536,
      "theoretical_loss": 3.7482128942175574,
      "tokens_seen": 760086528
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003887261785356068,
      "loss": 2.9138,
      "theoretical_loss": 3.7481805621099404,
      "tokens_seen": 760152064
    },
    {
      "epoch": 0.23,
      "objective/train/advantage_avg": 0.49367496371269226,
      "objective/train/docs_used": 438864,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.6903672218322754,
      "objective/train/original_loss": 2.6903676986694336,
      "objective/train/theoretical_loss": 3.748148233570115,
      "objective/train/tokens_used": 780677600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24379558861255646,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506068468093872,
      "objective/train/weighted_lm_loss": 2.826375961303711,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 3.748148233570115,
      "tokens_seen": 760217600
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038871614844533604,
      "loss": 2.8139,
      "theoretical_loss": 3.748148233570115,
      "tokens_seen": 760217600
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038870611835506517,
      "loss": 2.8558,
      "theoretical_loss": 3.7481159085973794,
      "tokens_seen": 760283136
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003886960882647944,
      "loss": 3.2589,
      "theoretical_loss": 3.7480835871910334,
      "tokens_seen": 760348672
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003886860581745236,
      "loss": 3.231,
      "theoretical_loss": 3.748051269350376,
      "tokens_seen": 760414208
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038867602808425277,
      "loss": 3.0525,
      "theoretical_loss": 3.7480189550747065,
      "tokens_seen": 760479744
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038866599799398195,
      "loss": 2.9251,
      "theoretical_loss": 3.7479866443633245,
      "tokens_seen": 760545280
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003886559679037112,
      "loss": 2.9993,
      "theoretical_loss": 3.7479543372155293,
      "tokens_seen": 760610816
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003886459378134403,
      "loss": 3.1674,
      "theoretical_loss": 3.747922033630622,
      "tokens_seen": 760676352
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038863590772316955,
      "loss": 2.9913,
      "theoretical_loss": 3.7478897336079013,
      "tokens_seen": 760741888
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003886258776328987,
      "loss": 2.9375,
      "theoretical_loss": 3.747857437146669,
      "tokens_seen": 760807424
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003886158475426279,
      "loss": 3.1737,
      "theoretical_loss": 3.7478251442462245,
      "tokens_seen": 760872960
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003886058174523571,
      "loss": 3.0285,
      "theoretical_loss": 3.747792854905869,
      "tokens_seen": 760938496
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038859578736208627,
      "loss": 2.887,
      "theoretical_loss": 3.7477605691249045,
      "tokens_seen": 761004032
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038858575727181545,
      "loss": 2.91,
      "theoretical_loss": 3.7477282869026305,
      "tokens_seen": 761069568
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038857572718154463,
      "loss": 3.0705,
      "theoretical_loss": 3.747696008238349,
      "tokens_seen": 761135104
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003885656970912738,
      "loss": 2.8284,
      "theoretical_loss": 3.7476637331313616,
      "tokens_seen": 761200640
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038855566700100305,
      "loss": 2.9528,
      "theoretical_loss": 3.74763146158097,
      "tokens_seen": 761266176
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003885456369107322,
      "loss": 3.1948,
      "theoretical_loss": 3.7475991935864768,
      "tokens_seen": 761331712
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003885356068204614,
      "loss": 2.8576,
      "theoretical_loss": 3.747566929147183,
      "tokens_seen": 761397248
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003885255767301906,
      "loss": 3.2672,
      "theoretical_loss": 3.747534668262391,
      "tokens_seen": 761462784
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003885155466399198,
      "loss": 2.8973,
      "theoretical_loss": 3.7475024109314043,
      "tokens_seen": 761528320
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038850551654964896,
      "loss": 2.9895,
      "theoretical_loss": 3.7474701571535256,
      "tokens_seen": 761593856
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038849548645937814,
      "loss": 2.9563,
      "theoretical_loss": 3.747437906928057,
      "tokens_seen": 761659392
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003884854563691073,
      "loss": 2.9607,
      "theoretical_loss": 3.747405660254302,
      "tokens_seen": 761724928
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038847542627883655,
      "loss": 2.9354,
      "theoretical_loss": 3.747373417131564,
      "tokens_seen": 761790464
    },
    {
      "epoch": 0.23,
      "objective/train/advantage_avg": 0.4939248859882355,
      "objective/train/docs_used": 439620,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.04797101020813,
      "objective/train/original_loss": 3.04797101020813,
      "objective/train/theoretical_loss": 3.7473411775591465,
      "objective/train/tokens_used": 782316000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2440507411956787,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506330728530884,
      "objective/train/weighted_lm_loss": 3.202017307281494,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 3.7473411775591465,
      "tokens_seen": 761856000
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003884653961885657,
      "loss": 2.9711,
      "theoretical_loss": 3.7473411775591465,
      "tokens_seen": 761856000
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003884553660982949,
      "loss": 3.3058,
      "theoretical_loss": 3.747308941536353,
      "tokens_seen": 761921536
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038844533600802404,
      "loss": 2.8123,
      "theoretical_loss": 3.7472767090624877,
      "tokens_seen": 761987072
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003884353059177533,
      "loss": 3.041,
      "theoretical_loss": 3.747244480136854,
      "tokens_seen": 762052608
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038842527582748246,
      "loss": 3.0797,
      "theoretical_loss": 3.7472122547587574,
      "tokens_seen": 762118144
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038841524573721164,
      "loss": 2.8855,
      "theoretical_loss": 3.7471800329275022,
      "tokens_seen": 762183680
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003884052156469408,
      "loss": 2.8167,
      "theoretical_loss": 3.747147814642392,
      "tokens_seen": 762249216
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038839518555667,
      "loss": 3.0914,
      "theoretical_loss": 3.747115599902733,
      "tokens_seen": 762314752
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003883851554663992,
      "loss": 2.9294,
      "theoretical_loss": 3.747083388707829,
      "tokens_seen": 762380288
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003883751253761284,
      "loss": 3.021,
      "theoretical_loss": 3.747051181056986,
      "tokens_seen": 762445824
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038836509528585755,
      "loss": 3.0369,
      "theoretical_loss": 3.7470189769495104,
      "tokens_seen": 762511360
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003883550651955868,
      "loss": 2.8221,
      "theoretical_loss": 3.746986776384706,
      "tokens_seen": 762576896
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038834503510531596,
      "loss": 3.0324,
      "theoretical_loss": 3.74695457936188,
      "tokens_seen": 762642432
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038833500501504514,
      "loss": 2.7398,
      "theoretical_loss": 3.746922385880338,
      "tokens_seen": 762707968
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003883249749247743,
      "loss": 3.0316,
      "theoretical_loss": 3.746890195939386,
      "tokens_seen": 762773504
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003883149448345035,
      "loss": 2.9395,
      "theoretical_loss": 3.7468580095383315,
      "tokens_seen": 762839040
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003883049147442327,
      "loss": 2.9457,
      "theoretical_loss": 3.7468258266764796,
      "tokens_seen": 762904576
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003882948846539619,
      "loss": 2.9824,
      "theoretical_loss": 3.746793647353139,
      "tokens_seen": 762970112
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038828485456369105,
      "loss": 3.1349,
      "theoretical_loss": 3.7467614715676154,
      "tokens_seen": 763035648
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003882748244734203,
      "loss": 3.0115,
      "theoretical_loss": 3.746729299319216,
      "tokens_seen": 763101184
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003882647943831494,
      "loss": 2.9693,
      "theoretical_loss": 3.7466971306072487,
      "tokens_seen": 763166720
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038825476429287865,
      "loss": 3.1232,
      "theoretical_loss": 3.7466649654310213,
      "tokens_seen": 763232256
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038824473420260783,
      "loss": 2.9248,
      "theoretical_loss": 3.7466328037898418,
      "tokens_seen": 763297792
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.000388234704112337,
      "loss": 2.8849,
      "theoretical_loss": 3.746600645683017,
      "tokens_seen": 763363328
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003882246740220662,
      "loss": 2.882,
      "theoretical_loss": 3.7465684911098567,
      "tokens_seen": 763428864
    },
    {
      "epoch": 0.23,
      "objective/train/advantage_avg": 0.4943615198135376,
      "objective/train/docs_used": 440741,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.7477872371673584,
      "objective/train/original_loss": 2.7477872371673584,
      "objective/train/theoretical_loss": 3.7465363400696683,
      "objective/train/tokens_used": 783954400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2444666028022766,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506788492202759,
      "objective/train/weighted_lm_loss": 2.886955976486206,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0304546356201172,
      "theoretical_loss": 3.7465363400696683,
      "tokens_seen": 763494400
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038821464393179537,
      "loss": 2.9248,
      "theoretical_loss": 3.7465363400696683,
      "tokens_seen": 763494400
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038820461384152455,
      "loss": 3.032,
      "theoretical_loss": 3.74650419256176,
      "tokens_seen": 763559936
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003881945837512538,
      "loss": 2.9646,
      "theoretical_loss": 3.746472048585442,
      "tokens_seen": 763625472
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003881845536609829,
      "loss": 2.632,
      "theoretical_loss": 3.7464399081400224,
      "tokens_seen": 763691008
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038817452357071215,
      "loss": 3.1089,
      "theoretical_loss": 3.7464077712248107,
      "tokens_seen": 763756544
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038816449348044133,
      "loss": 2.9853,
      "theoretical_loss": 3.7463756378391158,
      "tokens_seen": 763822080
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003881544633901705,
      "loss": 2.8801,
      "theoretical_loss": 3.746343507982248,
      "tokens_seen": 763887616
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003881444332998997,
      "loss": 2.9225,
      "theoretical_loss": 3.7463113816535163,
      "tokens_seen": 763953152
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003881344032096289,
      "loss": 3.276,
      "theoretical_loss": 3.746279258852231,
      "tokens_seen": 764018688
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003881243731193581,
      "loss": 2.6946,
      "theoretical_loss": 3.7462471395777026,
      "tokens_seen": 764084224
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003881143430290873,
      "loss": 3.0471,
      "theoretical_loss": 3.746215023829241,
      "tokens_seen": 764149760
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038810431293881647,
      "loss": 3.0011,
      "theoretical_loss": 3.746182911606157,
      "tokens_seen": 764215296
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038809428284854565,
      "loss": 2.8231,
      "theoretical_loss": 3.7461508029077613,
      "tokens_seen": 764280832
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038808425275827483,
      "loss": 3.0382,
      "theoretical_loss": 3.7461186977333645,
      "tokens_seen": 764346368
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.000388074222668004,
      "loss": 2.9711,
      "theoretical_loss": 3.7460865960822782,
      "tokens_seen": 764411904
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038806419257773325,
      "loss": 2.8677,
      "theoretical_loss": 3.7460544979538137,
      "tokens_seen": 764477440
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003880541624874624,
      "loss": 3.047,
      "theoretical_loss": 3.7460224033472826,
      "tokens_seen": 764542976
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003880441323971916,
      "loss": 3.0996,
      "theoretical_loss": 3.7459903122619957,
      "tokens_seen": 764608512
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003880341023069208,
      "loss": 3.0464,
      "theoretical_loss": 3.745958224697266,
      "tokens_seen": 764674048
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038802407221665,
      "loss": 2.8279,
      "theoretical_loss": 3.745926140652405,
      "tokens_seen": 764739584
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038801404212637916,
      "loss": 2.8991,
      "theoretical_loss": 3.7458940601267257,
      "tokens_seen": 764805120
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038800401203610834,
      "loss": 2.9569,
      "theoretical_loss": 3.745861983119539,
      "tokens_seen": 764870656
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003879939819458375,
      "loss": 3.1643,
      "theoretical_loss": 3.7458299096301597,
      "tokens_seen": 764936192
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038798395185556675,
      "loss": 2.7977,
      "theoretical_loss": 3.7457978396578993,
      "tokens_seen": 765001728
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003879739217652959,
      "loss": 2.9777,
      "theoretical_loss": 3.7457657732020713,
      "tokens_seen": 765067264
    },
    {
      "epoch": 0.23,
      "objective/train/advantage_avg": 0.49375119805336,
      "objective/train/docs_used": 441342,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.8227758407592773,
      "objective/train/original_loss": 2.8227756023406982,
      "objective/train/theoretical_loss": 3.7457337102619888,
      "objective/train/tokens_used": 785592800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24385963380336761,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506147146224976,
      "objective/train/weighted_lm_loss": 2.9654924869537354,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 3.7457337102619888,
      "tokens_seen": 765132800
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003879638916750251,
      "loss": 2.9531,
      "theoretical_loss": 3.7457337102619888,
      "tokens_seen": 765132800
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038795386158475424,
      "loss": 3.019,
      "theoretical_loss": 3.745701650836965,
      "tokens_seen": 765198336
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003879438314944835,
      "loss": 2.9556,
      "theoretical_loss": 3.745669594926314,
      "tokens_seen": 765263872
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038793380140421266,
      "loss": 2.6558,
      "theoretical_loss": 3.7456375425293493,
      "tokens_seen": 765329408
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038792377131394184,
      "loss": 2.9847,
      "theoretical_loss": 3.7456054936453853,
      "tokens_seen": 765394944
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.000387913741223671,
      "loss": 2.7819,
      "theoretical_loss": 3.745573448273736,
      "tokens_seen": 765460480
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003879037111334002,
      "loss": 2.9594,
      "theoretical_loss": 3.745541406413716,
      "tokens_seen": 765526016
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003878936810431294,
      "loss": 3.2658,
      "theoretical_loss": 3.7455093680646394,
      "tokens_seen": 765591552
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003878836509528586,
      "loss": 3.0511,
      "theoretical_loss": 3.7454773332258213,
      "tokens_seen": 765657088
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038787362086258775,
      "loss": 2.9729,
      "theoretical_loss": 3.745445301896577,
      "tokens_seen": 765722624
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.000387863590772317,
      "loss": 3.2146,
      "theoretical_loss": 3.7454132740762214,
      "tokens_seen": 765788160
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038785356068204616,
      "loss": 2.9712,
      "theoretical_loss": 3.74538124976407,
      "tokens_seen": 765853696
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038784353059177534,
      "loss": 2.9711,
      "theoretical_loss": 3.745349228959438,
      "tokens_seen": 765919232
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003878335005015045,
      "loss": 2.9393,
      "theoretical_loss": 3.745317211661642,
      "tokens_seen": 765984768
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003878234704112337,
      "loss": 3.001,
      "theoretical_loss": 3.7452851978699964,
      "tokens_seen": 766050304
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003878134403209629,
      "loss": 2.7182,
      "theoretical_loss": 3.7452531875838186,
      "tokens_seen": 766115840
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003878034102306921,
      "loss": 3.1996,
      "theoretical_loss": 3.7452211808024254,
      "tokens_seen": 766181376
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038779338014042125,
      "loss": 2.8449,
      "theoretical_loss": 3.7451891775251323,
      "tokens_seen": 766246912
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003877833500501505,
      "loss": 2.8763,
      "theoretical_loss": 3.7451571777512553,
      "tokens_seen": 766312448
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003877733199598796,
      "loss": 3.0183,
      "theoretical_loss": 3.7451251814801134,
      "tokens_seen": 766377984
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038776328986960885,
      "loss": 2.9318,
      "theoretical_loss": 3.7450931887110226,
      "tokens_seen": 766443520
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038775325977933803,
      "loss": 2.9273,
      "theoretical_loss": 3.7450611994433,
      "tokens_seen": 766509056
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003877432296890672,
      "loss": 3.0193,
      "theoretical_loss": 3.745029213676263,
      "tokens_seen": 766574592
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003877331995987964,
      "loss": 2.993,
      "theoretical_loss": 3.7449972314092297,
      "tokens_seen": 766640128
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038772316950852557,
      "loss": 2.717,
      "theoretical_loss": 3.744965252641518,
      "tokens_seen": 766705664
    },
    {
      "epoch": 0.23,
      "objective/train/advantage_avg": 0.49413394927978516,
      "objective/train/docs_used": 442736,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.2514359951019287,
      "objective/train/original_loss": 3.2514359951019287,
      "objective/train/theoretical_loss": 3.7449332773724455,
      "objective/train/tokens_used": 787231200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24422357976436615,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506548881530762,
      "objective/train/weighted_lm_loss": 3.4160921573638916,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0451053380966187,
      "theoretical_loss": 3.7449332773724455,
      "tokens_seen": 766771200
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038771313941825475,
      "loss": 3.1092,
      "theoretical_loss": 3.7449332773724455,
      "tokens_seen": 766771200
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.000387703109327984,
      "loss": 2.9361,
      "theoretical_loss": 3.7449013056013314,
      "tokens_seen": 766836736
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003876930792377131,
      "loss": 2.9052,
      "theoretical_loss": 3.744869337327493,
      "tokens_seen": 766902272
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038768304914744235,
      "loss": 3.0811,
      "theoretical_loss": 3.7448373725502497,
      "tokens_seen": 766967808
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038767301905717153,
      "loss": 3.2189,
      "theoretical_loss": 3.7448054112689197,
      "tokens_seen": 767033344
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003876629889669007,
      "loss": 2.9358,
      "theoretical_loss": 3.7447734534828223,
      "tokens_seen": 767098880
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003876529588766299,
      "loss": 3.1529,
      "theoretical_loss": 3.744741499191277,
      "tokens_seen": 767164416
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003876429287863591,
      "loss": 2.9914,
      "theoretical_loss": 3.7447095483936037,
      "tokens_seen": 767229952
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038763289869608826,
      "loss": 2.8604,
      "theoretical_loss": 3.74467760108912,
      "tokens_seen": 767295488
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003876228686058175,
      "loss": 2.859,
      "theoretical_loss": 3.7446456572771476,
      "tokens_seen": 767361024
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003876128385155466,
      "loss": 2.8482,
      "theoretical_loss": 3.744613716957006,
      "tokens_seen": 767426560
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038760280842527585,
      "loss": 2.8416,
      "theoretical_loss": 3.744581780128015,
      "tokens_seen": 767492096
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.000387592778335005,
      "loss": 2.8756,
      "theoretical_loss": 3.7445498467894947,
      "tokens_seen": 767557632
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003875827482447342,
      "loss": 3.0067,
      "theoretical_loss": 3.7445179169407665,
      "tokens_seen": 767623168
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003875727181544634,
      "loss": 3.1097,
      "theoretical_loss": 3.7444859905811505,
      "tokens_seen": 767688704
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003875626880641926,
      "loss": 2.9952,
      "theoretical_loss": 3.7444540677099676,
      "tokens_seen": 767754240
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038755265797392176,
      "loss": 2.9605,
      "theoretical_loss": 3.7444221483265396,
      "tokens_seen": 767819776
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.000387542627883651,
      "loss": 2.7591,
      "theoretical_loss": 3.744390232430187,
      "tokens_seen": 767885312
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003875325977933801,
      "loss": 2.9009,
      "theoretical_loss": 3.7443583200202317,
      "tokens_seen": 767950848
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038752256770310936,
      "loss": 3.0335,
      "theoretical_loss": 3.7443264110959955,
      "tokens_seen": 768016384
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003875125376128385,
      "loss": 2.9695,
      "theoretical_loss": 3.7442945056568,
      "tokens_seen": 768081920
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003875025075225677,
      "loss": 2.665,
      "theoretical_loss": 3.7442626037019675,
      "tokens_seen": 768147456
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003874924774322969,
      "loss": 2.9339,
      "theoretical_loss": 3.7442307052308195,
      "tokens_seen": 768212992
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003874824473420261,
      "loss": 3.1987,
      "theoretical_loss": 3.744198810242679,
      "tokens_seen": 768278528
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038747241725175526,
      "loss": 2.8359,
      "theoretical_loss": 3.744166918736869,
      "tokens_seen": 768344064
    },
    {
      "epoch": 0.23,
      "objective/train/advantage_avg": 0.49445682764053345,
      "objective/train/docs_used": 443303,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.520820140838623,
      "objective/train/original_loss": 2.520820140838623,
      "objective/train/theoretical_loss": 3.7441350307127115,
      "objective/train/tokens_used": 788869600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24455107748508453,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506888628005981,
      "objective/train/weighted_lm_loss": 2.6485342979431152,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0363596677780151,
      "theoretical_loss": 3.7441350307127115,
      "tokens_seen": 768409600
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038746238716148444,
      "loss": 2.8877,
      "theoretical_loss": 3.7441350307127115,
      "tokens_seen": 768409600
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003874523570712136,
      "loss": 3.2208,
      "theoretical_loss": 3.7441031461695307,
      "tokens_seen": 768475136
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038744232698094286,
      "loss": 2.873,
      "theoretical_loss": 3.7440712651066486,
      "tokens_seen": 768540672
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.000387432296890672,
      "loss": 3.0154,
      "theoretical_loss": 3.7440393875233893,
      "tokens_seen": 768606208
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003874222668004012,
      "loss": 2.8562,
      "theoretical_loss": 3.744007513419076,
      "tokens_seen": 768671744
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038741223671013035,
      "loss": 2.9157,
      "theoretical_loss": 3.743975642793032,
      "tokens_seen": 768737280
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003874022066198596,
      "loss": 2.8355,
      "theoretical_loss": 3.743943775644582,
      "tokens_seen": 768802816
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038739217652958876,
      "loss": 3.0705,
      "theoretical_loss": 3.7439119119730506,
      "tokens_seen": 768868352
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038738214643931795,
      "loss": 2.8503,
      "theoretical_loss": 3.7438800517777615,
      "tokens_seen": 768933888
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003873721163490472,
      "loss": 3.2415,
      "theoretical_loss": 3.743848195058039,
      "tokens_seen": 768999424
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038736208625877636,
      "loss": 2.9573,
      "theoretical_loss": 3.7438163418132078,
      "tokens_seen": 769064960
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038735205616850554,
      "loss": 3.0088,
      "theoretical_loss": 3.743784492042593,
      "tokens_seen": 769130496
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003873420260782347,
      "loss": 2.988,
      "theoretical_loss": 3.7437526457455204,
      "tokens_seen": 769196032
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003873319959879639,
      "loss": 2.8457,
      "theoretical_loss": 3.743720802921314,
      "tokens_seen": 769261568
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003873219658976931,
      "loss": 2.6675,
      "theoretical_loss": 3.7436889635693005,
      "tokens_seen": 769327104
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003873119358074223,
      "loss": 2.7609,
      "theoretical_loss": 3.743657127688804,
      "tokens_seen": 769392640
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038730190571715145,
      "loss": 2.6871,
      "theoretical_loss": 3.743625295279152,
      "tokens_seen": 769458176
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003872918756268807,
      "loss": 2.8747,
      "theoretical_loss": 3.7435934663396693,
      "tokens_seen": 769523712
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003872818455366098,
      "loss": 2.8733,
      "theoretical_loss": 3.743561640869683,
      "tokens_seen": 769589248
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038727181544633905,
      "loss": 3.0265,
      "theoretical_loss": 3.7435298188685184,
      "tokens_seen": 769654784
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038726178535606823,
      "loss": 2.9138,
      "theoretical_loss": 3.743498000335504,
      "tokens_seen": 769720320
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003872517552657974,
      "loss": 3.0216,
      "theoretical_loss": 3.7434661852699644,
      "tokens_seen": 769785856
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003872417251755266,
      "loss": 3.2407,
      "theoretical_loss": 3.743434373671228,
      "tokens_seen": 769851392
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038723169508525577,
      "loss": 3.2388,
      "theoretical_loss": 3.7434025655386214,
      "tokens_seen": 769916928
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038722166499498495,
      "loss": 2.9083,
      "theoretical_loss": 3.743370760871472,
      "tokens_seen": 769982464
    },
    {
      "epoch": 0.23,
      "objective/train/advantage_avg": 0.4914449155330658,
      "objective/train/docs_used": 444724,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.7034969329833984,
      "objective/train/original_loss": 2.7034964561462402,
      "objective/train/theoretical_loss": 3.7433389596691073,
      "objective/train/tokens_used": 790508000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2416326403617859,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0503727197647095,
      "objective/train/weighted_lm_loss": 2.8395755290985107,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0363596677780151,
      "theoretical_loss": 3.7433389596691073,
      "tokens_seen": 770048000
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003872116349047142,
      "loss": 3.0799,
      "theoretical_loss": 3.7433389596691073,
      "tokens_seen": 770048000
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003872016048144433,
      "loss": 2.7873,
      "theoretical_loss": 3.7433071619308556,
      "tokens_seen": 770113536
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038719157472417255,
      "loss": 2.8901,
      "theoretical_loss": 3.743275367656044,
      "tokens_seen": 770179072
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038718154463390173,
      "loss": 2.9546,
      "theoretical_loss": 3.7432435768440007,
      "tokens_seen": 770244608
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003871715145436309,
      "loss": 2.8229,
      "theoretical_loss": 3.7432117894940546,
      "tokens_seen": 770310144
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003871614844533601,
      "loss": 3.0548,
      "theoretical_loss": 3.7431800056055335,
      "tokens_seen": 770375680
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003871514543630893,
      "loss": 3.243,
      "theoretical_loss": 3.743148225177766,
      "tokens_seen": 770441216
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038714142427281846,
      "loss": 2.9007,
      "theoretical_loss": 3.743116448210082,
      "tokens_seen": 770506752
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003871313941825477,
      "loss": 2.8207,
      "theoretical_loss": 3.7430846747018096,
      "tokens_seen": 770572288
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003871213640922768,
      "loss": 3.102,
      "theoretical_loss": 3.743052904652278,
      "tokens_seen": 770637824
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038711133400200605,
      "loss": 2.9041,
      "theoretical_loss": 3.7430211380608167,
      "tokens_seen": 770703360
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003871013039117352,
      "loss": 2.8519,
      "theoretical_loss": 3.742989374926756,
      "tokens_seen": 770768896
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003870912738214644,
      "loss": 2.8251,
      "theoretical_loss": 3.7429576152494244,
      "tokens_seen": 770834432
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003870812437311936,
      "loss": 2.9117,
      "theoretical_loss": 3.742925859028153,
      "tokens_seen": 770899968
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003870712136409228,
      "loss": 2.926,
      "theoretical_loss": 3.7428941062622716,
      "tokens_seen": 770965504
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038706118355065196,
      "loss": 3.2416,
      "theoretical_loss": 3.74286235695111,
      "tokens_seen": 771031040
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003870511534603812,
      "loss": 3.0153,
      "theoretical_loss": 3.742830611094,
      "tokens_seen": 771096576
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003870411233701103,
      "loss": 2.8841,
      "theoretical_loss": 3.7427988686902713,
      "tokens_seen": 771162112
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038703109327983956,
      "loss": 3.3916,
      "theoretical_loss": 3.7427671297392546,
      "tokens_seen": 771227648
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003870210631895687,
      "loss": 2.7927,
      "theoretical_loss": 3.742735394240282,
      "tokens_seen": 771293184
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003870110330992979,
      "loss": 3.141,
      "theoretical_loss": 3.7427036621926835,
      "tokens_seen": 771358720
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003870010030090271,
      "loss": 2.9371,
      "theoretical_loss": 3.7426719335957914,
      "tokens_seen": 771424256
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003869909729187563,
      "loss": 2.9912,
      "theoretical_loss": 3.742640208448938,
      "tokens_seen": 771489792
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038698094282848546,
      "loss": 2.9147,
      "theoretical_loss": 3.742608486751453,
      "tokens_seen": 771555328
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038697091273821464,
      "loss": 2.877,
      "theoretical_loss": 3.742576768502671,
      "tokens_seen": 771620864
    },
    {
      "epoch": 0.23,
      "objective/train/advantage_avg": 0.49528515338897705,
      "objective/train/docs_used": 445347,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.815812587738037,
      "objective/train/original_loss": 2.815812587738037,
      "objective/train/theoretical_loss": 3.7425450537019223,
      "objective/train/tokens_used": 792146400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2453644871711731,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507758855819702,
      "objective/train/weighted_lm_loss": 2.958768606185913,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 3.7425450537019223,
      "tokens_seen": 771686400
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003869608826479438,
      "loss": 2.8991,
      "theoretical_loss": 3.7425450537019223,
      "tokens_seen": 771686400
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038695085255767306,
      "loss": 2.636,
      "theoretical_loss": 3.74251334234854,
      "tokens_seen": 771751936
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003869408224674022,
      "loss": 3.1329,
      "theoretical_loss": 3.742481634441857,
      "tokens_seen": 771817472
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003869307923771314,
      "loss": 2.9772,
      "theoretical_loss": 3.742449929981205,
      "tokens_seen": 771883008
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038692076228686055,
      "loss": 3.0371,
      "theoretical_loss": 3.742418228965918,
      "tokens_seen": 771948544
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003869107321965898,
      "loss": 2.967,
      "theoretical_loss": 3.742386531395329,
      "tokens_seen": 772014080
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038690070210631896,
      "loss": 2.8573,
      "theoretical_loss": 3.7423548372687705,
      "tokens_seen": 772079616
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038689067201604815,
      "loss": 2.8036,
      "theoretical_loss": 3.7423231465855773,
      "tokens_seen": 772145152
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038688064192577733,
      "loss": 3.2418,
      "theoretical_loss": 3.742291459345082,
      "tokens_seen": 772210688
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038687061183550656,
      "loss": 3.028,
      "theoretical_loss": 3.7422597755466187,
      "tokens_seen": 772276224
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003868605817452357,
      "loss": 3.2471,
      "theoretical_loss": 3.742228095189522,
      "tokens_seen": 772341760
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003868505516549649,
      "loss": 2.8973,
      "theoretical_loss": 3.7421964182731253,
      "tokens_seen": 772407296
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038684052156469405,
      "loss": 2.8339,
      "theoretical_loss": 3.7421647447967636,
      "tokens_seen": 772472832
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003868304914744233,
      "loss": 3.2601,
      "theoretical_loss": 3.7421330747597716,
      "tokens_seen": 772538368
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038682046138415247,
      "loss": 2.8738,
      "theoretical_loss": 3.742101408161484,
      "tokens_seen": 772603904
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038681043129388165,
      "loss": 3.3015,
      "theoretical_loss": 3.7420697450012357,
      "tokens_seen": 772669440
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038680040120361083,
      "loss": 2.9184,
      "theoretical_loss": 3.742038085278362,
      "tokens_seen": 772734976
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038679037111334,
      "loss": 3.0247,
      "theoretical_loss": 3.742006428992198,
      "tokens_seen": 772800512
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003867803410230692,
      "loss": 2.8484,
      "theoretical_loss": 3.7419747761420794,
      "tokens_seen": 772866048
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038677031093279843,
      "loss": 2.8659,
      "theoretical_loss": 3.7419431267273415,
      "tokens_seen": 772931584
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038676028084252755,
      "loss": 3.06,
      "theoretical_loss": 3.741911480747321,
      "tokens_seen": 772997120
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003867502507522568,
      "loss": 3.0705,
      "theoretical_loss": 3.7418798382013536,
      "tokens_seen": 773062656
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003867402206619859,
      "loss": 2.9296,
      "theoretical_loss": 3.741848199088776,
      "tokens_seen": 773128192
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038673019057171515,
      "loss": 3.0648,
      "theoretical_loss": 3.741816563408924,
      "tokens_seen": 773193728
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038672016048144433,
      "loss": 3.1215,
      "theoretical_loss": 3.7417849311611344,
      "tokens_seen": 773259264
    },
    {
      "epoch": 0.23,
      "objective/train/advantage_avg": 0.49385911226272583,
      "objective/train/docs_used": 446623,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.980241537094116,
      "objective/train/original_loss": 2.980241298675537,
      "objective/train/theoretical_loss": 3.7417533023447445,
      "objective/train/tokens_used": 793784800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24400487542152405,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506263971328735,
      "objective/train/weighted_lm_loss": 3.1312270164489746,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0338950157165527,
      "theoretical_loss": 3.7417533023447445,
      "tokens_seen": 773324800
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003867101303911735,
      "loss": 2.9415,
      "theoretical_loss": 3.7417533023447445,
      "tokens_seen": 773324800
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003867001003009027,
      "loss": 2.7165,
      "theoretical_loss": 3.7417216769590906,
      "tokens_seen": 773390336
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038669007021063193,
      "loss": 2.7759,
      "theoretical_loss": 3.7416900550035104,
      "tokens_seen": 773455872
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038668004012036106,
      "loss": 2.9061,
      "theoretical_loss": 3.741658436477341,
      "tokens_seen": 773521408
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003866700100300903,
      "loss": 2.7986,
      "theoretical_loss": 3.7416268213799206,
      "tokens_seen": 773586944
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003866599799398194,
      "loss": 2.8788,
      "theoretical_loss": 3.741595209710586,
      "tokens_seen": 773652480
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038664994984954866,
      "loss": 3.0044,
      "theoretical_loss": 3.741563601468676,
      "tokens_seen": 773718016
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038663991975927784,
      "loss": 3.0554,
      "theoretical_loss": 3.7415319966535288,
      "tokens_seen": 773783552
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.000386629889669007,
      "loss": 2.8574,
      "theoretical_loss": 3.741500395264481,
      "tokens_seen": 773849088
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038661985957873625,
      "loss": 2.9773,
      "theoretical_loss": 3.7414687973008736,
      "tokens_seen": 773914624
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003866098294884654,
      "loss": 2.8798,
      "theoretical_loss": 3.741437202762043,
      "tokens_seen": 773980160
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003865997993981946,
      "loss": 2.8156,
      "theoretical_loss": 3.74140561164733,
      "tokens_seen": 774045696
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003865897693079238,
      "loss": 2.6046,
      "theoretical_loss": 3.7413740239560727,
      "tokens_seen": 774111232
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.000386579739217653,
      "loss": 2.8683,
      "theoretical_loss": 3.7413424396876103,
      "tokens_seen": 774176768
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038656970912738216,
      "loss": 2.9315,
      "theoretical_loss": 3.741310858841282,
      "tokens_seen": 774242304
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003865596790371114,
      "loss": 2.8163,
      "theoretical_loss": 3.7412792814164275,
      "tokens_seen": 774307840
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003865496489468405,
      "loss": 3.1657,
      "theoretical_loss": 3.741247707412387,
      "tokens_seen": 774373376
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038653961885656976,
      "loss": 2.9491,
      "theoretical_loss": 3.7412161368285,
      "tokens_seen": 774438912
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003865295887662989,
      "loss": 2.7825,
      "theoretical_loss": 3.7411845696641075,
      "tokens_seen": 774504448
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003865195586760281,
      "loss": 2.9628,
      "theoretical_loss": 3.741153005918549,
      "tokens_seen": 774569984
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003865095285857573,
      "loss": 3.2633,
      "theoretical_loss": 3.7411214455911646,
      "tokens_seen": 774635520
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003864994984954865,
      "loss": 3.1619,
      "theoretical_loss": 3.741089888681296,
      "tokens_seen": 774701056
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038648946840521566,
      "loss": 3.1609,
      "theoretical_loss": 3.741058335188284,
      "tokens_seen": 774766592
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038647943831494484,
      "loss": 2.9583,
      "theoretical_loss": 3.741026785111469,
      "tokens_seen": 774832128
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.000386469408224674,
      "loss": 2.7248,
      "theoretical_loss": 3.7409952384501923,
      "tokens_seen": 774897664
    },
    {
      "epoch": 0.23,
      "objective/train/advantage_avg": 0.49400949478149414,
      "objective/train/docs_used": 447277,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.1595654487609863,
      "objective/train/original_loss": 3.1595652103424072,
      "objective/train/theoretical_loss": 3.7409636952037957,
      "objective/train/tokens_used": 795423200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24411392211914062,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506418943405151,
      "objective/train/weighted_lm_loss": 3.31968355178833,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0437887907028198,
      "theoretical_loss": 3.7409636952037957,
      "tokens_seen": 774963200
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038645937813440326,
      "loss": 3.1316,
      "theoretical_loss": 3.7409636952037957,
      "tokens_seen": 774963200
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003864493480441324,
      "loss": 2.9298,
      "theoretical_loss": 3.740932155371621,
      "tokens_seen": 775028736
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.0003864393179538616,
      "loss": 2.8261,
      "theoretical_loss": 3.7409006189530096,
      "tokens_seen": 775094272
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038642928786359075,
      "loss": 2.8018,
      "theoretical_loss": 3.7408690859473035,
      "tokens_seen": 775159808
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038641925777332,
      "loss": 2.9881,
      "theoretical_loss": 3.7408375563538447,
      "tokens_seen": 775225344
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038640922768304917,
      "loss": 2.8456,
      "theoretical_loss": 3.740806030171976,
      "tokens_seen": 775290880
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038639919759277835,
      "loss": 2.8911,
      "theoretical_loss": 3.7407745074010395,
      "tokens_seen": 775356416
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038638916750250753,
      "loss": 2.8664,
      "theoretical_loss": 3.7407429880403775,
      "tokens_seen": 775421952
    },
    {
      "epoch": 0.23,
      "learning_rate": 0.00038637913741223676,
      "loss": 2.8526,
      "theoretical_loss": 3.7407114720893344,
      "tokens_seen": 775487488
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003863691073219659,
      "loss": 2.8768,
      "theoretical_loss": 3.740679959547252,
      "tokens_seen": 775553024
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003863590772316951,
      "loss": 2.9738,
      "theoretical_loss": 3.740648450413474,
      "tokens_seen": 775618560
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038634904714142425,
      "loss": 2.8687,
      "theoretical_loss": 3.7406169446873436,
      "tokens_seen": 775684096
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003863390170511535,
      "loss": 2.8911,
      "theoretical_loss": 3.740585442368204,
      "tokens_seen": 775749632
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038632898696088267,
      "loss": 3.0819,
      "theoretical_loss": 3.7405539434554003,
      "tokens_seen": 775815168
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038631895687061185,
      "loss": 2.6989,
      "theoretical_loss": 3.7405224479482753,
      "tokens_seen": 775880704
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038630892678034103,
      "loss": 2.7665,
      "theoretical_loss": 3.740490955846173,
      "tokens_seen": 775946240
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003862988966900702,
      "loss": 2.868,
      "theoretical_loss": 3.740459467148439,
      "tokens_seen": 776011776
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003862888665997994,
      "loss": 2.8765,
      "theoretical_loss": 3.740427981854417,
      "tokens_seen": 776077312
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038627883650952863,
      "loss": 2.8575,
      "theoretical_loss": 3.7403964999634516,
      "tokens_seen": 776142848
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038626880641925775,
      "loss": 2.7003,
      "theoretical_loss": 3.7403650214748874,
      "tokens_seen": 776208384
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.000386258776328987,
      "loss": 2.7287,
      "theoretical_loss": 3.7403335463880705,
      "tokens_seen": 776273920
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003862487462387161,
      "loss": 3.2116,
      "theoretical_loss": 3.7403020747023454,
      "tokens_seen": 776339456
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038623871614844535,
      "loss": 3.0604,
      "theoretical_loss": 3.740270606417058,
      "tokens_seen": 776404992
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038622868605817453,
      "loss": 2.652,
      "theoretical_loss": 3.7402391415315535,
      "tokens_seen": 776470528
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003862186559679037,
      "loss": 3.0029,
      "theoretical_loss": 3.740207680045178,
      "tokens_seen": 776536064
    },
    {
      "epoch": 0.24,
      "objective/train/advantage_avg": 0.49554675817489624,
      "objective/train/docs_used": 447858,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.4326231479644775,
      "objective/train/original_loss": 2.4326233863830566,
      "objective/train/theoretical_loss": 3.7401762219572765,
      "objective/train/tokens_used": 797061600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24560746550559998,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0508031845092773,
      "objective/train/weighted_lm_loss": 2.556159496307373,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 3.7401762219572765,
      "tokens_seen": 776601600
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003862086258776329,
      "loss": 2.9691,
      "theoretical_loss": 3.7401762219572765,
      "tokens_seen": 776601600
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038619859578736213,
      "loss": 3.0625,
      "theoretical_loss": 3.740144767267197,
      "tokens_seen": 776667136
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038618856569709126,
      "loss": 2.8676,
      "theoretical_loss": 3.740113315974284,
      "tokens_seen": 776732672
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003861785356068205,
      "loss": 2.9282,
      "theoretical_loss": 3.7400818680778847,
      "tokens_seen": 776798208
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003861685055165496,
      "loss": 3.0946,
      "theoretical_loss": 3.740050423577346,
      "tokens_seen": 776863744
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038615847542627886,
      "loss": 3.1572,
      "theoretical_loss": 3.740018982472015,
      "tokens_seen": 776929280
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038614844533600804,
      "loss": 2.8627,
      "theoretical_loss": 3.739987544761238,
      "tokens_seen": 776994816
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003861384152457372,
      "loss": 2.9508,
      "theoretical_loss": 3.739956110444363,
      "tokens_seen": 777060352
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003861283851554664,
      "loss": 3.0787,
      "theoretical_loss": 3.739924679520737,
      "tokens_seen": 777125888
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003861183550651956,
      "loss": 2.8512,
      "theoretical_loss": 3.739893251989707,
      "tokens_seen": 777191424
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038610832497492476,
      "loss": 2.9211,
      "theoretical_loss": 3.739861827850622,
      "tokens_seen": 777256960
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.000386098294884654,
      "loss": 3.1149,
      "theoretical_loss": 3.739830407102829,
      "tokens_seen": 777322496
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003860882647943831,
      "loss": 2.9426,
      "theoretical_loss": 3.739798989745677,
      "tokens_seen": 777388032
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038607823470411236,
      "loss": 3.1342,
      "theoretical_loss": 3.7397675757785134,
      "tokens_seen": 777453568
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003860682046138415,
      "loss": 2.877,
      "theoretical_loss": 3.7397361652006875,
      "tokens_seen": 777519104
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003860581745235707,
      "loss": 3.0659,
      "theoretical_loss": 3.7397047580115474,
      "tokens_seen": 777584640
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003860481444332999,
      "loss": 3.1598,
      "theoretical_loss": 3.739673354210442,
      "tokens_seen": 777650176
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003860381143430291,
      "loss": 3.0575,
      "theoretical_loss": 3.7396419537967214,
      "tokens_seen": 777715712
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038602808425275826,
      "loss": 2.9678,
      "theoretical_loss": 3.739610556769733,
      "tokens_seen": 777781248
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003860180541624875,
      "loss": 2.8449,
      "theoretical_loss": 3.7395791631288278,
      "tokens_seen": 777846784
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003860080240722166,
      "loss": 2.9369,
      "theoretical_loss": 3.7395477728733537,
      "tokens_seen": 777912320
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038599799398194586,
      "loss": 2.9902,
      "theoretical_loss": 3.7395163860026623,
      "tokens_seen": 777977856
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.000385987963891675,
      "loss": 2.8236,
      "theoretical_loss": 3.7394850025161026,
      "tokens_seen": 778043392
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003859779338014042,
      "loss": 2.8846,
      "theoretical_loss": 3.7394536224130244,
      "tokens_seen": 778108928
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003859679037111334,
      "loss": 2.9157,
      "theoretical_loss": 3.7394222456927793,
      "tokens_seen": 778174464
    },
    {
      "epoch": 0.24,
      "objective/train/advantage_avg": 0.4932505190372467,
      "objective/train/docs_used": 448758,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.0165815353393555,
      "objective/train/original_loss": 3.0165815353393555,
      "objective/train/theoretical_loss": 3.739390872354716,
      "objective/train/tokens_used": 798700000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24335677921772003,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505621433258057,
      "objective/train/weighted_lm_loss": 3.168957471847534,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.046392798423767,
      "theoretical_loss": 3.739390872354716,
      "tokens_seen": 778240000
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003859578736208626,
      "loss": 2.8335,
      "theoretical_loss": 3.739390872354716,
      "tokens_seen": 778240000
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038594784353059177,
      "loss": 2.964,
      "theoretical_loss": 3.7393595023981865,
      "tokens_seen": 778305536
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038593781344032095,
      "loss": 3.027,
      "theoretical_loss": 3.739328135822541,
      "tokens_seen": 778371072
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038592778335005013,
      "loss": 2.6553,
      "theoretical_loss": 3.7392967726271307,
      "tokens_seen": 778436608
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038591775325977937,
      "loss": 3.064,
      "theoretical_loss": 3.7392654128113074,
      "tokens_seen": 778502144
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003859077231695085,
      "loss": 3.0317,
      "theoretical_loss": 3.7392340563744213,
      "tokens_seen": 778567680
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038589769307923773,
      "loss": 2.9255,
      "theoretical_loss": 3.7392027033158244,
      "tokens_seen": 778633216
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038588766298896685,
      "loss": 2.9455,
      "theoretical_loss": 3.7391713536348687,
      "tokens_seen": 778698752
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003858776328986961,
      "loss": 2.737,
      "theoretical_loss": 3.739140007330906,
      "tokens_seen": 778764288
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003858676028084253,
      "loss": 2.957,
      "theoretical_loss": 3.7391086644032887,
      "tokens_seen": 778829824
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038585757271815445,
      "loss": 3.097,
      "theoretical_loss": 3.739077324851368,
      "tokens_seen": 778895360
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003858475426278837,
      "loss": 3.03,
      "theoretical_loss": 3.7390459886744973,
      "tokens_seen": 778960896
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038583751253761287,
      "loss": 2.9537,
      "theoretical_loss": 3.7390146558720296,
      "tokens_seen": 779026432
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038582748244734205,
      "loss": 2.7507,
      "theoretical_loss": 3.738983326443316,
      "tokens_seen": 779091968
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038581745235707123,
      "loss": 2.9648,
      "theoretical_loss": 3.7389520003877115,
      "tokens_seen": 779157504
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003858074222668004,
      "loss": 3.1079,
      "theoretical_loss": 3.7389206777045683,
      "tokens_seen": 779223040
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003857973921765296,
      "loss": 2.845,
      "theoretical_loss": 3.738889358393239,
      "tokens_seen": 779288576
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038578736208625883,
      "loss": 2.9487,
      "theoretical_loss": 3.738858042453079,
      "tokens_seen": 779354112
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038577733199598796,
      "loss": 3.0626,
      "theoretical_loss": 3.7388267298834403,
      "tokens_seen": 779419648
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003857673019057172,
      "loss": 3.1143,
      "theoretical_loss": 3.7387954206836773,
      "tokens_seen": 779485184
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003857572718154463,
      "loss": 3.1429,
      "theoretical_loss": 3.738764114853144,
      "tokens_seen": 779550720
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038574724172517555,
      "loss": 3.135,
      "theoretical_loss": 3.738732812391195,
      "tokens_seen": 779616256
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038573721163490473,
      "loss": 3.0407,
      "theoretical_loss": 3.738701513297184,
      "tokens_seen": 779681792
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003857271815446339,
      "loss": 3.0771,
      "theoretical_loss": 3.7386702175704665,
      "tokens_seen": 779747328
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003857171514543631,
      "loss": 2.9697,
      "theoretical_loss": 3.738638925210397,
      "tokens_seen": 779812864
    },
    {
      "epoch": 0.24,
      "objective/train/advantage_avg": 0.49543261528015137,
      "objective/train/docs_used": 449318,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.886277437210083,
      "objective/train/original_loss": 2.886277675628662,
      "objective/train/theoretical_loss": 3.73860763621633,
      "objective/train/tokens_used": 800338400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24550488591194153,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507912635803223,
      "objective/train/weighted_lm_loss": 3.0329267978668213,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0338950157165527,
      "theoretical_loss": 3.73860763621633,
      "tokens_seen": 779878400
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038570712136409233,
      "loss": 3.107,
      "theoretical_loss": 3.73860763621633,
      "tokens_seen": 779878400
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038569709127382146,
      "loss": 2.9368,
      "theoretical_loss": 3.738576350587621,
      "tokens_seen": 779943936
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003856870611835507,
      "loss": 2.9069,
      "theoretical_loss": 3.7385450683236248,
      "tokens_seen": 780009472
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003856770310932798,
      "loss": 2.8371,
      "theoretical_loss": 3.7385137894236973,
      "tokens_seen": 780075008
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038566700100300906,
      "loss": 2.9291,
      "theoretical_loss": 3.7384825138871944,
      "tokens_seen": 780140544
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038565697091273824,
      "loss": 3.0515,
      "theoretical_loss": 3.738451241713472,
      "tokens_seen": 780206080
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003856469408224674,
      "loss": 3.1807,
      "theoretical_loss": 3.7384199729018848,
      "tokens_seen": 780271616
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003856369107321966,
      "loss": 2.9102,
      "theoretical_loss": 3.73838870745179,
      "tokens_seen": 780337152
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003856268806419258,
      "loss": 2.9945,
      "theoretical_loss": 3.7383574453625448,
      "tokens_seen": 780402688
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038561685055165496,
      "loss": 3.1011,
      "theoretical_loss": 3.738326186633504,
      "tokens_seen": 780468224
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003856068204613842,
      "loss": 3.0751,
      "theoretical_loss": 3.7382949312640252,
      "tokens_seen": 780533760
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003855967903711133,
      "loss": 2.6619,
      "theoretical_loss": 3.7382636792534654,
      "tokens_seen": 780599296
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038558676028084256,
      "loss": 2.7391,
      "theoretical_loss": 3.7382324306011814,
      "tokens_seen": 780664832
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003855767301905717,
      "loss": 2.8863,
      "theoretical_loss": 3.7382011853065307,
      "tokens_seen": 780730368
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003855667001003009,
      "loss": 2.8197,
      "theoretical_loss": 3.73816994336887,
      "tokens_seen": 780795904
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003855566700100301,
      "loss": 3.0415,
      "theoretical_loss": 3.7381387047875583,
      "tokens_seen": 780861440
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003855466399197593,
      "loss": 2.9698,
      "theoretical_loss": 3.7381074695619523,
      "tokens_seen": 780926976
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038553660982948846,
      "loss": 3.1836,
      "theoretical_loss": 3.73807623769141,
      "tokens_seen": 780992512
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003855265797392177,
      "loss": 2.8221,
      "theoretical_loss": 3.7380450091752904,
      "tokens_seen": 781058048
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003855165496489468,
      "loss": 3.0841,
      "theoretical_loss": 3.73801378401295,
      "tokens_seen": 781123584
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038550651955867606,
      "loss": 3.0257,
      "theoretical_loss": 3.737982562203749,
      "tokens_seen": 781189120
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003854964894684052,
      "loss": 2.9411,
      "theoretical_loss": 3.7379513437470457,
      "tokens_seen": 781254656
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003854864593781344,
      "loss": 3.088,
      "theoretical_loss": 3.737920128642199,
      "tokens_seen": 781320192
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003854764292878636,
      "loss": 2.9611,
      "theoretical_loss": 3.737888916888567,
      "tokens_seen": 781385728
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003854663991975928,
      "loss": 2.8133,
      "theoretical_loss": 3.7378577084855102,
      "tokens_seen": 781451264
    },
    {
      "epoch": 0.24,
      "objective/train/advantage_avg": 0.49570170044898987,
      "objective/train/docs_used": 450726,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.3402411937713623,
      "objective/train/original_loss": 2.340240955352783,
      "objective/train/theoretical_loss": 3.7378265034323865,
      "objective/train/tokens_used": 801976800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24576336145401,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0508193969726562,
      "objective/train/weighted_lm_loss": 2.459167718887329,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0408108234405518,
      "theoretical_loss": 3.7378265034323865,
      "tokens_seen": 781516800
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038545636910732197,
      "loss": 2.9704,
      "theoretical_loss": 3.7378265034323865,
      "tokens_seen": 781516800
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038544633901705115,
      "loss": 3.0101,
      "theoretical_loss": 3.737795301728557,
      "tokens_seen": 781582336
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038543630892678033,
      "loss": 2.8406,
      "theoretical_loss": 3.7377641033733804,
      "tokens_seen": 781647872
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038542627883650957,
      "loss": 3.1946,
      "theoretical_loss": 3.7377329083662167,
      "tokens_seen": 781713408
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003854162487462387,
      "loss": 3.1066,
      "theoretical_loss": 3.737701716706426,
      "tokens_seen": 781778944
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038540621865596793,
      "loss": 2.8436,
      "theoretical_loss": 3.7376705283933696,
      "tokens_seen": 781844480
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038539618856569705,
      "loss": 3.1591,
      "theoretical_loss": 3.737639343426406,
      "tokens_seen": 781910016
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003853861584754263,
      "loss": 2.9514,
      "theoretical_loss": 3.7376081618048973,
      "tokens_seen": 781975552
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038537612838515547,
      "loss": 2.9591,
      "theoretical_loss": 3.7375769835282036,
      "tokens_seen": 782041088
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038536609829488465,
      "loss": 3.2328,
      "theoretical_loss": 3.7375458085956863,
      "tokens_seen": 782106624
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038535606820461383,
      "loss": 3.1014,
      "theoretical_loss": 3.737514637006706,
      "tokens_seen": 782172160
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038534603811434307,
      "loss": 2.7782,
      "theoretical_loss": 3.737483468760624,
      "tokens_seen": 782237696
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003853360080240722,
      "loss": 2.8513,
      "theoretical_loss": 3.737452303856803,
      "tokens_seen": 782303232
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038532597793380143,
      "loss": 2.8646,
      "theoretical_loss": 3.737421142294603,
      "tokens_seen": 782368768
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038531594784353056,
      "loss": 2.9648,
      "theoretical_loss": 3.7373899840733866,
      "tokens_seen": 782434304
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003853059177532598,
      "loss": 3.3887,
      "theoretical_loss": 3.737358829192516,
      "tokens_seen": 782499840
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.000385295887662989,
      "loss": 2.9748,
      "theoretical_loss": 3.7373276776513533,
      "tokens_seen": 782565376
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038528585757271816,
      "loss": 2.946,
      "theoretical_loss": 3.73729652944926,
      "tokens_seen": 782630912
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038527582748244734,
      "loss": 2.9002,
      "theoretical_loss": 3.7372653845855996,
      "tokens_seen": 782696448
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003852657973921765,
      "loss": 2.9074,
      "theoretical_loss": 3.737234243059735,
      "tokens_seen": 782761984
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003852557673019057,
      "loss": 2.8742,
      "theoretical_loss": 3.7372031048710275,
      "tokens_seen": 782827520
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038524573721163493,
      "loss": 3.0443,
      "theoretical_loss": 3.737171970018842,
      "tokens_seen": 782893056
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038523570712136406,
      "loss": 3.106,
      "theoretical_loss": 3.737140838502541,
      "tokens_seen": 782958592
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003852256770310933,
      "loss": 2.9554,
      "theoretical_loss": 3.7371097103214876,
      "tokens_seen": 783024128
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003852156469408225,
      "loss": 2.923,
      "theoretical_loss": 3.7370785854750457,
      "tokens_seen": 783089664
    },
    {
      "epoch": 0.24,
      "objective/train/advantage_avg": 0.4935194253921509,
      "objective/train/docs_used": 451239,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.809216260910034,
      "objective/train/original_loss": 2.809216022491455,
      "objective/train/theoretical_loss": 3.737047463962579,
      "objective/train/tokens_used": 803615200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2437388151884079,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505908727645874,
      "objective/train/weighted_lm_loss": 2.9512674808502197,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0396549701690674,
      "theoretical_loss": 3.737047463962579,
      "tokens_seen": 783155200
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038520561685055166,
      "loss": 3.2392,
      "theoretical_loss": 3.737047463962579,
      "tokens_seen": 783155200
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038519558676028084,
      "loss": 2.9733,
      "theoretical_loss": 3.7370163457834513,
      "tokens_seen": 783220736
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038518555667001,
      "loss": 3.2528,
      "theoretical_loss": 3.7369852309370275,
      "tokens_seen": 783286272
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003851755265797392,
      "loss": 3.0901,
      "theoretical_loss": 3.7369541194226708,
      "tokens_seen": 783351808
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038516549648946844,
      "loss": 2.9679,
      "theoretical_loss": 3.736923011239746,
      "tokens_seen": 783417344
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038515546639919756,
      "loss": 2.7598,
      "theoretical_loss": 3.7368919063876174,
      "tokens_seen": 783482880
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003851454363089268,
      "loss": 2.9333,
      "theoretical_loss": 3.7368608048656506,
      "tokens_seen": 783548416
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003851354062186559,
      "loss": 3.139,
      "theoretical_loss": 3.7368297066732103,
      "tokens_seen": 783613952
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038512537612838516,
      "loss": 3.0138,
      "theoretical_loss": 3.7367986118096614,
      "tokens_seen": 783679488
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003851153460381144,
      "loss": 2.875,
      "theoretical_loss": 3.7367675202743698,
      "tokens_seen": 783745024
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003851053159478435,
      "loss": 3.1365,
      "theoretical_loss": 3.7367364320666994,
      "tokens_seen": 783810560
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038509528585757276,
      "loss": 2.8663,
      "theoretical_loss": 3.7367053471860183,
      "tokens_seen": 783876096
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003850852557673019,
      "loss": 3.0337,
      "theoretical_loss": 3.73667426563169,
      "tokens_seen": 783941632
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003850752256770311,
      "loss": 2.8417,
      "theoretical_loss": 3.7366431874030823,
      "tokens_seen": 784007168
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003850651955867603,
      "loss": 2.6598,
      "theoretical_loss": 3.73661211249956,
      "tokens_seen": 784072704
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003850551654964895,
      "loss": 3.1044,
      "theoretical_loss": 3.7365810409204903,
      "tokens_seen": 784138240
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038504513540621866,
      "loss": 2.9327,
      "theoretical_loss": 3.73654997266524,
      "tokens_seen": 784203776
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003850351053159479,
      "loss": 3.071,
      "theoretical_loss": 3.736518907733175,
      "tokens_seen": 784269312
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.000385025075225677,
      "loss": 3.017,
      "theoretical_loss": 3.736487846123663,
      "tokens_seen": 784334848
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038501504513540626,
      "loss": 2.7982,
      "theoretical_loss": 3.7364567878360706,
      "tokens_seen": 784400384
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003850050150451354,
      "loss": 3.0946,
      "theoretical_loss": 3.7364257328697645,
      "tokens_seen": 784465920
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003849949849548646,
      "loss": 3.0194,
      "theoretical_loss": 3.7363946812241133,
      "tokens_seen": 784531456
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003849849548645938,
      "loss": 2.7892,
      "theoretical_loss": 3.7363636328984837,
      "tokens_seen": 784596992
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.000384974924774323,
      "loss": 2.9274,
      "theoretical_loss": 3.7363325878922433,
      "tokens_seen": 784662528
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038496489468405217,
      "loss": 3.1109,
      "theoretical_loss": 3.7363015462047606,
      "tokens_seen": 784728064
    },
    {
      "epoch": 0.24,
      "objective/train/advantage_avg": 0.4937745928764343,
      "objective/train/docs_used": 452482,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.9381747245788574,
      "objective/train/original_loss": 2.9381744861602783,
      "objective/train/theoretical_loss": 3.736270507835404,
      "objective/train/tokens_used": 805253600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24389216303825378,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506173372268677,
      "objective/train/weighted_lm_loss": 3.0869300365448,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0338950157165527,
      "theoretical_loss": 3.736270507835404,
      "tokens_seen": 784793600
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038495486459378135,
      "loss": 3.0808,
      "theoretical_loss": 3.736270507835404,
      "tokens_seen": 784793600
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038494483450351053,
      "loss": 2.9059,
      "theoretical_loss": 3.736239472783541,
      "tokens_seen": 784859136
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038493480441323977,
      "loss": 3.0178,
      "theoretical_loss": 3.7362084410485403,
      "tokens_seen": 784924672
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003849247743229689,
      "loss": 2.9907,
      "theoretical_loss": 3.736177412629771,
      "tokens_seen": 784990208
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038491474423269813,
      "loss": 2.8811,
      "theoretical_loss": 3.736146387526601,
      "tokens_seen": 785055744
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038490471414242725,
      "loss": 3.1258,
      "theoretical_loss": 3.7361153657384003,
      "tokens_seen": 785121280
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003848946840521565,
      "loss": 3.0455,
      "theoretical_loss": 3.736084347264537,
      "tokens_seen": 785186816
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038488465396188567,
      "loss": 2.8923,
      "theoretical_loss": 3.7360533321043805,
      "tokens_seen": 785252352
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038487462387161485,
      "loss": 3.0139,
      "theoretical_loss": 3.7360223202573013,
      "tokens_seen": 785317888
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038486459378134403,
      "loss": 2.72,
      "theoretical_loss": 3.7359913117226684,
      "tokens_seen": 785383424
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038485456369107327,
      "loss": 2.8236,
      "theoretical_loss": 3.7359603064998517,
      "tokens_seen": 785448960
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003848445336008024,
      "loss": 3.0911,
      "theoretical_loss": 3.735929304588221,
      "tokens_seen": 785514496
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038483450351053163,
      "loss": 2.9757,
      "theoretical_loss": 3.7358983059871465,
      "tokens_seen": 785580032
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038482447342026076,
      "loss": 3.1422,
      "theoretical_loss": 3.7358673106959985,
      "tokens_seen": 785645568
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038481444332999,
      "loss": 3.022,
      "theoretical_loss": 3.7358363187141483,
      "tokens_seen": 785711104
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003848044132397192,
      "loss": 3.1214,
      "theoretical_loss": 3.7358053300409653,
      "tokens_seen": 785776640
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038479438314944836,
      "loss": 2.7874,
      "theoretical_loss": 3.7357743446758214,
      "tokens_seen": 785842176
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038478435305917754,
      "loss": 3.0404,
      "theoretical_loss": 3.735743362618088,
      "tokens_seen": 785907712
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003847743229689067,
      "loss": 3.1147,
      "theoretical_loss": 3.735712383867135,
      "tokens_seen": 785973248
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003847642928786359,
      "loss": 2.793,
      "theoretical_loss": 3.7356814084223338,
      "tokens_seen": 786038784
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038475426278836513,
      "loss": 2.7326,
      "theoretical_loss": 3.735650436283057,
      "tokens_seen": 786104320
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038474423269809426,
      "loss": 2.893,
      "theoretical_loss": 3.735619467448676,
      "tokens_seen": 786169856
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003847342026078235,
      "loss": 3.0168,
      "theoretical_loss": 3.735588501918562,
      "tokens_seen": 786235392
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003847241725175527,
      "loss": 2.7378,
      "theoretical_loss": 3.7355575396920884,
      "tokens_seen": 786300928
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038471414242728186,
      "loss": 3.1912,
      "theoretical_loss": 3.7355265807686258,
      "tokens_seen": 786366464
    },
    {
      "epoch": 0.24,
      "objective/train/advantage_avg": 0.4949711561203003,
      "objective/train/docs_used": 453141,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.6298294067382812,
      "objective/train/original_loss": 2.6298294067382812,
      "objective/train/theoretical_loss": 3.735495625147548,
      "objective/train/tokens_used": 806892000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2450433075428009,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050742745399475,
      "objective/train/weighted_lm_loss": 2.763174533843994,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0442860126495361,
      "theoretical_loss": 3.735495625147548,
      "tokens_seen": 786432000
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038470411233701104,
      "loss": 2.8746,
      "theoretical_loss": 3.735495625147548,
      "tokens_seen": 786432000
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003846940822467402,
      "loss": 2.8364,
      "theoretical_loss": 3.7354646728282264,
      "tokens_seen": 786497536
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003846840521564694,
      "loss": 2.6615,
      "theoretical_loss": 3.735433723810035,
      "tokens_seen": 786563072
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038467402206619864,
      "loss": 2.9275,
      "theoretical_loss": 3.7354027780923458,
      "tokens_seen": 786628608
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038466399197592776,
      "loss": 2.9329,
      "theoretical_loss": 3.7353718356745325,
      "tokens_seen": 786694144
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.000384653961885657,
      "loss": 2.9909,
      "theoretical_loss": 3.735340896555968,
      "tokens_seen": 786759680
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003846439317953861,
      "loss": 2.9989,
      "theoretical_loss": 3.735309960736026,
      "tokens_seen": 786825216
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038463390170511536,
      "loss": 2.8126,
      "theoretical_loss": 3.7352790282140793,
      "tokens_seen": 786890752
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038462387161484454,
      "loss": 3.0656,
      "theoretical_loss": 3.7352480989895023,
      "tokens_seen": 786956288
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003846138415245737,
      "loss": 2.8602,
      "theoretical_loss": 3.7352171730616694,
      "tokens_seen": 787021824
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003846038114343029,
      "loss": 3.1257,
      "theoretical_loss": 3.735186250429954,
      "tokens_seen": 787087360
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003845937813440321,
      "loss": 3.1315,
      "theoretical_loss": 3.735155331093731,
      "tokens_seen": 787152896
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038458375125376127,
      "loss": 2.8571,
      "theoretical_loss": 3.7351244150523737,
      "tokens_seen": 787218432
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003845737211634905,
      "loss": 2.9689,
      "theoretical_loss": 3.7350935023052583,
      "tokens_seen": 787283968
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038456369107321963,
      "loss": 2.766,
      "theoretical_loss": 3.735062592851759,
      "tokens_seen": 787349504
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038455366098294886,
      "loss": 2.867,
      "theoretical_loss": 3.73503168669125,
      "tokens_seen": 787415040
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038454363089267805,
      "loss": 2.9159,
      "theoretical_loss": 3.735000783823107,
      "tokens_seen": 787480576
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038453360080240723,
      "loss": 2.9399,
      "theoretical_loss": 3.7349698842467056,
      "tokens_seen": 787546112
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003845235707121364,
      "loss": 3.0392,
      "theoretical_loss": 3.7349389879614217,
      "tokens_seen": 787611648
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003845135406218656,
      "loss": 3.0065,
      "theoretical_loss": 3.7349080949666296,
      "tokens_seen": 787677184
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038450351053159477,
      "loss": 2.7702,
      "theoretical_loss": 3.7348772052617063,
      "tokens_seen": 787742720
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.000384493480441324,
      "loss": 2.7723,
      "theoretical_loss": 3.734846318846027,
      "tokens_seen": 787808256
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038448345035105313,
      "loss": 2.6394,
      "theoretical_loss": 3.7348154357189682,
      "tokens_seen": 787873792
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038447342026078237,
      "loss": 3.182,
      "theoretical_loss": 3.7347845558799064,
      "tokens_seen": 787939328
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003844633901705115,
      "loss": 2.8641,
      "theoretical_loss": 3.734753679328218,
      "tokens_seen": 788004864
    },
    {
      "epoch": 0.24,
      "objective/train/advantage_avg": 0.49207961559295654,
      "objective/train/docs_used": 453953,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.894120931625366,
      "objective/train/original_loss": 2.894120931625366,
      "objective/train/theoretical_loss": 3.7347228060632798,
      "objective/train/tokens_used": 808530400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24224688112735748,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0504392385482788,
      "objective/train/weighted_lm_loss": 3.0400209426879883,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0432155132293701,
      "theoretical_loss": 3.7347228060632798,
      "tokens_seen": 788070400
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038445336008024073,
      "loss": 2.8905,
      "theoretical_loss": 3.7347228060632798,
      "tokens_seen": 788070400
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003844433299899699,
      "loss": 2.9526,
      "theoretical_loss": 3.7346919360844684,
      "tokens_seen": 788135936
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003844332998996991,
      "loss": 2.7383,
      "theoretical_loss": 3.734661069391161,
      "tokens_seen": 788201472
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003844232698094283,
      "loss": 3.1129,
      "theoretical_loss": 3.7346302059827345,
      "tokens_seen": 788267008
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038441323971915745,
      "loss": 2.7446,
      "theoretical_loss": 3.734599345858567,
      "tokens_seen": 788332544
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038440320962888664,
      "loss": 2.795,
      "theoretical_loss": 3.7345684890180344,
      "tokens_seen": 788398080
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038439317953861587,
      "loss": 2.8518,
      "theoretical_loss": 3.7345376354605166,
      "tokens_seen": 788463616
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038438314944834505,
      "loss": 2.8075,
      "theoretical_loss": 3.7345067851853897,
      "tokens_seen": 788529152
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038437311935807423,
      "loss": 3.034,
      "theoretical_loss": 3.734475938192033,
      "tokens_seen": 788594688
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038436308926780347,
      "loss": 2.952,
      "theoretical_loss": 3.734445094479824,
      "tokens_seen": 788660224
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003843530591775326,
      "loss": 2.8488,
      "theoretical_loss": 3.734414254048141,
      "tokens_seen": 788725760
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038434302908726183,
      "loss": 2.9226,
      "theoretical_loss": 3.734383416896362,
      "tokens_seen": 788791296
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038433299899699096,
      "loss": 2.6966,
      "theoretical_loss": 3.7343525830238673,
      "tokens_seen": 788856832
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003843229689067202,
      "loss": 2.8599,
      "theoretical_loss": 3.7343217524300347,
      "tokens_seen": 788922368
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003843129388164494,
      "loss": 2.8344,
      "theoretical_loss": 3.7342909251142435,
      "tokens_seen": 788987904
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038430290872617856,
      "loss": 2.78,
      "theoretical_loss": 3.734260101075873,
      "tokens_seen": 789053440
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038429287863590774,
      "loss": 2.7741,
      "theoretical_loss": 3.7342292803143025,
      "tokens_seen": 789118976
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003842828485456369,
      "loss": 3.0889,
      "theoretical_loss": 3.7341984628289118,
      "tokens_seen": 789184512
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003842728184553661,
      "loss": 3.0021,
      "theoretical_loss": 3.73416764861908,
      "tokens_seen": 789250048
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038426278836509533,
      "loss": 2.9332,
      "theoretical_loss": 3.7341368376841872,
      "tokens_seen": 789315584
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038425275827482446,
      "loss": 3.0278,
      "theoretical_loss": 3.7341060300236144,
      "tokens_seen": 789381120
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003842427281845537,
      "loss": 2.753,
      "theoretical_loss": 3.73407522563674,
      "tokens_seen": 789446656
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003842326980942829,
      "loss": 2.9012,
      "theoretical_loss": 3.7340444245229465,
      "tokens_seen": 789512192
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038422266800401206,
      "loss": 2.8503,
      "theoretical_loss": 3.7340136266816133,
      "tokens_seen": 789577728
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038421263791374124,
      "loss": 2.9075,
      "theoretical_loss": 3.733982832112121,
      "tokens_seen": 789643264
    },
    {
      "epoch": 0.24,
      "objective/train/advantage_avg": 0.49534574151039124,
      "objective/train/docs_used": 455438,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.167574405670166,
      "objective/train/original_loss": 3.167574167251587,
      "objective/train/theoretical_loss": 3.7339520408138513,
      "objective/train/tokens_used": 810168800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24542570114135742,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507822036743164,
      "objective/train/weighted_lm_loss": 3.32832670211792,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0338950157165527,
      "theoretical_loss": 3.7339520408138513,
      "tokens_seen": 789708800
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003842026078234704,
      "loss": 3.0658,
      "theoretical_loss": 3.7339520408138513,
      "tokens_seen": 789708800
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003841925777331996,
      "loss": 3.0075,
      "theoretical_loss": 3.7339212527861845,
      "tokens_seen": 789774336
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038418254764292884,
      "loss": 2.814,
      "theoretical_loss": 3.733890468028502,
      "tokens_seen": 789839872
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038417251755265796,
      "loss": 2.9607,
      "theoretical_loss": 3.7338596865401854,
      "tokens_seen": 789905408
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003841624874623872,
      "loss": 2.7752,
      "theoretical_loss": 3.733828908320616,
      "tokens_seen": 789970944
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003841524573721163,
      "loss": 3.028,
      "theoretical_loss": 3.7337981333691763,
      "tokens_seen": 790036480
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038414242728184556,
      "loss": 2.856,
      "theoretical_loss": 3.733767361685248,
      "tokens_seen": 790102016
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038413239719157474,
      "loss": 3.0039,
      "theoretical_loss": 3.7337365932682123,
      "tokens_seen": 790167552
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003841223671013039,
      "loss": 2.7926,
      "theoretical_loss": 3.7337058281174524,
      "tokens_seen": 790233088
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003841123370110331,
      "loss": 3.0705,
      "theoretical_loss": 3.73367506623235,
      "tokens_seen": 790298624
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003841023069207623,
      "loss": 3.0442,
      "theoretical_loss": 3.7336443076122885,
      "tokens_seen": 790364160
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038409227683049147,
      "loss": 3.0974,
      "theoretical_loss": 3.7336135522566503,
      "tokens_seen": 790429696
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003840822467402207,
      "loss": 2.9761,
      "theoretical_loss": 3.733582800164818,
      "tokens_seen": 790495232
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038407221664994983,
      "loss": 2.9955,
      "theoretical_loss": 3.7335520513361753,
      "tokens_seen": 790560768
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038406218655967907,
      "loss": 3.0151,
      "theoretical_loss": 3.733521305770105,
      "tokens_seen": 790626304
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038405215646940825,
      "loss": 2.9804,
      "theoretical_loss": 3.73349056346599,
      "tokens_seen": 790691840
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038404212637913743,
      "loss": 2.9501,
      "theoretical_loss": 3.733459824423216,
      "tokens_seen": 790757376
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003840320962888666,
      "loss": 2.8309,
      "theoretical_loss": 3.7334290886411643,
      "tokens_seen": 790822912
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003840220661985958,
      "loss": 2.9269,
      "theoretical_loss": 3.7333983561192206,
      "tokens_seen": 790888448
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038401203610832497,
      "loss": 2.8589,
      "theoretical_loss": 3.7333676268567677,
      "tokens_seen": 790953984
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003840020060180542,
      "loss": 2.7199,
      "theoretical_loss": 3.7333369008531907,
      "tokens_seen": 791019520
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038399197592778333,
      "loss": 3.0518,
      "theoretical_loss": 3.733306178107873,
      "tokens_seen": 791085056
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038398194583751257,
      "loss": 2.8991,
      "theoretical_loss": 3.7332754586202013,
      "tokens_seen": 791150592
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003839719157472417,
      "loss": 2.767,
      "theoretical_loss": 3.7332447423895583,
      "tokens_seen": 791216128
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038396188565697093,
      "loss": 2.7963,
      "theoretical_loss": 3.73321402941533,
      "tokens_seen": 791281664
    },
    {
      "epoch": 0.24,
      "objective/train/advantage_avg": 0.49328505992889404,
      "objective/train/docs_used": 456184,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.9396188259124756,
      "objective/train/original_loss": 2.939619302749634,
      "objective/train/theoretical_loss": 3.733183319696901,
      "objective/train/tokens_used": 811807200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24351440370082855,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.05056631565094,
      "objective/train/weighted_lm_loss": 3.0882279872894287,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0253151655197144,
      "theoretical_loss": 3.733183319696901,
      "tokens_seen": 791347200
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003839518555667001,
      "loss": 3.0185,
      "theoretical_loss": 3.733183319696901,
      "tokens_seen": 791347200
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003839418254764293,
      "loss": 2.8346,
      "theoretical_loss": 3.7331526132336568,
      "tokens_seen": 791412736
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003839317953861585,
      "loss": 3.075,
      "theoretical_loss": 3.733121910024983,
      "tokens_seen": 791478272
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038392176529588765,
      "loss": 2.879,
      "theoretical_loss": 3.7330912100702647,
      "tokens_seen": 791543808
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038391173520561684,
      "loss": 2.8582,
      "theoretical_loss": 3.7330605133688883,
      "tokens_seen": 791609344
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038390170511534607,
      "loss": 3.0806,
      "theoretical_loss": 3.7330298199202394,
      "tokens_seen": 791674880
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003838916750250752,
      "loss": 2.8873,
      "theoretical_loss": 3.732999129723704,
      "tokens_seen": 791740416
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038388164493480443,
      "loss": 2.9605,
      "theoretical_loss": 3.7329684427786685,
      "tokens_seen": 791805952
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003838716148445336,
      "loss": 3.0263,
      "theoretical_loss": 3.73293775908452,
      "tokens_seen": 791871488
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003838615847542628,
      "loss": 3.0318,
      "theoretical_loss": 3.732907078640644,
      "tokens_seen": 791937024
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.000383851554663992,
      "loss": 3.042,
      "theoretical_loss": 3.7328764014464273,
      "tokens_seen": 792002560
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038384152457372116,
      "loss": 2.6533,
      "theoretical_loss": 3.732845727501258,
      "tokens_seen": 792068096
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038383149448345034,
      "loss": 2.7334,
      "theoretical_loss": 3.7328150568045224,
      "tokens_seen": 792133632
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003838214643931796,
      "loss": 2.65,
      "theoretical_loss": 3.732784389355608,
      "tokens_seen": 792199168
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003838114343029087,
      "loss": 2.7341,
      "theoretical_loss": 3.7327537251539016,
      "tokens_seen": 792264704
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038380140421263794,
      "loss": 2.9257,
      "theoretical_loss": 3.732723064198791,
      "tokens_seen": 792330240
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038379137412236706,
      "loss": 2.9082,
      "theoretical_loss": 3.732692406489665,
      "tokens_seen": 792395776
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003837813440320963,
      "loss": 2.8492,
      "theoretical_loss": 3.732661752025911,
      "tokens_seen": 792461312
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003837713139418255,
      "loss": 2.8821,
      "theoretical_loss": 3.732631100806916,
      "tokens_seen": 792526848
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038376128385155466,
      "loss": 2.9283,
      "theoretical_loss": 3.73260045283207,
      "tokens_seen": 792592384
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038375125376128384,
      "loss": 2.862,
      "theoretical_loss": 3.7325698081007603,
      "tokens_seen": 792657920
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003837412236710131,
      "loss": 3.0063,
      "theoretical_loss": 3.7325391666123764,
      "tokens_seen": 792723456
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003837311935807422,
      "loss": 3.2078,
      "theoretical_loss": 3.732508528366306,
      "tokens_seen": 792788992
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038372116349047144,
      "loss": 2.847,
      "theoretical_loss": 3.7324778933619385,
      "tokens_seen": 792854528
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038371113340020057,
      "loss": 2.6584,
      "theoretical_loss": 3.7324472615986632,
      "tokens_seen": 792920064
    },
    {
      "epoch": 0.24,
      "objective/train/advantage_avg": 0.4951176941394806,
      "objective/train/docs_used": 457386,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.2095656394958496,
      "objective/train/original_loss": 2.2095654010772705,
      "objective/train/theoretical_loss": 3.732416633075869,
      "objective/train/tokens_used": 813445600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24517838656902313,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507581233978271,
      "objective/train/weighted_lm_loss": 2.321723699569702,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0437887907028198,
      "theoretical_loss": 3.732416633075869,
      "tokens_seen": 792985600
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003837011033099298,
      "loss": 2.8988,
      "theoretical_loss": 3.732416633075869,
      "tokens_seen": 792985600
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.000383691073219659,
      "loss": 2.9121,
      "theoretical_loss": 3.732386007792946,
      "tokens_seen": 793051136
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038368104312938816,
      "loss": 3.0111,
      "theoretical_loss": 3.7323553857492824,
      "tokens_seen": 793116672
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038367101303911735,
      "loss": 2.9323,
      "theoretical_loss": 3.732324766944269,
      "tokens_seen": 793182208
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003836609829488465,
      "loss": 2.8522,
      "theoretical_loss": 3.732294151377296,
      "tokens_seen": 793247744
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003836509528585757,
      "loss": 2.8226,
      "theoretical_loss": 3.7322635390477537,
      "tokens_seen": 793313280
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038364092276830494,
      "loss": 2.9978,
      "theoretical_loss": 3.7322329299550305,
      "tokens_seen": 793378816
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003836308926780341,
      "loss": 3.155,
      "theoretical_loss": 3.7322023240985187,
      "tokens_seen": 793444352
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003836208625877633,
      "loss": 3.1049,
      "theoretical_loss": 3.7321717214776085,
      "tokens_seen": 793509888
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003836108324974925,
      "loss": 2.7514,
      "theoretical_loss": 3.73214112209169,
      "tokens_seen": 793575424
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038360080240722167,
      "loss": 2.9729,
      "theoretical_loss": 3.7321105259401546,
      "tokens_seen": 793640960
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003835907723169509,
      "loss": 2.9004,
      "theoretical_loss": 3.7320799330223933,
      "tokens_seen": 793706496
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038358074222668003,
      "loss": 2.7922,
      "theoretical_loss": 3.7320493433377973,
      "tokens_seen": 793772032
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038357071213640927,
      "loss": 2.8281,
      "theoretical_loss": 3.7320187568857586,
      "tokens_seen": 793837568
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038356068204613845,
      "loss": 3.013,
      "theoretical_loss": 3.731988173665668,
      "tokens_seen": 793903104
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038355065195586763,
      "loss": 2.939,
      "theoretical_loss": 3.731957593676917,
      "tokens_seen": 793968640
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003835406218655968,
      "loss": 2.9195,
      "theoretical_loss": 3.731927016918898,
      "tokens_seen": 794034176
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.000383530591775326,
      "loss": 2.8697,
      "theoretical_loss": 3.7318964433910033,
      "tokens_seen": 794099712
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038352056168505517,
      "loss": 2.9543,
      "theoretical_loss": 3.731865873092625,
      "tokens_seen": 794165248
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003835105315947844,
      "loss": 3.1318,
      "theoretical_loss": 3.7318353060231555,
      "tokens_seen": 794230784
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038350050150451353,
      "loss": 3.0071,
      "theoretical_loss": 3.7318047421819864,
      "tokens_seen": 794296320
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038349047141424277,
      "loss": 2.8556,
      "theoretical_loss": 3.731774181568512,
      "tokens_seen": 794361856
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003834804413239719,
      "loss": 2.8677,
      "theoretical_loss": 3.7317436241821245,
      "tokens_seen": 794427392
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038347041123370113,
      "loss": 3.0381,
      "theoretical_loss": 3.7317130700222174,
      "tokens_seen": 794492928
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003834603811434303,
      "loss": 2.6884,
      "theoretical_loss": 3.7316825190881824,
      "tokens_seen": 794558464
    },
    {
      "epoch": 0.24,
      "objective/train/advantage_avg": 0.49370622634887695,
      "objective/train/docs_used": 458050,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.436598539352417,
      "objective/train/original_loss": 2.436598777770996,
      "objective/train/theoretical_loss": 3.7316519713794145,
      "objective/train/tokens_used": 815084000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24382774531841278,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506101846694946,
      "objective/train/weighted_lm_loss": 2.5599100589752197,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0363596677780151,
      "theoretical_loss": 3.7316519713794145,
      "tokens_seen": 794624000
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003834503510531595,
      "loss": 2.9574,
      "theoretical_loss": 3.7316519713794145,
      "tokens_seen": 794624000
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003834403209628887,
      "loss": 3.0203,
      "theoretical_loss": 3.7316214268953063,
      "tokens_seen": 794689536
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038343029087261786,
      "loss": 3.0901,
      "theoretical_loss": 3.7315908856352524,
      "tokens_seen": 794755072
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038342026078234704,
      "loss": 2.7878,
      "theoretical_loss": 3.731560347598646,
      "tokens_seen": 794820608
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038341023069207627,
      "loss": 2.8065,
      "theoretical_loss": 3.7315298127848813,
      "tokens_seen": 794886144
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003834002006018054,
      "loss": 3.009,
      "theoretical_loss": 3.7314992811933525,
      "tokens_seen": 794951680
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038339017051153463,
      "loss": 2.7642,
      "theoretical_loss": 3.731468752823454,
      "tokens_seen": 795017216
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003833801404212638,
      "loss": 3.0111,
      "theoretical_loss": 3.73143822767458,
      "tokens_seen": 795082752
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.000383370110330993,
      "loss": 2.624,
      "theoretical_loss": 3.7314077057461255,
      "tokens_seen": 795148288
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003833600802407222,
      "loss": 2.831,
      "theoretical_loss": 3.731377187037486,
      "tokens_seen": 795213824
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038335005015045136,
      "loss": 2.9711,
      "theoretical_loss": 3.7313466715480557,
      "tokens_seen": 795279360
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038334002006018054,
      "loss": 2.7261,
      "theoretical_loss": 3.7313161592772293,
      "tokens_seen": 795344896
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003833299899699098,
      "loss": 3.1816,
      "theoretical_loss": 3.731285650224404,
      "tokens_seen": 795410432
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003833199598796389,
      "loss": 3.0197,
      "theoretical_loss": 3.7312551443889728,
      "tokens_seen": 795475968
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038330992978936814,
      "loss": 2.8368,
      "theoretical_loss": 3.731224641770334,
      "tokens_seen": 795541504
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038329989969909726,
      "loss": 3.0391,
      "theoretical_loss": 3.7311941423678814,
      "tokens_seen": 795607040
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003832898696088265,
      "loss": 2.8306,
      "theoretical_loss": 3.731163646181012,
      "tokens_seen": 795672576
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003832798395185557,
      "loss": 3.0419,
      "theoretical_loss": 3.7311331532091216,
      "tokens_seen": 795738112
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038326980942828486,
      "loss": 3.0662,
      "theoretical_loss": 3.7311026634516065,
      "tokens_seen": 795803648
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038325977933801404,
      "loss": 2.8662,
      "theoretical_loss": 3.7310721769078636,
      "tokens_seen": 795869184
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003832497492477433,
      "loss": 2.924,
      "theoretical_loss": 3.731041693577289,
      "tokens_seen": 795934720
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003832397191574724,
      "loss": 2.7543,
      "theoretical_loss": 3.7310112134592806,
      "tokens_seen": 796000256
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038322968906720164,
      "loss": 2.8363,
      "theoretical_loss": 3.730980736553234,
      "tokens_seen": 796065792
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038321965897693077,
      "loss": 2.8153,
      "theoretical_loss": 3.7309502628585474,
      "tokens_seen": 796131328
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038320962888666,
      "loss": 2.9253,
      "theoretical_loss": 3.7309197923746176,
      "tokens_seen": 796196864
    },
    {
      "epoch": 0.24,
      "objective/train/advantage_avg": 0.49609705805778503,
      "objective/train/docs_used": 459137,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.611326217651367,
      "objective/train/original_loss": 2.611326217651367,
      "objective/train/theoretical_loss": 3.7308893251008413,
      "objective/train/tokens_used": 816722400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24617154896259308,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0508612394332886,
      "objective/train/weighted_lm_loss": 2.7440502643585205,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 3.7308893251008413,
      "tokens_seen": 796262400
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003831995987963892,
      "loss": 2.9708,
      "theoretical_loss": 3.7308893251008413,
      "tokens_seen": 796262400
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038318956870611836,
      "loss": 2.9714,
      "theoretical_loss": 3.7308588610366176,
      "tokens_seen": 796327936
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038317953861584755,
      "loss": 3.1493,
      "theoretical_loss": 3.730828400181344,
      "tokens_seen": 796393472
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003831695085255767,
      "loss": 2.9627,
      "theoretical_loss": 3.7307979425344175,
      "tokens_seen": 796459008
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003831594784353059,
      "loss": 2.8159,
      "theoretical_loss": 3.7307674880952373,
      "tokens_seen": 796524544
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038314944834503514,
      "loss": 2.9034,
      "theoretical_loss": 3.730737036863201,
      "tokens_seen": 796590080
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038313941825476427,
      "loss": 2.9616,
      "theoretical_loss": 3.730706588837707,
      "tokens_seen": 796655616
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003831293881644935,
      "loss": 2.6063,
      "theoretical_loss": 3.730676144018154,
      "tokens_seen": 796721152
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038311935807422263,
      "loss": 2.9991,
      "theoretical_loss": 3.7306457024039417,
      "tokens_seen": 796786688
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038310932798395187,
      "loss": 2.7368,
      "theoretical_loss": 3.730615263994468,
      "tokens_seen": 796852224
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038309929789368105,
      "loss": 2.8245,
      "theoretical_loss": 3.730584828789132,
      "tokens_seen": 796917760
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038308926780341023,
      "loss": 2.4944,
      "theoretical_loss": 3.7305543967873334,
      "tokens_seen": 796983296
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003830792377131394,
      "loss": 2.9001,
      "theoretical_loss": 3.7305239679884714,
      "tokens_seen": 797048832
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038306920762286865,
      "loss": 2.5857,
      "theoretical_loss": 3.7304935423919456,
      "tokens_seen": 797114368
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003830591775325978,
      "loss": 2.9396,
      "theoretical_loss": 3.730463119997156,
      "tokens_seen": 797179904
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.000383049147442327,
      "loss": 2.8744,
      "theoretical_loss": 3.730432700803502,
      "tokens_seen": 797245440
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038303911735205614,
      "loss": 2.8633,
      "theoretical_loss": 3.730402284810384,
      "tokens_seen": 797310976
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038302908726178537,
      "loss": 2.9897,
      "theoretical_loss": 3.7303718720172023,
      "tokens_seen": 797376512
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038301905717151455,
      "loss": 2.8426,
      "theoretical_loss": 3.730341462423357,
      "tokens_seen": 797442048
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038300902708124373,
      "loss": 2.9777,
      "theoretical_loss": 3.7303110560282486,
      "tokens_seen": 797507584
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003829989969909729,
      "loss": 2.9796,
      "theoretical_loss": 3.7302806528312784,
      "tokens_seen": 797573120
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003829889669007021,
      "loss": 3.1144,
      "theoretical_loss": 3.730250252831847,
      "tokens_seen": 797638656
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003829789368104313,
      "loss": 2.8902,
      "theoretical_loss": 3.730219856029355,
      "tokens_seen": 797704192
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003829689067201605,
      "loss": 2.8786,
      "theoretical_loss": 3.730189462423204,
      "tokens_seen": 797769728
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038295887662988964,
      "loss": 2.5994,
      "theoretical_loss": 3.7301590720127953,
      "tokens_seen": 797835264
    },
    {
      "epoch": 0.24,
      "objective/train/advantage_avg": 0.49416235089302063,
      "objective/train/docs_used": 459607,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.41686749458313,
      "objective/train/original_loss": 3.416867256164551,
      "objective/train/theoretical_loss": 3.7301286847975295,
      "objective/train/tokens_used": 818360800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2442934215068817,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506579875946045,
      "objective/train/weighted_lm_loss": 3.590080976486206,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0408108234405518,
      "theoretical_loss": 3.7301286847975295,
      "tokens_seen": 797900800
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003829488465396189,
      "loss": 3.1907,
      "theoretical_loss": 3.7301286847975295,
      "tokens_seen": 797900800
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.000382938816449348,
      "loss": 3.244,
      "theoretical_loss": 3.7300983007768105,
      "tokens_seen": 797966336
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038292878635907724,
      "loss": 2.9552,
      "theoretical_loss": 3.730067919950038,
      "tokens_seen": 798031872
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003829187562688064,
      "loss": 3.0158,
      "theoretical_loss": 3.730037542316615,
      "tokens_seen": 798097408
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003829087261785356,
      "loss": 2.9366,
      "theoretical_loss": 3.730007167875944,
      "tokens_seen": 798162944
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003828986960882648,
      "loss": 3.0298,
      "theoretical_loss": 3.729976796627426,
      "tokens_seen": 798228480
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.000382888665997994,
      "loss": 2.7432,
      "theoretical_loss": 3.7299464285704644,
      "tokens_seen": 798294016
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003828786359077232,
      "loss": 3.0753,
      "theoretical_loss": 3.7299160637044624,
      "tokens_seen": 798359552
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003828686058174524,
      "loss": 2.8846,
      "theoretical_loss": 3.7298857020288216,
      "tokens_seen": 798425088
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038285857572718156,
      "loss": 2.8722,
      "theoretical_loss": 3.7298553435429453,
      "tokens_seen": 798490624
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038284854563691074,
      "loss": 2.8692,
      "theoretical_loss": 3.7298249882462375,
      "tokens_seen": 798556160
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038283851554664,
      "loss": 2.6399,
      "theoretical_loss": 3.7297946361381005,
      "tokens_seen": 798621696
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003828284854563691,
      "loss": 2.8332,
      "theoretical_loss": 3.7297642872179377,
      "tokens_seen": 798687232
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038281845536609834,
      "loss": 3.0856,
      "theoretical_loss": 3.7297339414851534,
      "tokens_seen": 798752768
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038280842527582746,
      "loss": 2.8049,
      "theoretical_loss": 3.7297035989391514,
      "tokens_seen": 798818304
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003827983951855567,
      "loss": 2.8928,
      "theoretical_loss": 3.729673259579335,
      "tokens_seen": 798883840
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003827883650952859,
      "loss": 2.964,
      "theoretical_loss": 3.729642923405108,
      "tokens_seen": 798949376
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038277833500501506,
      "loss": 2.9604,
      "theoretical_loss": 3.729612590415876,
      "tokens_seen": 799014912
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038276830491474424,
      "loss": 3.0606,
      "theoretical_loss": 3.7295822606110423,
      "tokens_seen": 799080448
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003827582748244735,
      "loss": 2.8797,
      "theoretical_loss": 3.7295519339900123,
      "tokens_seen": 799145984
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003827482447342026,
      "loss": 2.6028,
      "theoretical_loss": 3.7295216105521893,
      "tokens_seen": 799211520
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038273821464393184,
      "loss": 3.0062,
      "theoretical_loss": 3.72949129029698,
      "tokens_seen": 799277056
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038272818455366097,
      "loss": 2.8581,
      "theoretical_loss": 3.7294609732237882,
      "tokens_seen": 799342592
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003827181544633902,
      "loss": 2.6782,
      "theoretical_loss": 3.72943065933202,
      "tokens_seen": 799408128
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003827081243731194,
      "loss": 2.8036,
      "theoretical_loss": 3.7294003486210796,
      "tokens_seen": 799473664
    },
    {
      "epoch": 0.24,
      "objective/train/advantage_avg": 0.49555808305740356,
      "objective/train/docs_used": 460885,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.8115341663360596,
      "objective/train/original_loss": 2.8115339279174805,
      "objective/train/theoretical_loss": 3.729370041090373,
      "objective/train/tokens_used": 819999200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2456231564283371,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0508043766021729,
      "objective/train/weighted_lm_loss": 2.954429864883423,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0442860126495361,
      "theoretical_loss": 3.729370041090373,
      "tokens_seen": 799539200
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038269809428284856,
      "loss": 3.1341,
      "theoretical_loss": 3.729370041090373,
      "tokens_seen": 799539200
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038268806419257775,
      "loss": 2.9704,
      "theoretical_loss": 3.7293397367393073,
      "tokens_seen": 799604736
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003826780341023069,
      "loss": 2.9796,
      "theoretical_loss": 3.729309435567286,
      "tokens_seen": 799670272
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003826680040120361,
      "loss": 3.0708,
      "theoretical_loss": 3.729279137573717,
      "tokens_seen": 799735808
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038265797392176534,
      "loss": 2.8421,
      "theoretical_loss": 3.7292488427580053,
      "tokens_seen": 799801344
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038264794383149447,
      "loss": 2.6089,
      "theoretical_loss": 3.729218551119558,
      "tokens_seen": 799866880
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003826379137412237,
      "loss": 2.7237,
      "theoretical_loss": 3.729188262657781,
      "tokens_seen": 799932416
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038262788365095283,
      "loss": 2.9108,
      "theoretical_loss": 3.7291579773720813,
      "tokens_seen": 799997952
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038261785356068207,
      "loss": 2.8055,
      "theoretical_loss": 3.7291276952618655,
      "tokens_seen": 800063488
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038260782347041125,
      "loss": 3.0441,
      "theoretical_loss": 3.7290974163265407,
      "tokens_seen": 800129024
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038259779338014043,
      "loss": 2.8169,
      "theoretical_loss": 3.7290671405655145,
      "tokens_seen": 800194560
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003825877632898696,
      "loss": 2.8881,
      "theoretical_loss": 3.729036867978193,
      "tokens_seen": 800260096
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038257773319959885,
      "loss": 2.7553,
      "theoretical_loss": 3.7290065985639846,
      "tokens_seen": 800325632
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.000382567703109328,
      "loss": 2.9841,
      "theoretical_loss": 3.728976332322297,
      "tokens_seen": 800391168
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003825576730190572,
      "loss": 3.1071,
      "theoretical_loss": 3.728946069252537,
      "tokens_seen": 800456704
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038254764292878634,
      "loss": 2.8132,
      "theoretical_loss": 3.7289158093541133,
      "tokens_seen": 800522240
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038253761283851557,
      "loss": 2.8728,
      "theoretical_loss": 3.7288855526264344,
      "tokens_seen": 800587776
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038252758274824475,
      "loss": 3.1682,
      "theoretical_loss": 3.7288552990689077,
      "tokens_seen": 800653312
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038251755265797393,
      "loss": 2.977,
      "theoretical_loss": 3.7288250486809416,
      "tokens_seen": 800718848
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003825075225677031,
      "loss": 2.7988,
      "theoretical_loss": 3.7287948014619454,
      "tokens_seen": 800784384
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003824974924774323,
      "loss": 2.9487,
      "theoretical_loss": 3.728764557411327,
      "tokens_seen": 800849920
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003824874623871615,
      "loss": 2.8475,
      "theoretical_loss": 3.728734316528495,
      "tokens_seen": 800915456
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003824774322968907,
      "loss": 2.8986,
      "theoretical_loss": 3.72870407881286,
      "tokens_seen": 800980992
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038246740220661984,
      "loss": 2.8494,
      "theoretical_loss": 3.72867384426383,
      "tokens_seen": 801046528
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003824573721163491,
      "loss": 2.981,
      "theoretical_loss": 3.7286436128808145,
      "tokens_seen": 801112064
    },
    {
      "epoch": 0.24,
      "objective/train/advantage_avg": 0.49346858263015747,
      "objective/train/docs_used": 461399,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.1561880111694336,
      "objective/train/original_loss": 3.1561882495880127,
      "objective/train/theoretical_loss": 3.7286133846632232,
      "objective/train/tokens_used": 821637600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24361135065555573,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505852699279785,
      "objective/train/weighted_lm_loss": 3.315882921218872,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0396549701690674,
      "theoretical_loss": 3.7286133846632232,
      "tokens_seen": 801177600
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003824473420260782,
      "loss": 2.774,
      "theoretical_loss": 3.7286133846632232,
      "tokens_seen": 801177600
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038243731193580744,
      "loss": 2.8467,
      "theoretical_loss": 3.7285831596104657,
      "tokens_seen": 801243136
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003824272818455366,
      "loss": 3.1331,
      "theoretical_loss": 3.7285529377219517,
      "tokens_seen": 801308672
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003824172517552658,
      "loss": 2.9522,
      "theoretical_loss": 3.7285227189970915,
      "tokens_seen": 801374208
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.000382407221664995,
      "loss": 3.0956,
      "theoretical_loss": 3.728492503435295,
      "tokens_seen": 801439744
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003823971915747242,
      "loss": 2.9414,
      "theoretical_loss": 3.7284622910359726,
      "tokens_seen": 801505280
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038238716148445334,
      "loss": 2.7881,
      "theoretical_loss": 3.728432081798535,
      "tokens_seen": 801570816
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003823771313941826,
      "loss": 2.9673,
      "theoretical_loss": 3.7284018757223913,
      "tokens_seen": 801636352
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003823671013039117,
      "loss": 2.9062,
      "theoretical_loss": 3.7283716728069547,
      "tokens_seen": 801701888
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038235707121364094,
      "loss": 2.9199,
      "theoretical_loss": 3.7283414730516347,
      "tokens_seen": 801767424
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003823470411233701,
      "loss": 2.8141,
      "theoretical_loss": 3.7283112764558424,
      "tokens_seen": 801832960
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003823370110330993,
      "loss": 2.8137,
      "theoretical_loss": 3.7282810830189894,
      "tokens_seen": 801898496
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003823269809428285,
      "loss": 2.8408,
      "theoretical_loss": 3.728250892740487,
      "tokens_seen": 801964032
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038231695085255766,
      "loss": 2.9341,
      "theoretical_loss": 3.728220705619747,
      "tokens_seen": 802029568
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038230692076228685,
      "loss": 2.9754,
      "theoretical_loss": 3.72819052165618,
      "tokens_seen": 802095104
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003822968906720161,
      "loss": 2.7399,
      "theoretical_loss": 3.7281603408491995,
      "tokens_seen": 802160640
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003822868605817452,
      "loss": 2.942,
      "theoretical_loss": 3.728130163198217,
      "tokens_seen": 802226176
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038227683049147444,
      "loss": 3.0363,
      "theoretical_loss": 3.728099988702644,
      "tokens_seen": 802291712
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038226680040120357,
      "loss": 2.831,
      "theoretical_loss": 3.728069817361894,
      "tokens_seen": 802357248
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003822567703109328,
      "loss": 3.0347,
      "theoretical_loss": 3.7280396491753782,
      "tokens_seen": 802422784
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.000382246740220662,
      "loss": 3.0656,
      "theoretical_loss": 3.72800948414251,
      "tokens_seen": 802488320
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038223671013039117,
      "loss": 2.8614,
      "theoretical_loss": 3.7279793222627022,
      "tokens_seen": 802553856
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038222668004012035,
      "loss": 3.0481,
      "theoretical_loss": 3.727949163535368,
      "tokens_seen": 802619392
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003822166499498496,
      "loss": 2.8595,
      "theoretical_loss": 3.7279190079599203,
      "tokens_seen": 802684928
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003822066198595787,
      "loss": 3.0146,
      "theoretical_loss": 3.727888855535772,
      "tokens_seen": 802750464
    },
    {
      "epoch": 0.24,
      "objective/train/advantage_avg": 0.49447205662727356,
      "objective/train/docs_used": 462084,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.0290043354034424,
      "objective/train/original_loss": 3.0290043354034424,
      "objective/train/theoretical_loss": 3.727858706262338,
      "objective/train/tokens_used": 823276000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24457135796546936,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506902933120728,
      "objective/train/weighted_lm_loss": 3.182662010192871,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0338950157165527,
      "theoretical_loss": 3.727858706262338,
      "tokens_seen": 802816000
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038219658976930795,
      "loss": 2.9095,
      "theoretical_loss": 3.727858706262338,
      "tokens_seen": 802816000
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003821865596790371,
      "loss": 2.8912,
      "theoretical_loss": 3.7278285601390295,
      "tokens_seen": 802881536
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003821765295887663,
      "loss": 3.0476,
      "theoretical_loss": 3.7277984171652623,
      "tokens_seen": 802947072
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003821664994984955,
      "loss": 3.1603,
      "theoretical_loss": 3.7277682773404495,
      "tokens_seen": 803012608
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038215646940822467,
      "loss": 3.0839,
      "theoretical_loss": 3.7277381406640053,
      "tokens_seen": 803078144
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038214643931795385,
      "loss": 3.007,
      "theoretical_loss": 3.7277080071353437,
      "tokens_seen": 803143680
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038213640922768303,
      "loss": 2.9459,
      "theoretical_loss": 3.72767787675388,
      "tokens_seen": 803209216
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038212637913741227,
      "loss": 2.8735,
      "theoretical_loss": 3.727647749519028,
      "tokens_seen": 803274752
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038211634904714145,
      "loss": 3.1105,
      "theoretical_loss": 3.7276176254302023,
      "tokens_seen": 803340288
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038210631895687063,
      "loss": 3.0176,
      "theoretical_loss": 3.727587504486818,
      "tokens_seen": 803405824
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003820962888665998,
      "loss": 3.1036,
      "theoretical_loss": 3.7275573866882903,
      "tokens_seen": 803471360
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038208625877632905,
      "loss": 2.9169,
      "theoretical_loss": 3.727527272034034,
      "tokens_seen": 803536896
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003820762286860582,
      "loss": 2.9471,
      "theoretical_loss": 3.7274971605234644,
      "tokens_seen": 803602432
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003820661985957874,
      "loss": 3.0383,
      "theoretical_loss": 3.7274670521559976,
      "tokens_seen": 803667968
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038205616850551654,
      "loss": 2.9254,
      "theoretical_loss": 3.727436946931049,
      "tokens_seen": 803733504
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038204613841524577,
      "loss": 2.7996,
      "theoretical_loss": 3.727406844848034,
      "tokens_seen": 803799040
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038203610832497495,
      "loss": 3.011,
      "theoretical_loss": 3.7273767459063687,
      "tokens_seen": 803864576
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038202607823470413,
      "loss": 3.0531,
      "theoretical_loss": 3.7273466501054697,
      "tokens_seen": 803930112
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003820160481444333,
      "loss": 2.6671,
      "theoretical_loss": 3.7273165574447527,
      "tokens_seen": 803995648
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003820060180541625,
      "loss": 2.8581,
      "theoretical_loss": 3.7272864679236344,
      "tokens_seen": 804061184
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003819959879638917,
      "loss": 2.9126,
      "theoretical_loss": 3.7272563815415314,
      "tokens_seen": 804126720
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003819859578736209,
      "loss": 3.0057,
      "theoretical_loss": 3.727226298297861,
      "tokens_seen": 804192256
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038197592778335004,
      "loss": 2.7309,
      "theoretical_loss": 3.727196218192039,
      "tokens_seen": 804257792
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003819658976930793,
      "loss": 2.9454,
      "theoretical_loss": 3.727166141223482,
      "tokens_seen": 804323328
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003819558676028084,
      "loss": 3.0398,
      "theoretical_loss": 3.7271360673916094,
      "tokens_seen": 804388864
    },
    {
      "epoch": 0.24,
      "objective/train/advantage_avg": 0.4951366186141968,
      "objective/train/docs_used": 463313,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.9573004245758057,
      "objective/train/original_loss": 2.9573004245758057,
      "objective/train/theoretical_loss": 3.727105996695837,
      "objective/train/tokens_used": 824914400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24521282315254211,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050760269165039,
      "objective/train/weighted_lm_loss": 3.1072521209716797,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0417574644088745,
      "theoretical_loss": 3.727105996695837,
      "tokens_seen": 804454400
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038194583751253764,
      "loss": 3.0532,
      "theoretical_loss": 3.727105996695837,
      "tokens_seen": 804454400
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003819358074222668,
      "loss": 2.9877,
      "theoretical_loss": 3.727075929135583,
      "tokens_seen": 804519936
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.000381925777331996,
      "loss": 2.9834,
      "theoretical_loss": 3.727045864710264,
      "tokens_seen": 804585472
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003819157472417252,
      "loss": 2.8842,
      "theoretical_loss": 3.727015803419299,
      "tokens_seen": 804651008
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003819057171514544,
      "loss": 2.8845,
      "theoretical_loss": 3.7269857452621054,
      "tokens_seen": 804716544
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038189568706118354,
      "loss": 2.9425,
      "theoretical_loss": 3.7269556902381007,
      "tokens_seen": 804782080
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003818856569709128,
      "loss": 3.1322,
      "theoretical_loss": 3.726925638346705,
      "tokens_seen": 804847616
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003818756268806419,
      "loss": 2.9927,
      "theoretical_loss": 3.7268955895873352,
      "tokens_seen": 804913152
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038186559679037114,
      "loss": 3.0226,
      "theoretical_loss": 3.726865543959411,
      "tokens_seen": 804978688
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003818555667001003,
      "loss": 3.1136,
      "theoretical_loss": 3.72683550146235,
      "tokens_seen": 805044224
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003818455366098295,
      "loss": 2.9967,
      "theoretical_loss": 3.726805462095572,
      "tokens_seen": 805109760
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003818355065195587,
      "loss": 2.9615,
      "theoretical_loss": 3.726775425858496,
      "tokens_seen": 805175296
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038182547642928786,
      "loss": 3.0195,
      "theoretical_loss": 3.7267453927505407,
      "tokens_seen": 805240832
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038181544633901705,
      "loss": 2.9412,
      "theoretical_loss": 3.7267153627711256,
      "tokens_seen": 805306368
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003818054162487463,
      "loss": 3.0892,
      "theoretical_loss": 3.726685335919671,
      "tokens_seen": 805371904
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003817953861584754,
      "loss": 3.0126,
      "theoretical_loss": 3.7266553121955956,
      "tokens_seen": 805437440
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038178535606820464,
      "loss": 3.0046,
      "theoretical_loss": 3.72662529159832,
      "tokens_seen": 805502976
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038177532597793377,
      "loss": 3.0907,
      "theoretical_loss": 3.7265952741272637,
      "tokens_seen": 805568512
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.000381765295887663,
      "loss": 2.9259,
      "theoretical_loss": 3.726565259781847,
      "tokens_seen": 805634048
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003817552657973922,
      "loss": 3.0365,
      "theoretical_loss": 3.7265352485614907,
      "tokens_seen": 805699584
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038174523570712137,
      "loss": 2.9474,
      "theoretical_loss": 3.726505240465614,
      "tokens_seen": 805765120
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038173520561685055,
      "loss": 3.0016,
      "theoretical_loss": 3.726475235493639,
      "tokens_seen": 805830656
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003817251755265798,
      "loss": 2.9442,
      "theoretical_loss": 3.726445233644986,
      "tokens_seen": 805896192
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003817151454363089,
      "loss": 2.8486,
      "theoretical_loss": 3.726415234919075,
      "tokens_seen": 805961728
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038170511534603815,
      "loss": 2.9136,
      "theoretical_loss": 3.7263852393153285,
      "tokens_seen": 806027264
    },
    {
      "epoch": 0.24,
      "objective/train/advantage_avg": 0.4926786422729492,
      "objective/train/docs_used": 464591,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.546175956726074,
      "objective/train/original_loss": 2.5461761951446533,
      "objective/train/theoretical_loss": 3.7263552468331667,
      "objective/train/tokens_used": 826552800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24281136691570282,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050502061843872,
      "objective/train/weighted_lm_loss": 2.674635648727417,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0396549701690674,
      "theoretical_loss": 3.7263552468331667,
      "tokens_seen": 806092800
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003816950852557673,
      "loss": 2.8138,
      "theoretical_loss": 3.7263552468331667,
      "tokens_seen": 806092800
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003816850551654965,
      "loss": 2.8088,
      "theoretical_loss": 3.726325257472012,
      "tokens_seen": 806158336
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003816750250752257,
      "loss": 2.8499,
      "theoretical_loss": 3.7262952712312845,
      "tokens_seen": 806223872
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038166499498495487,
      "loss": 2.8571,
      "theoretical_loss": 3.7262652881104072,
      "tokens_seen": 806289408
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038165496489468405,
      "loss": 2.9006,
      "theoretical_loss": 3.7262353081088015,
      "tokens_seen": 806354944
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038164493480441323,
      "loss": 3.0579,
      "theoretical_loss": 3.726205331225889,
      "tokens_seen": 806420480
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003816349047141424,
      "loss": 2.7844,
      "theoretical_loss": 3.726175357461093,
      "tokens_seen": 806486016
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038162487462387165,
      "loss": 3.0122,
      "theoretical_loss": 3.7261453868138346,
      "tokens_seen": 806551552
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003816148445336008,
      "loss": 2.9097,
      "theoretical_loss": 3.726115419283537,
      "tokens_seen": 806617088
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038160481444333,
      "loss": 2.6704,
      "theoretical_loss": 3.7260854548696223,
      "tokens_seen": 806682624
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038159478435305914,
      "loss": 2.8447,
      "theoretical_loss": 3.7260554935715136,
      "tokens_seen": 806748160
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003815847542627884,
      "loss": 3.0943,
      "theoretical_loss": 3.726025535388634,
      "tokens_seen": 806813696
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038157472417251756,
      "loss": 3.0966,
      "theoretical_loss": 3.725995580320406,
      "tokens_seen": 806879232
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038156469408224674,
      "loss": 3.0909,
      "theoretical_loss": 3.7259656283662537,
      "tokens_seen": 806944768
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003815546639919759,
      "loss": 2.8725,
      "theoretical_loss": 3.7259356795256,
      "tokens_seen": 807010304
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038154463390170515,
      "loss": 2.8074,
      "theoretical_loss": 3.7259057337978687,
      "tokens_seen": 807075840
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003815346038114343,
      "loss": 3.0671,
      "theoretical_loss": 3.7258757911824825,
      "tokens_seen": 807141376
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003815245737211635,
      "loss": 3.0795,
      "theoretical_loss": 3.7258458516788666,
      "tokens_seen": 807206912
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038151454363089264,
      "loss": 3.1192,
      "theoretical_loss": 3.7258159152864447,
      "tokens_seen": 807272448
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003815045135406219,
      "loss": 2.8426,
      "theoretical_loss": 3.7257859820046404,
      "tokens_seen": 807337984
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038149448345035106,
      "loss": 2.9024,
      "theoretical_loss": 3.725756051832878,
      "tokens_seen": 807403520
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038148445336008024,
      "loss": 2.6787,
      "theoretical_loss": 3.725726124770583,
      "tokens_seen": 807469056
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003814744232698094,
      "loss": 2.8703,
      "theoretical_loss": 3.725696200817179,
      "tokens_seen": 807534592
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003814643931795386,
      "loss": 2.9405,
      "theoretical_loss": 3.725666279972091,
      "tokens_seen": 807600128
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003814543630892678,
      "loss": 3.0398,
      "theoretical_loss": 3.725636362234744,
      "tokens_seen": 807665664
    },
    {
      "epoch": 0.24,
      "objective/train/advantage_avg": 0.49482491612434387,
      "objective/train/docs_used": 465345,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.8255138397216797,
      "objective/train/original_loss": 2.8255138397216797,
      "objective/train/theoretical_loss": 3.725606447604563,
      "objective/train/tokens_used": 828191200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24489620327949524,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050727367401123,
      "objective/train/weighted_lm_loss": 2.9688191413879395,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0457526445388794,
      "theoretical_loss": 3.725606447604563,
      "tokens_seen": 807731200
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.000381444332998997,
      "loss": 3.0304,
      "theoretical_loss": 3.725606447604563,
      "tokens_seen": 807731200
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038143430290872614,
      "loss": 2.8242,
      "theoretical_loss": 3.725576536080974,
      "tokens_seen": 807796736
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003814242728184554,
      "loss": 2.8251,
      "theoretical_loss": 3.7255466276634013,
      "tokens_seen": 807862272
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038141424272818456,
      "loss": 2.7755,
      "theoretical_loss": 3.725516722351271,
      "tokens_seen": 807927808
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038140421263791374,
      "loss": 2.6377,
      "theoretical_loss": 3.725486820144009,
      "tokens_seen": 807993344
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003813941825476429,
      "loss": 3.0919,
      "theoretical_loss": 3.7254569210410398,
      "tokens_seen": 808058880
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003813841524573721,
      "loss": 3.0299,
      "theoretical_loss": 3.725427025041791,
      "tokens_seen": 808124416
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038137412236710134,
      "loss": 3.0038,
      "theoretical_loss": 3.7253971321456882,
      "tokens_seen": 808189952
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003813640922768305,
      "loss": 2.9687,
      "theoretical_loss": 3.7253672423521578,
      "tokens_seen": 808255488
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003813540621865597,
      "loss": 3.0106,
      "theoretical_loss": 3.725337355660626,
      "tokens_seen": 808321024
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.0003813440320962889,
      "loss": 2.91,
      "theoretical_loss": 3.7253074720705195,
      "tokens_seen": 808386560
    },
    {
      "epoch": 0.24,
      "learning_rate": 0.00038133400200601806,
      "loss": 2.8407,
      "theoretical_loss": 3.7252775915812655,
      "tokens_seen": 808452096
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00038132397191574725,
      "loss": 2.8453,
      "theoretical_loss": 3.72524771419229,
      "tokens_seen": 808517632
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003813139418254765,
      "loss": 2.7703,
      "theoretical_loss": 3.725217839903021,
      "tokens_seen": 808583168
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003813039117352056,
      "loss": 2.9211,
      "theoretical_loss": 3.725187968712885,
      "tokens_seen": 808648704
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00038129388164493484,
      "loss": 3.0396,
      "theoretical_loss": 3.72515810062131,
      "tokens_seen": 808714240
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00038128385155466397,
      "loss": 3.0987,
      "theoretical_loss": 3.7251282356277233,
      "tokens_seen": 808779776
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003812738214643932,
      "loss": 3.1493,
      "theoretical_loss": 3.725098373731552,
      "tokens_seen": 808845312
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003812637913741224,
      "loss": 2.8111,
      "theoretical_loss": 3.7250685149322247,
      "tokens_seen": 808910848
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00038125376128385157,
      "loss": 3.0616,
      "theoretical_loss": 3.7250386592291695,
      "tokens_seen": 808976384
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00038124373119358075,
      "loss": 3.0779,
      "theoretical_loss": 3.725008806621813,
      "tokens_seen": 809041920
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00038123370110331,
      "loss": 3.1385,
      "theoretical_loss": 3.7249789571095855,
      "tokens_seen": 809107456
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003812236710130391,
      "loss": 2.9554,
      "theoretical_loss": 3.7249491106919144,
      "tokens_seen": 809172992
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00038121364092276835,
      "loss": 2.7275,
      "theoretical_loss": 3.7249192673682283,
      "tokens_seen": 809238528
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003812036108324975,
      "loss": 3.0731,
      "theoretical_loss": 3.7248894271379562,
      "tokens_seen": 809304064
    },
    {
      "epoch": 0.25,
      "objective/train/advantage_avg": 0.49464473128318787,
      "objective/train/docs_used": 466409,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.7988381385803223,
      "objective/train/original_loss": 2.798837900161743,
      "objective/train/theoretical_loss": 3.724859590000527,
      "objective/train/tokens_used": 829829600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24475401639938354,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507086515426636,
      "objective/train/weighted_lm_loss": 2.9408507347106934,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0338950157165527,
      "theoretical_loss": 3.724859590000527,
      "tokens_seen": 809369600
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003811935807422267,
      "loss": 3.1726,
      "theoretical_loss": 3.724859590000527,
      "tokens_seen": 809369600
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003811835506519559,
      "loss": 2.8487,
      "theoretical_loss": 3.7248297559553696,
      "tokens_seen": 809435136
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00038117352056168507,
      "loss": 2.9122,
      "theoretical_loss": 3.724799925001913,
      "tokens_seen": 809500672
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00038116349047141425,
      "loss": 3.1532,
      "theoretical_loss": 3.7247700971395865,
      "tokens_seen": 809566208
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00038115346038114343,
      "loss": 2.8558,
      "theoretical_loss": 3.7247402723678205,
      "tokens_seen": 809631744
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003811434302908726,
      "loss": 3.0254,
      "theoretical_loss": 3.7247104506860436,
      "tokens_seen": 809697280
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00038113340020060185,
      "loss": 3.0352,
      "theoretical_loss": 3.724680632093686,
      "tokens_seen": 809762816
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.000381123370110331,
      "loss": 3.2752,
      "theoretical_loss": 3.7246508165901777,
      "tokens_seen": 809828352
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003811133400200602,
      "loss": 2.917,
      "theoretical_loss": 3.7246210041749492,
      "tokens_seen": 809893888
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00038110330992978934,
      "loss": 2.9477,
      "theoretical_loss": 3.7245911948474295,
      "tokens_seen": 809959424
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003810932798395186,
      "loss": 2.8284,
      "theoretical_loss": 3.72456138860705,
      "tokens_seen": 810024960
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00038108324974924776,
      "loss": 2.7958,
      "theoretical_loss": 3.7245315854532417,
      "tokens_seen": 810090496
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00038107321965897694,
      "loss": 2.8835,
      "theoretical_loss": 3.724501785385434,
      "tokens_seen": 810156032
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003810631895687061,
      "loss": 2.9584,
      "theoretical_loss": 3.7244719884030593,
      "tokens_seen": 810221568
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00038105315947843535,
      "loss": 2.7311,
      "theoretical_loss": 3.7244421945055466,
      "tokens_seen": 810287104
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003810431293881645,
      "loss": 2.9331,
      "theoretical_loss": 3.7244124036923294,
      "tokens_seen": 810352640
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003810330992978937,
      "loss": 2.9788,
      "theoretical_loss": 3.7243826159628366,
      "tokens_seen": 810418176
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00038102306920762284,
      "loss": 2.9912,
      "theoretical_loss": 3.7243528313165015,
      "tokens_seen": 810483712
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003810130391173521,
      "loss": 3.0493,
      "theoretical_loss": 3.7243230497527553,
      "tokens_seen": 810549248
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00038100300902708126,
      "loss": 2.8376,
      "theoretical_loss": 3.724293271271029,
      "tokens_seen": 810614784
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00038099297893681044,
      "loss": 2.873,
      "theoretical_loss": 3.7242634958707557,
      "tokens_seen": 810680320
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003809829488465396,
      "loss": 2.9287,
      "theoretical_loss": 3.724233723551367,
      "tokens_seen": 810745856
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003809729187562688,
      "loss": 2.6344,
      "theoretical_loss": 3.7242039543122942,
      "tokens_seen": 810811392
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.000380962888665998,
      "loss": 3.0241,
      "theoretical_loss": 3.724174188152971,
      "tokens_seen": 810876928
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003809528585757272,
      "loss": 2.7998,
      "theoretical_loss": 3.7241444250728293,
      "tokens_seen": 810942464
    },
    {
      "epoch": 0.25,
      "objective/train/advantage_avg": 0.49091657996177673,
      "objective/train/docs_used": 466961,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.847186803817749,
      "objective/train/original_loss": 2.84718656539917,
      "objective/train/theoretical_loss": 3.7241146650713013,
      "objective/train/tokens_used": 831468000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24117104709148407,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0503175258636475,
      "objective/train/weighted_lm_loss": 2.9903175830841064,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 3.7241146650713013,
      "tokens_seen": 811008000
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00038094282848545635,
      "loss": 2.9203,
      "theoretical_loss": 3.7241146650713013,
      "tokens_seen": 811008000
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003809327983951856,
      "loss": 2.9606,
      "theoretical_loss": 3.724084908147821,
      "tokens_seen": 811073536
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00038092276830491476,
      "loss": 2.9169,
      "theoretical_loss": 3.72405515430182,
      "tokens_seen": 811139072
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00038091273821464394,
      "loss": 2.8703,
      "theoretical_loss": 3.7240254035327327,
      "tokens_seen": 811204608
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003809027081243731,
      "loss": 3.0348,
      "theoretical_loss": 3.7239956558399916,
      "tokens_seen": 811270144
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003808926780341023,
      "loss": 3.0307,
      "theoretical_loss": 3.7239659112230306,
      "tokens_seen": 811335680
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003808826479438315,
      "loss": 3.1405,
      "theoretical_loss": 3.7239361696812825,
      "tokens_seen": 811401216
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003808726178535607,
      "loss": 3.0396,
      "theoretical_loss": 3.7239064312141816,
      "tokens_seen": 811466752
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00038086258776328985,
      "loss": 2.9683,
      "theoretical_loss": 3.723876695821162,
      "tokens_seen": 811532288
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003808525576730191,
      "loss": 3.1039,
      "theoretical_loss": 3.723846963501657,
      "tokens_seen": 811597824
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003808425275827482,
      "loss": 3.0559,
      "theoretical_loss": 3.723817234255101,
      "tokens_seen": 811663360
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00038083249749247745,
      "loss": 3.2975,
      "theoretical_loss": 3.723787508080929,
      "tokens_seen": 811728896
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003808224674022066,
      "loss": 3.0926,
      "theoretical_loss": 3.7237577849785746,
      "tokens_seen": 811794432
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003808124373119358,
      "loss": 3.0162,
      "theoretical_loss": 3.7237280649474727,
      "tokens_seen": 811859968
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.000380802407221665,
      "loss": 2.7878,
      "theoretical_loss": 3.7236983479870585,
      "tokens_seen": 811925504
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00038079237713139417,
      "loss": 3.076,
      "theoretical_loss": 3.723668634096766,
      "tokens_seen": 811991040
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00038078234704112335,
      "loss": 2.8853,
      "theoretical_loss": 3.7236389232760314,
      "tokens_seen": 812056576
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003807723169508526,
      "loss": 3.1055,
      "theoretical_loss": 3.7236092155242893,
      "tokens_seen": 812122112
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003807622868605817,
      "loss": 3.0946,
      "theoretical_loss": 3.723579510840975,
      "tokens_seen": 812187648
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00038075225677031095,
      "loss": 2.9028,
      "theoretical_loss": 3.7235498092255237,
      "tokens_seen": 812253184
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00038074222668004013,
      "loss": 3.0539,
      "theoretical_loss": 3.723520110677372,
      "tokens_seen": 812318720
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003807321965897693,
      "loss": 2.9335,
      "theoretical_loss": 3.723490415195955,
      "tokens_seen": 812384256
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003807221664994985,
      "loss": 2.635,
      "theoretical_loss": 3.7234607227807084,
      "tokens_seen": 812449792
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003807121364092277,
      "loss": 3.0543,
      "theoretical_loss": 3.723431033431069,
      "tokens_seen": 812515328
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00038070210631895685,
      "loss": 2.9091,
      "theoretical_loss": 3.7234013471464733,
      "tokens_seen": 812580864
    },
    {
      "epoch": 0.25,
      "objective/train/advantage_avg": 0.49542108178138733,
      "objective/train/docs_used": 468484,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.7025020122528076,
      "objective/train/original_loss": 2.7025020122528076,
      "objective/train/theoretical_loss": 3.7233716639263568,
      "objective/train/tokens_used": 833106400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24550051987171173,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507901906967163,
      "objective/train/weighted_lm_loss": 2.8397324085235596,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0425469875335693,
      "theoretical_loss": 3.7233716639263568,
      "tokens_seen": 812646400
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003806920762286861,
      "loss": 2.8862,
      "theoretical_loss": 3.7233716639263568,
      "tokens_seen": 812646400
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003806820461384152,
      "loss": 2.7517,
      "theoretical_loss": 3.7233419837701565,
      "tokens_seen": 812711936
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00038067201604814445,
      "loss": 3.0537,
      "theoretical_loss": 3.7233123066773097,
      "tokens_seen": 812777472
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003806619859578736,
      "loss": 2.8571,
      "theoretical_loss": 3.723282632647252,
      "tokens_seen": 812843008
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003806519558676028,
      "loss": 3.0124,
      "theoretical_loss": 3.723252961679421,
      "tokens_seen": 812908544
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.000380641925777332,
      "loss": 3.1109,
      "theoretical_loss": 3.7232232937732546,
      "tokens_seen": 812974080
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003806318956870612,
      "loss": 3.1213,
      "theoretical_loss": 3.723193628928189,
      "tokens_seen": 813039616
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003806218655967904,
      "loss": 2.9821,
      "theoretical_loss": 3.723163967143662,
      "tokens_seen": 813105152
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00038061183550651954,
      "loss": 2.9721,
      "theoretical_loss": 3.7231343084191115,
      "tokens_seen": 813170688
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003806018054162488,
      "loss": 2.9857,
      "theoretical_loss": 3.723104652753975,
      "tokens_seen": 813236224
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00038059177532597796,
      "loss": 3.0237,
      "theoretical_loss": 3.7230750001476904,
      "tokens_seen": 813301760
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00038058174523570714,
      "loss": 3.0477,
      "theoretical_loss": 3.723045350599696,
      "tokens_seen": 813367296
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003805717151454363,
      "loss": 2.97,
      "theoretical_loss": 3.7230157041094296,
      "tokens_seen": 813432832
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00038056168505516555,
      "loss": 2.9826,
      "theoretical_loss": 3.7229860606763294,
      "tokens_seen": 813498368
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003805516549648947,
      "loss": 2.8904,
      "theoretical_loss": 3.722956420299835,
      "tokens_seen": 813563904
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003805416248746239,
      "loss": 2.9558,
      "theoretical_loss": 3.7229267829793837,
      "tokens_seen": 813629440
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00038053159478435304,
      "loss": 2.9302,
      "theoretical_loss": 3.7228971487144147,
      "tokens_seen": 813694976
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003805215646940823,
      "loss": 3.0662,
      "theoretical_loss": 3.7228675175043673,
      "tokens_seen": 813760512
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00038051153460381146,
      "loss": 2.9918,
      "theoretical_loss": 3.7228378893486807,
      "tokens_seen": 813826048
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00038050150451354064,
      "loss": 2.9848,
      "theoretical_loss": 3.7228082642467935,
      "tokens_seen": 813891584
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003804914744232698,
      "loss": 2.8726,
      "theoretical_loss": 3.7227786421981452,
      "tokens_seen": 813957120
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.000380481444332999,
      "loss": 2.9434,
      "theoretical_loss": 3.722749023202176,
      "tokens_seen": 814022656
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003804714142427282,
      "loss": 3.0699,
      "theoretical_loss": 3.722719407258325,
      "tokens_seen": 814088192
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003804613841524574,
      "loss": 2.9554,
      "theoretical_loss": 3.7226897943660315,
      "tokens_seen": 814153728
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00038045135406218655,
      "loss": 2.9475,
      "theoretical_loss": 3.722660184524737,
      "tokens_seen": 814219264
    },
    {
      "epoch": 0.25,
      "objective/train/advantage_avg": 0.49395135045051575,
      "objective/train/docs_used": 469177,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.9180617332458496,
      "objective/train/original_loss": 2.9180617332458496,
      "objective/train/theoretical_loss": 3.722630577733881,
      "objective/train/tokens_used": 834744800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24407067894935608,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506359338760376,
      "objective/train/weighted_lm_loss": 3.065896987915039,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0363596677780151,
      "theoretical_loss": 3.722630577733881,
      "tokens_seen": 814284800
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003804413239719158,
      "loss": 2.6715,
      "theoretical_loss": 3.722630577733881,
      "tokens_seen": 814284800
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00038043129388164496,
      "loss": 2.8138,
      "theoretical_loss": 3.7226009739929022,
      "tokens_seen": 814350336
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00038042126379137414,
      "loss": 2.9153,
      "theoretical_loss": 3.7225713733012435,
      "tokens_seen": 814415872
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003804112337011033,
      "loss": 2.9989,
      "theoretical_loss": 3.7225417756583434,
      "tokens_seen": 814481408
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003804012036108325,
      "loss": 3.1663,
      "theoretical_loss": 3.722512181063644,
      "tokens_seen": 814546944
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003803911735205617,
      "loss": 2.8929,
      "theoretical_loss": 3.722482589516585,
      "tokens_seen": 814612480
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003803811434302909,
      "loss": 3.0855,
      "theoretical_loss": 3.722453001016609,
      "tokens_seen": 814678016
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00038037111334002005,
      "loss": 3.0356,
      "theoretical_loss": 3.722423415563156,
      "tokens_seen": 814743552
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003803610832497493,
      "loss": 2.9778,
      "theoretical_loss": 3.7223938331556674,
      "tokens_seen": 814809088
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003803510531594784,
      "loss": 3.0444,
      "theoretical_loss": 3.722364253793585,
      "tokens_seen": 814874624
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00038034102306920765,
      "loss": 3.0135,
      "theoretical_loss": 3.7223346774763497,
      "tokens_seen": 814940160
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00038033099297893683,
      "loss": 3.0075,
      "theoretical_loss": 3.7223051042034037,
      "tokens_seen": 815005696
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.000380320962888666,
      "loss": 2.9442,
      "theoretical_loss": 3.7222755339741895,
      "tokens_seen": 815071232
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003803109327983952,
      "loss": 2.9982,
      "theoretical_loss": 3.7222459667881482,
      "tokens_seen": 815136768
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00038030090270812437,
      "loss": 3.0716,
      "theoretical_loss": 3.7222164026447224,
      "tokens_seen": 815202304
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00038029087261785355,
      "loss": 2.9563,
      "theoretical_loss": 3.7221868415433543,
      "tokens_seen": 815267840
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003802808425275828,
      "loss": 2.8414,
      "theoretical_loss": 3.7221572834834866,
      "tokens_seen": 815333376
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003802708124373119,
      "loss": 2.8317,
      "theoretical_loss": 3.7221277284645615,
      "tokens_seen": 815398912
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00038026078234704115,
      "loss": 2.8859,
      "theoretical_loss": 3.722098176486022,
      "tokens_seen": 815464448
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00038025075225677033,
      "loss": 3.009,
      "theoretical_loss": 3.7220686275473116,
      "tokens_seen": 815529984
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003802407221664995,
      "loss": 3.0251,
      "theoretical_loss": 3.722039081647873,
      "tokens_seen": 815595520
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003802306920762287,
      "loss": 2.6212,
      "theoretical_loss": 3.7220095387871486,
      "tokens_seen": 815661056
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003802206619859579,
      "loss": 2.9833,
      "theoretical_loss": 3.7219799989645828,
      "tokens_seen": 815726592
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00038021063189568705,
      "loss": 3.0254,
      "theoretical_loss": 3.7219504621796187,
      "tokens_seen": 815792128
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003802006018054163,
      "loss": 3.1409,
      "theoretical_loss": 3.7219209284317,
      "tokens_seen": 815857664
    },
    {
      "epoch": 0.25,
      "objective/train/advantage_avg": 0.49331632256507874,
      "objective/train/docs_used": 469961,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.7106359004974365,
      "objective/train/original_loss": 2.7106356620788574,
      "objective/train/theoretical_loss": 3.7218913977202703,
      "objective/train/tokens_used": 836383200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24342210590839386,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505690574645996,
      "objective/train/weighted_lm_loss": 2.8477463722229004,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0442860126495361,
      "theoretical_loss": 3.7218913977202703,
      "tokens_seen": 815923200
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003801905717151454,
      "loss": 2.6726,
      "theoretical_loss": 3.7218913977202703,
      "tokens_seen": 815923200
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00038018054162487465,
      "loss": 2.7972,
      "theoretical_loss": 3.7218618700447736,
      "tokens_seen": 815988736
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003801705115346038,
      "loss": 2.952,
      "theoretical_loss": 3.7218323454046547,
      "tokens_seen": 816054272
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.000380160481444333,
      "loss": 2.9864,
      "theoretical_loss": 3.721802823799357,
      "tokens_seen": 816119808
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003801504513540622,
      "loss": 3.0238,
      "theoretical_loss": 3.721773305228325,
      "tokens_seen": 816185344
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003801404212637914,
      "loss": 3.0939,
      "theoretical_loss": 3.7217437896910037,
      "tokens_seen": 816250880
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00038013039117352056,
      "loss": 2.9148,
      "theoretical_loss": 3.7217142771868374,
      "tokens_seen": 816316416
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00038013039117352056,
      "loss": 3.1672,
      "theoretical_loss": 3.721684767715271,
      "tokens_seen": 816381952
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00038012036108324974,
      "loss": 2.8647,
      "theoretical_loss": 3.7216552612757496,
      "tokens_seen": 816447488
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003801103309929789,
      "loss": 3.1614,
      "theoretical_loss": 3.721625757867718,
      "tokens_seen": 816513024
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00038010030090270816,
      "loss": 2.9348,
      "theoretical_loss": 3.7215962574906216,
      "tokens_seen": 816578560
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003800902708124373,
      "loss": 3.016,
      "theoretical_loss": 3.721566760143906,
      "tokens_seen": 816644096
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003800802407221665,
      "loss": 2.719,
      "theoretical_loss": 3.721537265827017,
      "tokens_seen": 816709632
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003800702106318957,
      "loss": 2.8392,
      "theoretical_loss": 3.7215077745393996,
      "tokens_seen": 816775168
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003800601805416249,
      "loss": 2.594,
      "theoretical_loss": 3.7214782862805,
      "tokens_seen": 816840704
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00038005015045135406,
      "loss": 2.8187,
      "theoretical_loss": 3.721448801049764,
      "tokens_seen": 816906240
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00038004012036108324,
      "loss": 3.0614,
      "theoretical_loss": 3.721419318846638,
      "tokens_seen": 816971776
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003800300902708124,
      "loss": 2.9639,
      "theoretical_loss": 3.721389839670568,
      "tokens_seen": 817037312
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00038002006018054166,
      "loss": 2.7783,
      "theoretical_loss": 3.7213603635210015,
      "tokens_seen": 817102848
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003800100300902708,
      "loss": 2.783,
      "theoretical_loss": 3.7213308903973834,
      "tokens_seen": 817168384
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00038,
      "loss": 2.8579,
      "theoretical_loss": 3.7213014202991612,
      "tokens_seen": 817233920
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037998996990972915,
      "loss": 2.7067,
      "theoretical_loss": 3.7212719532257825,
      "tokens_seen": 817299456
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003799799398194584,
      "loss": 2.9881,
      "theoretical_loss": 3.721242489176693,
      "tokens_seen": 817364992
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037996990972918756,
      "loss": 3.0582,
      "theoretical_loss": 3.721213028151341,
      "tokens_seen": 817430528
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037995987963891675,
      "loss": 2.9171,
      "theoretical_loss": 3.721183570149173,
      "tokens_seen": 817496064
    },
    {
      "epoch": 0.25,
      "objective/train/advantage_avg": 0.4947288930416107,
      "objective/train/docs_used": 471203,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.782205581665039,
      "objective/train/original_loss": 2.78220534324646,
      "objective/train/theoretical_loss": 3.721154115169637,
      "objective/train/tokens_used": 838021600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24481016397476196,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507172346115112,
      "objective/train/weighted_lm_loss": 2.9231717586517334,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0304546356201172,
      "theoretical_loss": 3.721154115169637,
      "tokens_seen": 817561600
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003799498495486459,
      "loss": 2.8939,
      "theoretical_loss": 3.721154115169637,
      "tokens_seen": 817561600
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037993981945837516,
      "loss": 3.0062,
      "theoretical_loss": 3.72112466321218,
      "tokens_seen": 817627136
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003799297893681043,
      "loss": 3.1195,
      "theoretical_loss": 3.72109521427625,
      "tokens_seen": 817692672
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003799197592778335,
      "loss": 2.8732,
      "theoretical_loss": 3.721065768361295,
      "tokens_seen": 817758208
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037990972918756265,
      "loss": 2.8947,
      "theoretical_loss": 3.721036325466763,
      "tokens_seen": 817823744
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003798996990972919,
      "loss": 3.2079,
      "theoretical_loss": 3.7210068855921024,
      "tokens_seen": 817889280
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037988966900702107,
      "loss": 3.0225,
      "theoretical_loss": 3.720977448736761,
      "tokens_seen": 817954816
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037987963891675025,
      "loss": 2.9291,
      "theoretical_loss": 3.7209480149001877,
      "tokens_seen": 818020352
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003798696088264795,
      "loss": 2.9851,
      "theoretical_loss": 3.720918584081831,
      "tokens_seen": 818085888
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003798595787362086,
      "loss": 3.1586,
      "theoretical_loss": 3.72088915628114,
      "tokens_seen": 818151424
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037984954864593785,
      "loss": 3.0951,
      "theoretical_loss": 3.720859731497563,
      "tokens_seen": 818216960
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037983951855566703,
      "loss": 2.895,
      "theoretical_loss": 3.720830309730549,
      "tokens_seen": 818282496
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003798294884653962,
      "loss": 3.0073,
      "theoretical_loss": 3.7208008909795476,
      "tokens_seen": 818348032
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003798194583751254,
      "loss": 2.7553,
      "theoretical_loss": 3.7207714752440078,
      "tokens_seen": 818413568
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037980942828485457,
      "loss": 3.0556,
      "theoretical_loss": 3.72074206252338,
      "tokens_seen": 818479104
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037979939819458375,
      "loss": 2.7832,
      "theoretical_loss": 3.720712652817112,
      "tokens_seen": 818544640
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.000379789368104313,
      "loss": 2.8073,
      "theoretical_loss": 3.720683246124655,
      "tokens_seen": 818610176
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003797793380140421,
      "loss": 3.0211,
      "theoretical_loss": 3.7206538424454587,
      "tokens_seen": 818675712
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037976930792377135,
      "loss": 3.0912,
      "theoretical_loss": 3.7206244417789733,
      "tokens_seen": 818741248
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037975927783350053,
      "loss": 2.9811,
      "theoretical_loss": 3.720595044124649,
      "tokens_seen": 818806784
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003797492477432297,
      "loss": 3.0856,
      "theoretical_loss": 3.7205656494819355,
      "tokens_seen": 818872320
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003797392176529589,
      "loss": 2.905,
      "theoretical_loss": 3.7205362578502834,
      "tokens_seen": 818937856
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003797291875626881,
      "loss": 2.9996,
      "theoretical_loss": 3.720506869229144,
      "tokens_seen": 819003392
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037971915747241725,
      "loss": 3.008,
      "theoretical_loss": 3.7204774836179673,
      "tokens_seen": 819068928
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003797091273821465,
      "loss": 2.7749,
      "theoretical_loss": 3.7204481010162045,
      "tokens_seen": 819134464
    },
    {
      "debugging/Self-BLEU-5": 0.23022853119814032,
      "debugging/distinct-1-grams": 0.7520081804979636,
      "debugging/distinct-2-grams": 0.9205321507760532,
      "debugging/entropy-1-grams": 4.962017398067815,
      "debugging/entropy-2-grams": 5.4064809924091275,
      "debugging/length": 494.3333333333333,
      "debugging/num_segments": 3,
      "debugging/raw_token_scores_avg": 0.004860436078161001,
      "debugging/raw_token_scores_std": 0.008490439504384995,
      "debugging/score": 0.006766520013802623,
      "debugging/score_std": 0.00814926718169,
      "epoch": 0.25,
      "objective/train/advantage_avg": 0.4951414465904236,
      "objective/train/docs_used": 471797,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.117976427078247,
      "objective/train/original_loss": 3.1179757118225098,
      "objective/train/theoretical_loss": 3.7204187214233073,
      "objective/train/tokens_used": 839660000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24523712694644928,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507608652114868,
      "objective/train/weighted_lm_loss": 3.276245355606079,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0304546356201172,
      "theoretical_loss": 3.7204187214233073,
      "tokens_seen": 819200000
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003796990972918756,
      "loss": 3.0255,
      "theoretical_loss": 3.7204187214233073,
      "tokens_seen": 819200000
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037968906720160485,
      "loss": 3.008,
      "theoretical_loss": 3.7203893448387264,
      "tokens_seen": 819265536
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.000379679037111334,
      "loss": 2.4814,
      "theoretical_loss": 3.720359971261913,
      "tokens_seen": 819331072
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003796690070210632,
      "loss": 3.0407,
      "theoretical_loss": 3.7203306006923187,
      "tokens_seen": 819396608
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003796589769307924,
      "loss": 2.8323,
      "theoretical_loss": 3.720301233129395,
      "tokens_seen": 819462144
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003796489468405216,
      "loss": 2.8319,
      "theoretical_loss": 3.720271868572594,
      "tokens_seen": 819527680
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037963891675025076,
      "loss": 2.8199,
      "theoretical_loss": 3.720242507021368,
      "tokens_seen": 819593216
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037962888665997994,
      "loss": 2.9572,
      "theoretical_loss": 3.720213148475168,
      "tokens_seen": 819658752
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003796188565697091,
      "loss": 3.0861,
      "theoretical_loss": 3.7201837929334465,
      "tokens_seen": 819724288
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037960882647943836,
      "loss": 2.9087,
      "theoretical_loss": 3.720154440395657,
      "tokens_seen": 819789824
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003795987963891675,
      "loss": 2.953,
      "theoretical_loss": 3.7201250908612504,
      "tokens_seen": 819855360
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003795887662988967,
      "loss": 3.0596,
      "theoretical_loss": 3.7200957443296803,
      "tokens_seen": 819920896
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003795787362086259,
      "loss": 2.8559,
      "theoretical_loss": 3.7200664008004,
      "tokens_seen": 819986432
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003795687061183551,
      "loss": 2.8587,
      "theoretical_loss": 3.720037060272861,
      "tokens_seen": 820051968
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037955867602808426,
      "loss": 2.9241,
      "theoretical_loss": 3.720007722746517,
      "tokens_seen": 820117504
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037954864593781344,
      "loss": 3.0852,
      "theoretical_loss": 3.719978388220822,
      "tokens_seen": 820183040
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003795386158475426,
      "loss": 2.9186,
      "theoretical_loss": 3.7199490566952287,
      "tokens_seen": 820248576
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037952858575727186,
      "loss": 2.9509,
      "theoretical_loss": 3.71991972816919,
      "tokens_seen": 820314112
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.000379518555667001,
      "loss": 2.9066,
      "theoretical_loss": 3.7198904026421604,
      "tokens_seen": 820379648
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003795085255767302,
      "loss": 2.7275,
      "theoretical_loss": 3.719861080113594,
      "tokens_seen": 820445184
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037949849548645935,
      "loss": 3.0378,
      "theoretical_loss": 3.7198317605829434,
      "tokens_seen": 820510720
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003794884653961886,
      "loss": 3.1758,
      "theoretical_loss": 3.7198024440496638,
      "tokens_seen": 820576256
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037947843530591776,
      "loss": 2.9461,
      "theoretical_loss": 3.7197731305132087,
      "tokens_seen": 820641792
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037946840521564695,
      "loss": 3.1083,
      "theoretical_loss": 3.7197438199730337,
      "tokens_seen": 820707328
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003794583751253761,
      "loss": 2.8721,
      "theoretical_loss": 3.7197145124285917,
      "tokens_seen": 820772864
    },
    {
      "epoch": 0.25,
      "objective/train/advantage_avg": 0.49458253383636475,
      "objective/train/docs_used": 473281,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.821866989135742,
      "objective/train/original_loss": 2.821866989135742,
      "objective/train/theoretical_loss": 3.7196852078793388,
      "objective/train/tokens_used": 841298400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24468523263931274,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507019758224487,
      "objective/train/weighted_lm_loss": 2.9647021293640137,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0253151655197144,
      "theoretical_loss": 3.7196852078793388,
      "tokens_seen": 820838400
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037944834503510536,
      "loss": 3.0879,
      "theoretical_loss": 3.7196852078793388,
      "tokens_seen": 820838400
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003794383149448345,
      "loss": 3.1875,
      "theoretical_loss": 3.7196559063247285,
      "tokens_seen": 820903936
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003794282848545637,
      "loss": 2.9159,
      "theoretical_loss": 3.719626607764217,
      "tokens_seen": 820969472
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037941825476429285,
      "loss": 2.8344,
      "theoretical_loss": 3.7195973121972585,
      "tokens_seen": 821035008
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003794082246740221,
      "loss": 3.077,
      "theoretical_loss": 3.719568019623308,
      "tokens_seen": 821100544
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037939819458375127,
      "loss": 2.8459,
      "theoretical_loss": 3.7195387300418217,
      "tokens_seen": 821166080
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037938816449348045,
      "loss": 2.9603,
      "theoretical_loss": 3.719509443452255,
      "tokens_seen": 821231616
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037937813440320963,
      "loss": 2.7662,
      "theoretical_loss": 3.719480159854063,
      "tokens_seen": 821297152
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003793681043129388,
      "loss": 2.8312,
      "theoretical_loss": 3.7194508792467023,
      "tokens_seen": 821362688
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.000379358074222668,
      "loss": 2.7198,
      "theoretical_loss": 3.719421601629628,
      "tokens_seen": 821428224
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037934804413239723,
      "loss": 2.8484,
      "theoretical_loss": 3.7193923270022964,
      "tokens_seen": 821493760
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037933801404212635,
      "loss": 2.9813,
      "theoretical_loss": 3.7193630553641643,
      "tokens_seen": 821559296
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003793279839518556,
      "loss": 3.1382,
      "theoretical_loss": 3.719333786714687,
      "tokens_seen": 821624832
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003793179538615847,
      "loss": 2.9949,
      "theoretical_loss": 3.7193045210533215,
      "tokens_seen": 821690368
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037930792377131395,
      "loss": 3.0328,
      "theoretical_loss": 3.7192752583795254,
      "tokens_seen": 821755904
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037929789368104313,
      "loss": 2.7242,
      "theoretical_loss": 3.719245998692754,
      "tokens_seen": 821821440
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003792878635907723,
      "loss": 3.0078,
      "theoretical_loss": 3.719216741992465,
      "tokens_seen": 821886976
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003792778335005015,
      "loss": 2.9959,
      "theoretical_loss": 3.7191874882781155,
      "tokens_seen": 821952512
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037926780341023073,
      "loss": 3.0404,
      "theoretical_loss": 3.7191582375491623,
      "tokens_seen": 822018048
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037925777331995986,
      "loss": 3.0461,
      "theoretical_loss": 3.7191289898050632,
      "tokens_seen": 822083584
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003792477432296891,
      "loss": 3.0691,
      "theoretical_loss": 3.719099745045275,
      "tokens_seen": 822149120
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003792377131394182,
      "loss": 3.0342,
      "theoretical_loss": 3.719070503269257,
      "tokens_seen": 822214656
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037922768304914746,
      "loss": 2.9443,
      "theoretical_loss": 3.719041264476465,
      "tokens_seen": 822280192
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037921765295887664,
      "loss": 3.0896,
      "theoretical_loss": 3.719012028666358,
      "tokens_seen": 822345728
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003792076228686058,
      "loss": 2.9536,
      "theoretical_loss": 3.718982795838394,
      "tokens_seen": 822411264
    },
    {
      "epoch": 0.25,
      "objective/train/advantage_avg": 0.4928377866744995,
      "objective/train/docs_used": 473825,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.0089595317840576,
      "objective/train/original_loss": 3.008960008621216,
      "objective/train/theoretical_loss": 3.718953565992031,
      "objective/train/tokens_used": 842936800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24299494922161102,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505189895629883,
      "objective/train/weighted_lm_loss": 3.160900354385376,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0304546356201172,
      "theoretical_loss": 3.718953565992031,
      "tokens_seen": 822476800
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.000379197592778335,
      "loss": 2.8832,
      "theoretical_loss": 3.718953565992031,
      "tokens_seen": 822476800
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003791875626880642,
      "loss": 2.9099,
      "theoretical_loss": 3.718924339126727,
      "tokens_seen": 822542336
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037917753259779336,
      "loss": 2.9728,
      "theoretical_loss": 3.718895115241941,
      "tokens_seen": 822607872
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003791675025075226,
      "loss": 3.0232,
      "theoretical_loss": 3.7188658943371316,
      "tokens_seen": 822673408
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003791574724172517,
      "loss": 2.9655,
      "theoretical_loss": 3.7188366764117577,
      "tokens_seen": 822738944
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037914744232698096,
      "loss": 3.036,
      "theoretical_loss": 3.718807461465278,
      "tokens_seen": 822804480
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003791374122367101,
      "loss": 2.9552,
      "theoretical_loss": 3.7187782494971517,
      "tokens_seen": 822870016
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003791273821464393,
      "loss": 2.9869,
      "theoretical_loss": 3.7187490405068377,
      "tokens_seen": 822935552
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037911735205616856,
      "loss": 2.8628,
      "theoretical_loss": 3.718719834493795,
      "tokens_seen": 823001088
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003791073219658977,
      "loss": 3.1358,
      "theoretical_loss": 3.718690631457484,
      "tokens_seen": 823066624
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003790972918756269,
      "loss": 3.1293,
      "theoretical_loss": 3.7186614313973645,
      "tokens_seen": 823132160
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003790872617853561,
      "loss": 2.9471,
      "theoretical_loss": 3.7186322343128952,
      "tokens_seen": 823197696
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003790772316950853,
      "loss": 2.9245,
      "theoretical_loss": 3.7186030402035364,
      "tokens_seen": 823263232
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037906720160481446,
      "loss": 3.0664,
      "theoretical_loss": 3.7185738490687483,
      "tokens_seen": 823328768
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037905717151454364,
      "loss": 2.9173,
      "theoretical_loss": 3.718544660907991,
      "tokens_seen": 823394304
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003790471414242728,
      "loss": 2.7726,
      "theoretical_loss": 3.718515475720725,
      "tokens_seen": 823459840
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037903711133400206,
      "loss": 3.1051,
      "theoretical_loss": 3.7184862935064102,
      "tokens_seen": 823525376
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003790270812437312,
      "loss": 2.7553,
      "theoretical_loss": 3.718457114264508,
      "tokens_seen": 823590912
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003790170511534604,
      "loss": 2.8911,
      "theoretical_loss": 3.7184279379944787,
      "tokens_seen": 823656448
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037900702106318955,
      "loss": 2.9274,
      "theoretical_loss": 3.7183987646957832,
      "tokens_seen": 823721984
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003789969909729188,
      "loss": 2.8392,
      "theoretical_loss": 3.718369594367883,
      "tokens_seen": 823787520
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037898696088264796,
      "loss": 3.0634,
      "theoretical_loss": 3.7183404270102383,
      "tokens_seen": 823853056
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037897693079237715,
      "loss": 3.1138,
      "theoretical_loss": 3.7183112626223114,
      "tokens_seen": 823918592
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003789669007021063,
      "loss": 2.7579,
      "theoretical_loss": 3.718282101203563,
      "tokens_seen": 823984128
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037895687061183556,
      "loss": 2.8456,
      "theoretical_loss": 3.718252942753455,
      "tokens_seen": 824049664
    },
    {
      "epoch": 0.25,
      "objective/train/advantage_avg": 0.49377477169036865,
      "objective/train/docs_used": 475209,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.0270330905914307,
      "objective/train/original_loss": 3.0270330905914307,
      "objective/train/theoretical_loss": 3.718223787271449,
      "objective/train/tokens_used": 844575200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24388201534748077,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506173372268677,
      "objective/train/weighted_lm_loss": 3.180253028869629,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0396549701690674,
      "theoretical_loss": 3.718223787271449,
      "tokens_seen": 824115200
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003789468405215647,
      "loss": 3.1148,
      "theoretical_loss": 3.718223787271449,
      "tokens_seen": 824115200
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003789368104312939,
      "loss": 2.8526,
      "theoretical_loss": 3.7181946347570074,
      "tokens_seen": 824180736
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037892678034102305,
      "loss": 3.0076,
      "theoretical_loss": 3.7181654852095924,
      "tokens_seen": 824246272
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003789167502507523,
      "loss": 2.9769,
      "theoretical_loss": 3.718136338628665,
      "tokens_seen": 824311808
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037890672016048147,
      "loss": 2.993,
      "theoretical_loss": 3.7181071950136877,
      "tokens_seen": 824377344
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037889669007021065,
      "loss": 2.6878,
      "theoretical_loss": 3.718078054364124,
      "tokens_seen": 824442880
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037888665997993983,
      "loss": 3.1302,
      "theoretical_loss": 3.7180489166794355,
      "tokens_seen": 824508416
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.000378876629889669,
      "loss": 3.0376,
      "theoretical_loss": 3.718019781959085,
      "tokens_seen": 824573952
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003788665997993982,
      "loss": 3.0002,
      "theoretical_loss": 3.717990650202536,
      "tokens_seen": 824639488
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037885656970912743,
      "loss": 3.0572,
      "theoretical_loss": 3.717961521409251,
      "tokens_seen": 824705024
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037884653961885655,
      "loss": 2.8752,
      "theoretical_loss": 3.7179323955786936,
      "tokens_seen": 824770560
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003788365095285858,
      "loss": 2.8239,
      "theoretical_loss": 3.717903272710326,
      "tokens_seen": 824836096
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003788264794383149,
      "loss": 2.8298,
      "theoretical_loss": 3.7178741528036126,
      "tokens_seen": 824901632
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037881644934804415,
      "loss": 2.8541,
      "theoretical_loss": 3.7178450358580166,
      "tokens_seen": 824967168
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037880641925777333,
      "loss": 2.775,
      "theoretical_loss": 3.717815921873002,
      "tokens_seen": 825032704
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003787963891675025,
      "loss": 2.8909,
      "theoretical_loss": 3.7177868108480316,
      "tokens_seen": 825098240
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003787863590772317,
      "loss": 2.9686,
      "theoretical_loss": 3.717757702782571,
      "tokens_seen": 825163776
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037877632898696093,
      "loss": 2.7503,
      "theoretical_loss": 3.7177285976760834,
      "tokens_seen": 825229312
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037876629889669006,
      "loss": 2.9247,
      "theoretical_loss": 3.7176994955280334,
      "tokens_seen": 825294848
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003787562688064193,
      "loss": 3.0679,
      "theoretical_loss": 3.7176703963378843,
      "tokens_seen": 825360384
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003787462387161484,
      "loss": 3.0441,
      "theoretical_loss": 3.717641300105102,
      "tokens_seen": 825425920
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037873620862587766,
      "loss": 2.8713,
      "theoretical_loss": 3.71761220682915,
      "tokens_seen": 825491456
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037872617853560684,
      "loss": 3.0619,
      "theoretical_loss": 3.717583116509495,
      "tokens_seen": 825556992
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.000378716148445336,
      "loss": 2.7972,
      "theoretical_loss": 3.7175540291455995,
      "tokens_seen": 825622528
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003787061183550652,
      "loss": 3.0082,
      "theoretical_loss": 3.7175249447369305,
      "tokens_seen": 825688064
    },
    {
      "epoch": 0.25,
      "objective/train/advantage_avg": 0.4932311177253723,
      "objective/train/docs_used": 475848,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.9069979190826416,
      "objective/train/original_loss": 2.9069979190826416,
      "objective/train/theoretical_loss": 3.7174958632829522,
      "objective/train/tokens_used": 846213600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24334900081157684,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505601167678833,
      "objective/train/weighted_lm_loss": 3.053992509841919,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0338950157165527,
      "theoretical_loss": 3.7174958632829522,
      "tokens_seen": 825753600
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003786960882647944,
      "loss": 3.1825,
      "theoretical_loss": 3.7174958632829522,
      "tokens_seen": 825753600
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037868605817452356,
      "loss": 3.134,
      "theoretical_loss": 3.71746678478313,
      "tokens_seen": 825819136
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003786760280842528,
      "loss": 2.8953,
      "theoretical_loss": 3.71743770923693,
      "tokens_seen": 825884672
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003786659979939819,
      "loss": 3.1676,
      "theoretical_loss": 3.7174086366438175,
      "tokens_seen": 825950208
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037865596790371116,
      "loss": 3.0914,
      "theoretical_loss": 3.7173795670032588,
      "tokens_seen": 826015744
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003786459378134403,
      "loss": 2.9534,
      "theoretical_loss": 3.717350500314719,
      "tokens_seen": 826081280
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003786359077231695,
      "loss": 2.9898,
      "theoretical_loss": 3.7173214365776643,
      "tokens_seen": 826146816
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003786258776328987,
      "loss": 2.8737,
      "theoretical_loss": 3.7172923757915615,
      "tokens_seen": 826212352
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003786158475426279,
      "loss": 3.0758,
      "theoretical_loss": 3.7172633179558763,
      "tokens_seen": 826277888
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037860581745235706,
      "loss": 3.0224,
      "theoretical_loss": 3.7172342630700754,
      "tokens_seen": 826343424
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003785957873620863,
      "loss": 3.0426,
      "theoretical_loss": 3.7172052111336256,
      "tokens_seen": 826408960
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003785857572718154,
      "loss": 2.958,
      "theoretical_loss": 3.7171761621459933,
      "tokens_seen": 826474496
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037857572718154466,
      "loss": 3.1097,
      "theoretical_loss": 3.717147116106646,
      "tokens_seen": 826540032
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003785656970912738,
      "loss": 2.9105,
      "theoretical_loss": 3.71711807301505,
      "tokens_seen": 826605568
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.000378555667001003,
      "loss": 2.8808,
      "theoretical_loss": 3.717089032870673,
      "tokens_seen": 826671104
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003785456369107322,
      "loss": 2.9985,
      "theoretical_loss": 3.717059995672982,
      "tokens_seen": 826736640
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003785356068204614,
      "loss": 2.9627,
      "theoretical_loss": 3.717030961421445,
      "tokens_seen": 826802176
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037852557673019057,
      "loss": 3.0775,
      "theoretical_loss": 3.717001930115529,
      "tokens_seen": 826867712
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037851554663991975,
      "loss": 3.0438,
      "theoretical_loss": 3.7169729017547013,
      "tokens_seen": 826933248
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037850551654964893,
      "loss": 2.9833,
      "theoretical_loss": 3.716943876338431,
      "tokens_seen": 826998784
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037849548645937816,
      "loss": 2.7657,
      "theoretical_loss": 3.716914853866185,
      "tokens_seen": 827064320
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003784854563691073,
      "loss": 2.8757,
      "theoretical_loss": 3.7168858343374325,
      "tokens_seen": 827129856
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003784754262788365,
      "loss": 2.7459,
      "theoretical_loss": 3.7168568177516415,
      "tokens_seen": 827195392
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037846539618856565,
      "loss": 2.911,
      "theoretical_loss": 3.716827804108279,
      "tokens_seen": 827260928
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003784553660982949,
      "loss": 3.2261,
      "theoretical_loss": 3.7167987934068156,
      "tokens_seen": 827326464
    },
    {
      "epoch": 0.25,
      "objective/train/advantage_avg": 0.4945508539676666,
      "objective/train/docs_used": 477214,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.6662185192108154,
      "objective/train/original_loss": 2.6662182807922363,
      "objective/train/theoretical_loss": 3.7167697856467186,
      "objective/train/tokens_used": 847852000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24463361501693726,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506986379623413,
      "objective/train/weighted_lm_loss": 2.8012759685516357,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0338950157165527,
      "theoretical_loss": 3.7167697856467186,
      "tokens_seen": 827392000
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037844533600802407,
      "loss": 2.902,
      "theoretical_loss": 3.7167697856467186,
      "tokens_seen": 827392000
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037843530591775325,
      "loss": 2.9618,
      "theoretical_loss": 3.7167407808274575,
      "tokens_seen": 827457536
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037842527582748243,
      "loss": 2.9073,
      "theoretical_loss": 3.716711778948501,
      "tokens_seen": 827523072
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037841524573721167,
      "loss": 3.0514,
      "theoretical_loss": 3.716682780009318,
      "tokens_seen": 827588608
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003784052156469408,
      "loss": 2.8054,
      "theoretical_loss": 3.7166537840093783,
      "tokens_seen": 827654144
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037839518555667003,
      "loss": 2.9166,
      "theoretical_loss": 3.7166247909481513,
      "tokens_seen": 827719680
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037838515546639916,
      "loss": 2.8938,
      "theoretical_loss": 3.716595800825106,
      "tokens_seen": 827785216
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003783751253761284,
      "loss": 3.1126,
      "theoretical_loss": 3.7165668136397123,
      "tokens_seen": 827850752
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037836509528585763,
      "loss": 2.89,
      "theoretical_loss": 3.7165378293914397,
      "tokens_seen": 827916288
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037835506519558675,
      "loss": 3.1495,
      "theoretical_loss": 3.7165088480797595,
      "tokens_seen": 827981824
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.000378345035105316,
      "loss": 2.8809,
      "theoretical_loss": 3.7164798697041395,
      "tokens_seen": 828047360
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003783350050150451,
      "loss": 2.9354,
      "theoretical_loss": 3.716450894264052,
      "tokens_seen": 828112896
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037832497492477435,
      "loss": 2.9157,
      "theoretical_loss": 3.7164219217589656,
      "tokens_seen": 828178432
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037831494483450353,
      "loss": 3.0413,
      "theoretical_loss": 3.716392952188352,
      "tokens_seen": 828243968
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003783049147442327,
      "loss": 2.905,
      "theoretical_loss": 3.716363985551682,
      "tokens_seen": 828309504
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003782948846539619,
      "loss": 2.9097,
      "theoretical_loss": 3.7163350218484252,
      "tokens_seen": 828375040
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037828485456369113,
      "loss": 2.8509,
      "theoretical_loss": 3.716306061078054,
      "tokens_seen": 828440576
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037827482447342026,
      "loss": 3.0339,
      "theoretical_loss": 3.7162771032400377,
      "tokens_seen": 828506112
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003782647943831495,
      "loss": 3.0581,
      "theoretical_loss": 3.7162481483338485,
      "tokens_seen": 828571648
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003782547642928786,
      "loss": 2.9137,
      "theoretical_loss": 3.716219196358958,
      "tokens_seen": 828637184
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037824473420260786,
      "loss": 2.7694,
      "theoretical_loss": 3.716190247314837,
      "tokens_seen": 828702720
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037823470411233704,
      "loss": 2.9973,
      "theoretical_loss": 3.716161301200957,
      "tokens_seen": 828768256
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003782246740220662,
      "loss": 3.1689,
      "theoretical_loss": 3.7161323580167904,
      "tokens_seen": 828833792
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003782146439317954,
      "loss": 2.9059,
      "theoretical_loss": 3.7161034177618086,
      "tokens_seen": 828899328
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003782046138415246,
      "loss": 2.7722,
      "theoretical_loss": 3.7160744804354833,
      "tokens_seen": 828964864
    },
    {
      "epoch": 0.25,
      "objective/train/advantage_avg": 0.4948139786720276,
      "objective/train/docs_used": 478014,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.0577733516693115,
      "objective/train/original_loss": 2.0577731132507324,
      "objective/train/theoretical_loss": 3.716045546037287,
      "objective/train/tokens_used": 849490400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24491508305072784,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507264137268066,
      "objective/train/weighted_lm_loss": 2.1620078086853027,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0352214574813843,
      "theoretical_loss": 3.716045546037287,
      "tokens_seen": 829030400
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037819458375125376,
      "loss": 2.6762,
      "theoretical_loss": 3.716045546037287,
      "tokens_seen": 829030400
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.000378184553660983,
      "loss": 2.8712,
      "theoretical_loss": 3.7160166145666924,
      "tokens_seen": 829095936
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003781745235707121,
      "loss": 2.9788,
      "theoretical_loss": 3.7159876860231713,
      "tokens_seen": 829161472
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037816449348044136,
      "loss": 2.8602,
      "theoretical_loss": 3.7159587604061963,
      "tokens_seen": 829227008
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003781544633901705,
      "loss": 2.9196,
      "theoretical_loss": 3.7159298377152403,
      "tokens_seen": 829292544
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003781444332998997,
      "loss": 2.9219,
      "theoretical_loss": 3.715900917949776,
      "tokens_seen": 829358080
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003781344032096289,
      "loss": 2.8034,
      "theoretical_loss": 3.7158720011092767,
      "tokens_seen": 829423616
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003781243731193581,
      "loss": 2.9423,
      "theoretical_loss": 3.715843087193215,
      "tokens_seen": 829489152
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037811434302908726,
      "loss": 3.0182,
      "theoretical_loss": 3.715814176201064,
      "tokens_seen": 829554688
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003781043129388165,
      "loss": 2.9104,
      "theoretical_loss": 3.715785268132297,
      "tokens_seen": 829620224
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003780942828485456,
      "loss": 2.9083,
      "theoretical_loss": 3.715756362986389,
      "tokens_seen": 829685760
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037808425275827486,
      "loss": 2.8163,
      "theoretical_loss": 3.7157274607628117,
      "tokens_seen": 829751296
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.000378074222668004,
      "loss": 3.0257,
      "theoretical_loss": 3.71569856146104,
      "tokens_seen": 829816832
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003780641925777332,
      "loss": 2.8551,
      "theoretical_loss": 3.7156696650805467,
      "tokens_seen": 829882368
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003780541624874624,
      "loss": 2.6453,
      "theoretical_loss": 3.7156407716208073,
      "tokens_seen": 829947904
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003780441323971916,
      "loss": 2.8287,
      "theoretical_loss": 3.715611881081295,
      "tokens_seen": 830013440
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037803410230692077,
      "loss": 3.0752,
      "theoretical_loss": 3.7155829934614846,
      "tokens_seen": 830078976
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037802407221664995,
      "loss": 2.8281,
      "theoretical_loss": 3.71555410876085,
      "tokens_seen": 830144512
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037801404212637913,
      "loss": 2.7386,
      "theoretical_loss": 3.7155252269788663,
      "tokens_seen": 830210048
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037800401203610836,
      "loss": 3.0893,
      "theoretical_loss": 3.7154963481150087,
      "tokens_seen": 830275584
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003779939819458375,
      "loss": 2.7091,
      "theoretical_loss": 3.7154674721687506,
      "tokens_seen": 830341120
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037798395185556673,
      "loss": 2.7918,
      "theoretical_loss": 3.715438599139568,
      "tokens_seen": 830406656
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037797392176529585,
      "loss": 3.0956,
      "theoretical_loss": 3.715409729026936,
      "tokens_seen": 830472192
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003779638916750251,
      "loss": 3.0128,
      "theoretical_loss": 3.7153808618303295,
      "tokens_seen": 830537728
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037795386158475427,
      "loss": 2.9754,
      "theoretical_loss": 3.7153519975492246,
      "tokens_seen": 830603264
    },
    {
      "epoch": 0.25,
      "objective/train/advantage_avg": 0.49475374817848206,
      "objective/train/docs_used": 479315,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.7887260913848877,
      "objective/train/original_loss": 2.7887258529663086,
      "objective/train/theoretical_loss": 3.7153231361830956,
      "objective/train/tokens_used": 851128800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24486500024795532,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507200956344604,
      "objective/train/weighted_lm_loss": 2.9301090240478516,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 3.7153231361830956,
      "tokens_seen": 830668800
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037794383149448345,
      "loss": 2.7198,
      "theoretical_loss": 3.7153231361830956,
      "tokens_seen": 830668800
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037793380140421263,
      "loss": 2.8006,
      "theoretical_loss": 3.7152942777314197,
      "tokens_seen": 830734336
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037792377131394187,
      "loss": 3.0497,
      "theoretical_loss": 3.7152654221936716,
      "tokens_seen": 830799872
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.000377913741223671,
      "loss": 2.8028,
      "theoretical_loss": 3.7152365695693277,
      "tokens_seen": 830865408
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037790371113340023,
      "loss": 2.9783,
      "theoretical_loss": 3.715207719857864,
      "tokens_seen": 830930944
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037789368104312936,
      "loss": 3.1719,
      "theoretical_loss": 3.7151788730587567,
      "tokens_seen": 830996480
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003778836509528586,
      "loss": 2.9419,
      "theoretical_loss": 3.715150029171482,
      "tokens_seen": 831062016
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003778736208625878,
      "loss": 2.8463,
      "theoretical_loss": 3.715121188195517,
      "tokens_seen": 831127552
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037786359077231695,
      "loss": 2.9683,
      "theoretical_loss": 3.7150923501303375,
      "tokens_seen": 831193088
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037785356068204614,
      "loss": 2.8705,
      "theoretical_loss": 3.7150635149754203,
      "tokens_seen": 831258624
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003778435305917753,
      "loss": 2.9995,
      "theoretical_loss": 3.7150346827302427,
      "tokens_seen": 831324160
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003778335005015045,
      "loss": 2.8506,
      "theoretical_loss": 3.715005853394282,
      "tokens_seen": 831389696
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037782347041123373,
      "loss": 2.9666,
      "theoretical_loss": 3.7149770269670146,
      "tokens_seen": 831455232
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037781344032096286,
      "loss": 2.7702,
      "theoretical_loss": 3.714948203447919,
      "tokens_seen": 831520768
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003778034102306921,
      "loss": 2.6948,
      "theoretical_loss": 3.714919382836471,
      "tokens_seen": 831586304
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003777933801404213,
      "loss": 2.9704,
      "theoretical_loss": 3.7148905651321495,
      "tokens_seen": 831651840
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037778335005015046,
      "loss": 3.2532,
      "theoretical_loss": 3.714861750334432,
      "tokens_seen": 831717376
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037777331995987964,
      "loss": 2.9136,
      "theoretical_loss": 3.714832938442796,
      "tokens_seen": 831782912
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003777632898696088,
      "loss": 2.664,
      "theoretical_loss": 3.7148041294567187,
      "tokens_seen": 831848448
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.000377753259779338,
      "loss": 2.8244,
      "theoretical_loss": 3.7147753233756795,
      "tokens_seen": 831913984
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037774322968906724,
      "loss": 2.5678,
      "theoretical_loss": 3.7147465201991565,
      "tokens_seen": 831979520
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037773319959879636,
      "loss": 2.8719,
      "theoretical_loss": 3.7147177199266275,
      "tokens_seen": 832045056
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003777231695085256,
      "loss": 2.8671,
      "theoretical_loss": 3.7146889225575714,
      "tokens_seen": 832110592
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003777131394182547,
      "loss": 2.9368,
      "theoretical_loss": 3.7146601280914666,
      "tokens_seen": 832176128
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037770310932798396,
      "loss": 3.0479,
      "theoretical_loss": 3.714631336527792,
      "tokens_seen": 832241664
    },
    {
      "epoch": 0.25,
      "objective/train/advantage_avg": 0.494020938873291,
      "objective/train/docs_used": 479893,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.8296380043029785,
      "objective/train/original_loss": 2.829637289047241,
      "objective/train/theoretical_loss": 3.7146025478660274,
      "objective/train/tokens_used": 852767200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24410906434059143,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506430864334106,
      "objective/train/weighted_lm_loss": 2.972784996032715,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.044721245765686,
      "theoretical_loss": 3.7146025478660274,
      "tokens_seen": 832307200
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037769307923771314,
      "loss": 3.0235,
      "theoretical_loss": 3.7146025478660274,
      "tokens_seen": 832307200
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003776830491474423,
      "loss": 2.6617,
      "theoretical_loss": 3.7145737621056503,
      "tokens_seen": 832372736
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003776730190571715,
      "loss": 3.0439,
      "theoretical_loss": 3.7145449792461402,
      "tokens_seen": 832438272
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003776629889669007,
      "loss": 3.0967,
      "theoretical_loss": 3.7145161992869773,
      "tokens_seen": 832503808
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037765295887662987,
      "loss": 2.8531,
      "theoretical_loss": 3.7144874222276405,
      "tokens_seen": 832569344
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003776429287863591,
      "loss": 3.0828,
      "theoretical_loss": 3.71445864806761,
      "tokens_seen": 832634880
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037763289869608823,
      "loss": 2.8337,
      "theoretical_loss": 3.714429876806364,
      "tokens_seen": 832700416
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037762286860581746,
      "loss": 2.9281,
      "theoretical_loss": 3.7144011084433837,
      "tokens_seen": 832765952
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003776128385155467,
      "loss": 2.9518,
      "theoretical_loss": 3.714372342978149,
      "tokens_seen": 832831488
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003776028084252758,
      "loss": 2.8587,
      "theoretical_loss": 3.71434358041014,
      "tokens_seen": 832897024
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037759277833500506,
      "loss": 3.0926,
      "theoretical_loss": 3.714314820738836,
      "tokens_seen": 832962560
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003775827482447342,
      "loss": 2.8433,
      "theoretical_loss": 3.7142860639637187,
      "tokens_seen": 833028096
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003775727181544634,
      "loss": 3.0495,
      "theoretical_loss": 3.7142573100842675,
      "tokens_seen": 833093632
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003775626880641926,
      "loss": 3.0251,
      "theoretical_loss": 3.714228559099964,
      "tokens_seen": 833159168
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003775526579739218,
      "loss": 2.9272,
      "theoretical_loss": 3.714199811010289,
      "tokens_seen": 833224704
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037754262788365097,
      "loss": 2.9278,
      "theoretical_loss": 3.7141710658147224,
      "tokens_seen": 833290240
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037753259779338015,
      "loss": 2.7099,
      "theoretical_loss": 3.714142323512746,
      "tokens_seen": 833355776
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037752256770310933,
      "loss": 2.8828,
      "theoretical_loss": 3.714113584103841,
      "tokens_seen": 833421312
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037751253761283857,
      "loss": 3.0479,
      "theoretical_loss": 3.7140848475874892,
      "tokens_seen": 833486848
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003775025075225677,
      "loss": 2.8762,
      "theoretical_loss": 3.714056113963171,
      "tokens_seen": 833552384
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037749247743229693,
      "loss": 3.0219,
      "theoretical_loss": 3.714027383230369,
      "tokens_seen": 833617920
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037748244734202605,
      "loss": 2.9176,
      "theoretical_loss": 3.713998655388564,
      "tokens_seen": 833683456
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003774724172517553,
      "loss": 3.0127,
      "theoretical_loss": 3.7139699304372384,
      "tokens_seen": 833748992
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037746238716148447,
      "loss": 2.7288,
      "theoretical_loss": 3.713941208375874,
      "tokens_seen": 833814528
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037745235707121365,
      "loss": 2.6843,
      "theoretical_loss": 3.7139124892039534,
      "tokens_seen": 833880064
    },
    {
      "epoch": 0.25,
      "objective/train/advantage_avg": 0.49421626329421997,
      "objective/train/docs_used": 480416,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.9947309494018555,
      "objective/train/original_loss": 2.9947311878204346,
      "objective/train/theoretical_loss": 3.7138837729209584,
      "objective/train/tokens_used": 854405600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24431642889976501,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506635904312134,
      "objective/train/weighted_lm_loss": 3.1466007232666016,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0454468727111816,
      "theoretical_loss": 3.7138837729209584,
      "tokens_seen": 833945600
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037744232698094283,
      "loss": 2.8238,
      "theoretical_loss": 3.7138837729209584,
      "tokens_seen": 833945600
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037743229689067207,
      "loss": 2.8835,
      "theoretical_loss": 3.7138550595263715,
      "tokens_seen": 834011136
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003774222668004012,
      "loss": 2.8678,
      "theoretical_loss": 3.7138263490196755,
      "tokens_seen": 834076672
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037741223671013043,
      "loss": 2.7976,
      "theoretical_loss": 3.713797641400353,
      "tokens_seen": 834142208
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037740220661985956,
      "loss": 2.9535,
      "theoretical_loss": 3.7137689366678863,
      "tokens_seen": 834207744
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003773921765295888,
      "loss": 3.0323,
      "theoretical_loss": 3.7137402348217585,
      "tokens_seen": 834273280
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.000377382146439318,
      "loss": 2.9974,
      "theoretical_loss": 3.713711535861453,
      "tokens_seen": 834338816
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037737211634904715,
      "loss": 2.9654,
      "theoretical_loss": 3.713682839786453,
      "tokens_seen": 834404352
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037736208625877634,
      "loss": 2.9863,
      "theoretical_loss": 3.713654146596242,
      "tokens_seen": 834469888
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003773520561685055,
      "loss": 2.6215,
      "theoretical_loss": 3.713625456290303,
      "tokens_seen": 834535424
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003773420260782347,
      "loss": 2.8719,
      "theoretical_loss": 3.7135967688681193,
      "tokens_seen": 834600960
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037733199598796393,
      "loss": 3.0925,
      "theoretical_loss": 3.713568084329175,
      "tokens_seen": 834666496
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037732196589769306,
      "loss": 2.9069,
      "theoretical_loss": 3.7135394026729553,
      "tokens_seen": 834732032
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003773119358074223,
      "loss": 2.8257,
      "theoretical_loss": 3.713510723898942,
      "tokens_seen": 834797568
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003773019057171515,
      "loss": 2.7502,
      "theoretical_loss": 3.71348204800662,
      "tokens_seen": 834863104
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037729187562688066,
      "loss": 2.9206,
      "theoretical_loss": 3.7134533749954746,
      "tokens_seen": 834928640
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037728184553660984,
      "loss": 2.943,
      "theoretical_loss": 3.713424704864989,
      "tokens_seen": 834994176
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.000377271815446339,
      "loss": 2.8074,
      "theoretical_loss": 3.7133960376146478,
      "tokens_seen": 835059712
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003772617853560682,
      "loss": 2.8739,
      "theoretical_loss": 3.7133673732439365,
      "tokens_seen": 835125248
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037725175526579744,
      "loss": 3.1619,
      "theoretical_loss": 3.713338711752339,
      "tokens_seen": 835190784
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037724172517552656,
      "loss": 2.8866,
      "theoretical_loss": 3.713310053139341,
      "tokens_seen": 835256320
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003772316950852558,
      "loss": 2.83,
      "theoretical_loss": 3.713281397404427,
      "tokens_seen": 835321856
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003772216649949849,
      "loss": 2.9423,
      "theoretical_loss": 3.7132527445470815,
      "tokens_seen": 835387392
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037721163490471416,
      "loss": 2.9056,
      "theoretical_loss": 3.713224094566792,
      "tokens_seen": 835452928
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037720160481444334,
      "loss": 2.82,
      "theoretical_loss": 3.7131954474630424,
      "tokens_seen": 835518464
    },
    {
      "epoch": 0.25,
      "objective/train/advantage_avg": 0.4946676194667816,
      "objective/train/docs_used": 481156,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.883741855621338,
      "objective/train/original_loss": 2.883742332458496,
      "objective/train/theoretical_loss": 3.713166803235318,
      "objective/train/tokens_used": 856044000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24479395151138306,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507110357284546,
      "objective/train/weighted_lm_loss": 3.030078411102295,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0253151655197144,
      "theoretical_loss": 3.713166803235318,
      "tokens_seen": 835584000
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003771915747241725,
      "loss": 2.9098,
      "theoretical_loss": 3.713166803235318,
      "tokens_seen": 835584000
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003771815446339017,
      "loss": 2.8428,
      "theoretical_loss": 3.713138161883105,
      "tokens_seen": 835649536
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003771715145436309,
      "loss": 2.7714,
      "theoretical_loss": 3.7131095234058895,
      "tokens_seen": 835715072
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037716148445336007,
      "loss": 2.9683,
      "theoretical_loss": 3.713080887803158,
      "tokens_seen": 835780608
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003771514543630893,
      "loss": 3.0089,
      "theoretical_loss": 3.7130522550743947,
      "tokens_seen": 835846144
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037714142427281843,
      "loss": 3.0324,
      "theoretical_loss": 3.7130236252190882,
      "tokens_seen": 835911680
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037713139418254766,
      "loss": 2.9559,
      "theoretical_loss": 3.7129949982367236,
      "tokens_seen": 835977216
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037712136409227685,
      "loss": 3.1785,
      "theoretical_loss": 3.712966374126787,
      "tokens_seen": 836042752
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.000377111334002006,
      "loss": 2.9285,
      "theoretical_loss": 3.7129377528887657,
      "tokens_seen": 836108288
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003771013039117352,
      "loss": 2.9597,
      "theoretical_loss": 3.712909134522147,
      "tokens_seen": 836173824
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003770912738214644,
      "loss": 2.803,
      "theoretical_loss": 3.7128805190264167,
      "tokens_seen": 836239360
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037708124373119357,
      "loss": 2.6898,
      "theoretical_loss": 3.7128519064010628,
      "tokens_seen": 836304896
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003770712136409228,
      "loss": 2.82,
      "theoretical_loss": 3.712823296645572,
      "tokens_seen": 836370432
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037706118355065193,
      "loss": 2.8893,
      "theoretical_loss": 3.7127946897594315,
      "tokens_seen": 836435968
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037705115346038117,
      "loss": 2.8663,
      "theoretical_loss": 3.7127660857421287,
      "tokens_seen": 836501504
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003770411233701103,
      "loss": 2.8802,
      "theoretical_loss": 3.712737484593152,
      "tokens_seen": 836567040
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037703109327983953,
      "loss": 2.9865,
      "theoretical_loss": 3.712708886311988,
      "tokens_seen": 836632576
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003770210631895687,
      "loss": 2.5236,
      "theoretical_loss": 3.712680290898125,
      "tokens_seen": 836698112
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003770110330992979,
      "loss": 2.9553,
      "theoretical_loss": 3.712651698351051,
      "tokens_seen": 836763648
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003770010030090271,
      "loss": 2.9209,
      "theoretical_loss": 3.7126231086702544,
      "tokens_seen": 836829184
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037699097291875625,
      "loss": 3.0291,
      "theoretical_loss": 3.7125945218552228,
      "tokens_seen": 836894720
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037698094282848544,
      "loss": 2.8776,
      "theoretical_loss": 3.7125659379054454,
      "tokens_seen": 836960256
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037697091273821467,
      "loss": 3.0879,
      "theoretical_loss": 3.71253735682041,
      "tokens_seen": 837025792
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003769608826479438,
      "loss": 3.0002,
      "theoretical_loss": 3.7125087785996045,
      "tokens_seen": 837091328
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037695085255767303,
      "loss": 3.001,
      "theoretical_loss": 3.712480203242519,
      "tokens_seen": 837156864
    },
    {
      "epoch": 0.25,
      "objective/train/advantage_avg": 0.49484172463417053,
      "objective/train/docs_used": 482509,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.3554136753082275,
      "objective/train/original_loss": 2.3554131984710693,
      "objective/train/theoretical_loss": 3.7124516307486424,
      "objective/train/tokens_used": 857682400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24492113292217255,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507292747497559,
      "objective/train/weighted_lm_loss": 2.474759340286255,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 3.7124516307486424,
      "tokens_seen": 837222400
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003769408224674022,
      "loss": 2.7542,
      "theoretical_loss": 3.7124516307486424,
      "tokens_seen": 837222400
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003769307923771314,
      "loss": 2.8963,
      "theoretical_loss": 3.7124230611174633,
      "tokens_seen": 837287936
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003769207622868606,
      "loss": 2.9282,
      "theoretical_loss": 3.7123944943484704,
      "tokens_seen": 837353472
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037691073219658976,
      "loss": 2.9616,
      "theoretical_loss": 3.7123659304411536,
      "tokens_seen": 837419008
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037690070210631894,
      "loss": 2.7,
      "theoretical_loss": 3.7123373693950024,
      "tokens_seen": 837484544
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003768906720160482,
      "loss": 2.7892,
      "theoretical_loss": 3.712308811209505,
      "tokens_seen": 837550080
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003768806419257773,
      "loss": 2.9038,
      "theoretical_loss": 3.7122802558841532,
      "tokens_seen": 837615616
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037687061183550654,
      "loss": 3.1915,
      "theoretical_loss": 3.7122517034184352,
      "tokens_seen": 837681152
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003768605817452357,
      "loss": 2.7688,
      "theoretical_loss": 3.7122231538118413,
      "tokens_seen": 837746688
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003768505516549649,
      "loss": 2.7716,
      "theoretical_loss": 3.7121946070638625,
      "tokens_seen": 837812224
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037684052156469413,
      "loss": 3.0511,
      "theoretical_loss": 3.7121660631739877,
      "tokens_seen": 837877760
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037683049147442326,
      "loss": 2.986,
      "theoretical_loss": 3.712137522141708,
      "tokens_seen": 837943296
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003768204613841525,
      "loss": 2.9464,
      "theoretical_loss": 3.7121089839665133,
      "tokens_seen": 838008832
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003768104312938817,
      "loss": 3.0877,
      "theoretical_loss": 3.7120804486478947,
      "tokens_seen": 838074368
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037680040120361086,
      "loss": 2.8182,
      "theoretical_loss": 3.7120519161853425,
      "tokens_seen": 838139904
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037679037111334004,
      "loss": 2.8511,
      "theoretical_loss": 3.7120233865783483,
      "tokens_seen": 838205440
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003767803410230692,
      "loss": 3.0049,
      "theoretical_loss": 3.711994859826402,
      "tokens_seen": 838270976
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003767703109327984,
      "loss": 2.803,
      "theoretical_loss": 3.711966335928995,
      "tokens_seen": 838336512
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037676028084252764,
      "loss": 2.9038,
      "theoretical_loss": 3.7119378148856192,
      "tokens_seen": 838402048
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037675025075225676,
      "loss": 2.9161,
      "theoretical_loss": 3.711909296695766,
      "tokens_seen": 838467584
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.000376740220661986,
      "loss": 2.8038,
      "theoretical_loss": 3.7118807813589263,
      "tokens_seen": 838533120
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003767301905717151,
      "loss": 2.904,
      "theoretical_loss": 3.7118522688745914,
      "tokens_seen": 838598656
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037672016048144436,
      "loss": 3.176,
      "theoretical_loss": 3.7118237592422543,
      "tokens_seen": 838664192
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037671013039117354,
      "loss": 3.0653,
      "theoretical_loss": 3.711795252461406,
      "tokens_seen": 838729728
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003767001003009027,
      "loss": 2.8246,
      "theoretical_loss": 3.711766748531538,
      "tokens_seen": 838795264
    },
    {
      "epoch": 0.25,
      "objective/train/advantage_avg": 0.49396568536758423,
      "objective/train/docs_used": 483094,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.8039252758026123,
      "objective/train/original_loss": 2.8039252758026123,
      "objective/train/theoretical_loss": 3.7117382474521436,
      "objective/train/tokens_used": 859320800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24407237768173218,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506373643875122,
      "objective/train/weighted_lm_loss": 2.9460198879241943,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0396549701690674,
      "theoretical_loss": 3.7117382474521436,
      "tokens_seen": 838860800
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003766900702106319,
      "loss": 3.0565,
      "theoretical_loss": 3.7117382474521436,
      "tokens_seen": 838860800
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003766800401203611,
      "loss": 2.838,
      "theoretical_loss": 3.711709749222715,
      "tokens_seen": 838926336
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037667001003009027,
      "loss": 2.7314,
      "theoretical_loss": 3.7116812538427437,
      "tokens_seen": 838991872
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003766599799398195,
      "loss": 2.84,
      "theoretical_loss": 3.7116527613117234,
      "tokens_seen": 839057408
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037664994984954863,
      "loss": 2.7004,
      "theoretical_loss": 3.7116242716291454,
      "tokens_seen": 839122944
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037663991975927786,
      "loss": 2.9742,
      "theoretical_loss": 3.711595784794504,
      "tokens_seen": 839188480
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037662988966900705,
      "loss": 3.1112,
      "theoretical_loss": 3.711567300807291,
      "tokens_seen": 839254016
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003766198595787362,
      "loss": 2.7328,
      "theoretical_loss": 3.7115388196670005,
      "tokens_seen": 839319552
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003766098294884654,
      "loss": 3.0798,
      "theoretical_loss": 3.711510341373124,
      "tokens_seen": 839385088
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003765997993981946,
      "loss": 3.1508,
      "theoretical_loss": 3.711481865925157,
      "tokens_seen": 839450624
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037658976930792377,
      "loss": 2.8228,
      "theoretical_loss": 3.7114533933225915,
      "tokens_seen": 839516160
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.000376579739217653,
      "loss": 2.8997,
      "theoretical_loss": 3.711424923564921,
      "tokens_seen": 839581696
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037656970912738213,
      "loss": 2.9019,
      "theoretical_loss": 3.71139645665164,
      "tokens_seen": 839647232
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037655967903711137,
      "loss": 2.8301,
      "theoretical_loss": 3.7113679925822423,
      "tokens_seen": 839712768
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003765496489468405,
      "loss": 2.6726,
      "theoretical_loss": 3.711339531356221,
      "tokens_seen": 839778304
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037653961885656973,
      "loss": 2.878,
      "theoretical_loss": 3.7113110729730714,
      "tokens_seen": 839843840
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003765295887662989,
      "loss": 2.6885,
      "theoretical_loss": 3.7112826174322864,
      "tokens_seen": 839909376
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003765195586760281,
      "loss": 2.9333,
      "theoretical_loss": 3.7112541647333614,
      "tokens_seen": 839974912
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003765095285857573,
      "loss": 2.9703,
      "theoretical_loss": 3.7112257148757903,
      "tokens_seen": 840040448
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037649949849548645,
      "loss": 2.8329,
      "theoretical_loss": 3.7111972678590686,
      "tokens_seen": 840105984
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037648946840521564,
      "loss": 2.9888,
      "theoretical_loss": 3.71116882368269,
      "tokens_seen": 840171520
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037647943831494487,
      "loss": 2.9227,
      "theoretical_loss": 3.7111403823461497,
      "tokens_seen": 840237056
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.000376469408224674,
      "loss": 3.1217,
      "theoretical_loss": 3.711111943848943,
      "tokens_seen": 840302592
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037645937813440323,
      "loss": 3.0255,
      "theoretical_loss": 3.7110835081905647,
      "tokens_seen": 840368128
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003764493480441324,
      "loss": 2.8702,
      "theoretical_loss": 3.7110550753705103,
      "tokens_seen": 840433664
    },
    {
      "epoch": 0.25,
      "objective/train/advantage_avg": 0.49393555521965027,
      "objective/train/docs_used": 483926,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.04071044921875,
      "objective/train/original_loss": 3.040710926055908,
      "objective/train/theoretical_loss": 3.711026645388275,
      "objective/train/tokens_used": 860959200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24403563141822815,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506340265274048,
      "objective/train/weighted_lm_loss": 3.194624662399292,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0425469875335693,
      "theoretical_loss": 3.711026645388275,
      "tokens_seen": 840499200
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003764393179538616,
      "loss": 3.0236,
      "theoretical_loss": 3.711026645388275,
      "tokens_seen": 840499200
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003764292878635908,
      "loss": 3.2929,
      "theoretical_loss": 3.710998218243355,
      "tokens_seen": 840564736
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037641925777331996,
      "loss": 2.8829,
      "theoretical_loss": 3.7109697939352446,
      "tokens_seen": 840630272
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037640922768304914,
      "loss": 2.9493,
      "theoretical_loss": 3.7109413724634406,
      "tokens_seen": 840695808
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003763991975927784,
      "loss": 2.8531,
      "theoretical_loss": 3.710912953827439,
      "tokens_seen": 840761344
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003763891675025075,
      "loss": 2.8361,
      "theoretical_loss": 3.7108845380267352,
      "tokens_seen": 840826880
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037637913741223674,
      "loss": 2.9836,
      "theoretical_loss": 3.7108561250608254,
      "tokens_seen": 840892416
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037636910732196586,
      "loss": 2.8196,
      "theoretical_loss": 3.7108277149292066,
      "tokens_seen": 840957952
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003763590772316951,
      "loss": 3.1502,
      "theoretical_loss": 3.7107993076313748,
      "tokens_seen": 841023488
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003763490471414243,
      "loss": 2.7763,
      "theoretical_loss": 3.710770903166827,
      "tokens_seen": 841089024
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037633901705115346,
      "loss": 2.9051,
      "theoretical_loss": 3.7107425015350586,
      "tokens_seen": 841154560
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037632898696088264,
      "loss": 2.9023,
      "theoretical_loss": 3.710714102735568,
      "tokens_seen": 841220096
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.0003763189568706119,
      "loss": 2.6935,
      "theoretical_loss": 3.7106857067678503,
      "tokens_seen": 841285632
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.000376308926780341,
      "loss": 3.018,
      "theoretical_loss": 3.7106573136314047,
      "tokens_seen": 841351168
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037629889669007024,
      "loss": 3.1643,
      "theoretical_loss": 3.710628923325727,
      "tokens_seen": 841416704
    },
    {
      "epoch": 0.25,
      "learning_rate": 0.00037628886659979937,
      "loss": 2.9714,
      "theoretical_loss": 3.7106005358503147,
      "tokens_seen": 841482240
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003762788365095286,
      "loss": 2.9255,
      "theoretical_loss": 3.7105721512046657,
      "tokens_seen": 841547776
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003762688064192578,
      "loss": 2.8442,
      "theoretical_loss": 3.7105437693882775,
      "tokens_seen": 841613312
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037625877632898696,
      "loss": 2.9415,
      "theoretical_loss": 3.7105153904006474,
      "tokens_seen": 841678848
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037624874623871615,
      "loss": 3.0611,
      "theoretical_loss": 3.7104870142412736,
      "tokens_seen": 841744384
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003762387161484453,
      "loss": 3.1258,
      "theoretical_loss": 3.710458640909654,
      "tokens_seen": 841809920
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003762286860581745,
      "loss": 3.2346,
      "theoretical_loss": 3.7104302704052863,
      "tokens_seen": 841875456
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037621865596790374,
      "loss": 3.0993,
      "theoretical_loss": 3.7104019027276696,
      "tokens_seen": 841940992
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037620862587763287,
      "loss": 3.0138,
      "theoretical_loss": 3.7103735378763014,
      "tokens_seen": 842006528
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003761985957873621,
      "loss": 2.8194,
      "theoretical_loss": 3.71034517585068,
      "tokens_seen": 842072064
    },
    {
      "epoch": 0.26,
      "objective/train/advantage_avg": 0.49468711018562317,
      "objective/train/docs_used": 484283,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.724545478820801,
      "objective/train/original_loss": 2.724545955657959,
      "objective/train/theoretical_loss": 3.7103168166503053,
      "objective/train/tokens_used": 862597600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24476630985736847,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507129430770874,
      "objective/train/weighted_lm_loss": 2.862663507461548,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0425469875335693,
      "theoretical_loss": 3.7103168166503053,
      "tokens_seen": 842137600
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037618856569709123,
      "loss": 2.9302,
      "theoretical_loss": 3.7103168166503053,
      "tokens_seen": 842137600
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037617853560682047,
      "loss": 2.9006,
      "theoretical_loss": 3.7102884602746746,
      "tokens_seen": 842203136
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037616850551654965,
      "loss": 3.0129,
      "theoretical_loss": 3.710260106723288,
      "tokens_seen": 842268672
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037615847542627883,
      "loss": 3.082,
      "theoretical_loss": 3.7102317559956437,
      "tokens_seen": 842334208
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.000376148445336008,
      "loss": 3.1064,
      "theoretical_loss": 3.7102034080912407,
      "tokens_seen": 842399744
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037613841524573725,
      "loss": 2.7722,
      "theoretical_loss": 3.7101750630095793,
      "tokens_seen": 842465280
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037612838515546637,
      "loss": 3.0486,
      "theoretical_loss": 3.7101467207501573,
      "tokens_seen": 842530816
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003761183550651956,
      "loss": 2.7679,
      "theoretical_loss": 3.7101183813124754,
      "tokens_seen": 842596352
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003761083249749248,
      "loss": 3.1712,
      "theoretical_loss": 3.7100900446960328,
      "tokens_seen": 842661888
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037609829488465397,
      "loss": 2.8512,
      "theoretical_loss": 3.7100617109003293,
      "tokens_seen": 842727424
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003760882647943832,
      "loss": 2.819,
      "theoretical_loss": 3.710033379924865,
      "tokens_seen": 842792960
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037607823470411233,
      "loss": 3.1928,
      "theoretical_loss": 3.7100050517691403,
      "tokens_seen": 842858496
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037606820461384157,
      "loss": 2.9057,
      "theoretical_loss": 3.7099767264326537,
      "tokens_seen": 842924032
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003760581745235707,
      "loss": 2.9203,
      "theoretical_loss": 3.7099484039149075,
      "tokens_seen": 842989568
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037604814443329993,
      "loss": 2.9906,
      "theoretical_loss": 3.7099200842154003,
      "tokens_seen": 843055104
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003760381143430291,
      "loss": 2.7516,
      "theoretical_loss": 3.709891767333634,
      "tokens_seen": 843120640
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003760280842527583,
      "loss": 3.0991,
      "theoretical_loss": 3.709863453269109,
      "tokens_seen": 843186176
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003760180541624875,
      "loss": 2.8538,
      "theoretical_loss": 3.709835142021325,
      "tokens_seen": 843251712
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037600802407221665,
      "loss": 2.8192,
      "theoretical_loss": 3.7098068335897842,
      "tokens_seen": 843317248
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037599799398194584,
      "loss": 2.9637,
      "theoretical_loss": 3.7097785279739868,
      "tokens_seen": 843382784
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037598796389167507,
      "loss": 2.6801,
      "theoretical_loss": 3.7097502251734342,
      "tokens_seen": 843448320
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003759779338014042,
      "loss": 2.7647,
      "theoretical_loss": 3.7097219251876283,
      "tokens_seen": 843513856
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037596790371113343,
      "loss": 3.0895,
      "theoretical_loss": 3.7096936280160695,
      "tokens_seen": 843579392
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003759578736208626,
      "loss": 2.6648,
      "theoretical_loss": 3.7096653336582603,
      "tokens_seen": 843644928
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003759478435305918,
      "loss": 2.9967,
      "theoretical_loss": 3.7096370421137017,
      "tokens_seen": 843710464
    },
    {
      "epoch": 0.26,
      "objective/train/advantage_avg": 0.49458232522010803,
      "objective/train/docs_used": 485567,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.7407164573669434,
      "objective/train/original_loss": 2.740715980529785,
      "objective/train/theoretical_loss": 3.7096087533818953,
      "objective/train/tokens_used": 864236000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24470923840999603,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507022142410278,
      "objective/train/weighted_lm_loss": 2.879488468170166,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0363596677780151,
      "theoretical_loss": 3.7096087533818953,
      "tokens_seen": 843776000
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.000375937813440321,
      "loss": 2.8802,
      "theoretical_loss": 3.7096087533818953,
      "tokens_seen": 843776000
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037592778335005016,
      "loss": 3.0673,
      "theoretical_loss": 3.7095804674623443,
      "tokens_seen": 843841536
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037591775325977934,
      "loss": 2.7243,
      "theoretical_loss": 3.709552184354549,
      "tokens_seen": 843907072
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003759077231695086,
      "loss": 2.6681,
      "theoretical_loss": 3.709523904058013,
      "tokens_seen": 843972608
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003758976930792377,
      "loss": 2.8624,
      "theoretical_loss": 3.709495626572237,
      "tokens_seen": 844038144
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037588766298896694,
      "loss": 2.9648,
      "theoretical_loss": 3.709467351896726,
      "tokens_seen": 844103680
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037587763289869606,
      "loss": 2.9445,
      "theoretical_loss": 3.70943908003098,
      "tokens_seen": 844169216
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003758676028084253,
      "loss": 3.0815,
      "theoretical_loss": 3.709410810974503,
      "tokens_seen": 844234752
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003758575727181545,
      "loss": 2.8115,
      "theoretical_loss": 3.7093825447267976,
      "tokens_seen": 844300288
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037584754262788366,
      "loss": 2.7986,
      "theoretical_loss": 3.709354281287367,
      "tokens_seen": 844365824
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037583751253761284,
      "loss": 2.9801,
      "theoretical_loss": 3.7093260206557135,
      "tokens_seen": 844431360
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003758274824473421,
      "loss": 3.0657,
      "theoretical_loss": 3.7092977628313406,
      "tokens_seen": 844496896
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003758174523570712,
      "loss": 2.917,
      "theoretical_loss": 3.709269507813752,
      "tokens_seen": 844562432
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037580742226680044,
      "loss": 2.8646,
      "theoretical_loss": 3.7092412556024503,
      "tokens_seen": 844627968
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037579739217652957,
      "loss": 2.9136,
      "theoretical_loss": 3.70921300619694,
      "tokens_seen": 844693504
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003757873620862588,
      "loss": 3.0433,
      "theoretical_loss": 3.7091847595967247,
      "tokens_seen": 844759040
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.000375777331995988,
      "loss": 3.0332,
      "theoretical_loss": 3.7091565158013076,
      "tokens_seen": 844824576
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037576730190571716,
      "loss": 2.944,
      "theoretical_loss": 3.709128274810193,
      "tokens_seen": 844890112
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037575727181544635,
      "loss": 2.6729,
      "theoretical_loss": 3.709100036622885,
      "tokens_seen": 844955648
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003757472417251755,
      "loss": 3.0478,
      "theoretical_loss": 3.709071801238888,
      "tokens_seen": 845021184
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003757372116349047,
      "loss": 2.7408,
      "theoretical_loss": 3.709043568657706,
      "tokens_seen": 845086720
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037572718154463394,
      "loss": 2.7847,
      "theoretical_loss": 3.709015338878843,
      "tokens_seen": 845152256
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037571715145436307,
      "loss": 2.9523,
      "theoretical_loss": 3.7089871119018047,
      "tokens_seen": 845217792
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003757071213640923,
      "loss": 2.7073,
      "theoretical_loss": 3.7089588877260953,
      "tokens_seen": 845283328
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037569709127382143,
      "loss": 2.8822,
      "theoretical_loss": 3.708930666351219,
      "tokens_seen": 845348864
    },
    {
      "epoch": 0.26,
      "objective/train/advantage_avg": 0.49386337399482727,
      "objective/train/docs_used": 486205,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.2155978679656982,
      "objective/train/original_loss": 2.2155978679656982,
      "objective/train/theoretical_loss": 3.7089024477766817,
      "objective/train/tokens_used": 865874400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24402761459350586,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506268739700317,
      "objective/train/weighted_lm_loss": 2.3278040885925293,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0363596677780151,
      "theoretical_loss": 3.7089024477766817,
      "tokens_seen": 845414400
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037568706118355067,
      "loss": 3.0588,
      "theoretical_loss": 3.7089024477766817,
      "tokens_seen": 845414400
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037567703109327985,
      "loss": 2.9972,
      "theoretical_loss": 3.708874232001988,
      "tokens_seen": 845479936
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037566700100300903,
      "loss": 2.6786,
      "theoretical_loss": 3.7088460190266437,
      "tokens_seen": 845545472
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003756569709127382,
      "loss": 2.9946,
      "theoretical_loss": 3.708817808850153,
      "tokens_seen": 845611008
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037564694082246745,
      "loss": 2.8482,
      "theoretical_loss": 3.708789601472022,
      "tokens_seen": 845676544
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003756369107321966,
      "loss": 2.757,
      "theoretical_loss": 3.7087613968917563,
      "tokens_seen": 845742080
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003756268806419258,
      "loss": 2.8117,
      "theoretical_loss": 3.7087331951088616,
      "tokens_seen": 845807616
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037561685055165494,
      "loss": 2.8344,
      "theoretical_loss": 3.7087049961228438,
      "tokens_seen": 845873152
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037560682046138417,
      "loss": 2.8823,
      "theoretical_loss": 3.7086767999332086,
      "tokens_seen": 845938688
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037559679037111335,
      "loss": 2.9328,
      "theoretical_loss": 3.7086486065394624,
      "tokens_seen": 846004224
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037558676028084253,
      "loss": 2.8996,
      "theoretical_loss": 3.7086204159411116,
      "tokens_seen": 846069760
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003755767301905717,
      "loss": 3.0617,
      "theoretical_loss": 3.708592228137662,
      "tokens_seen": 846135296
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003755667001003009,
      "loss": 2.9462,
      "theoretical_loss": 3.7085640431286198,
      "tokens_seen": 846200832
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003755566700100301,
      "loss": 2.8768,
      "theoretical_loss": 3.708535860913493,
      "tokens_seen": 846266368
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003755466399197593,
      "loss": 2.9675,
      "theoretical_loss": 3.7085076814917866,
      "tokens_seen": 846331904
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037553660982948844,
      "loss": 2.7005,
      "theoretical_loss": 3.708479504863008,
      "tokens_seen": 846397440
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003755265797392177,
      "loss": 2.6741,
      "theoretical_loss": 3.708451331026665,
      "tokens_seen": 846462976
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003755165496489468,
      "loss": 2.9243,
      "theoretical_loss": 3.7084231599822637,
      "tokens_seen": 846528512
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037550651955867604,
      "loss": 2.9765,
      "theoretical_loss": 3.7083949917293118,
      "tokens_seen": 846594048
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003754964894684052,
      "loss": 3.1347,
      "theoretical_loss": 3.7083668262673166,
      "tokens_seen": 846659584
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003754864593781344,
      "loss": 2.9014,
      "theoretical_loss": 3.7083386635957853,
      "tokens_seen": 846725120
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003754764292878636,
      "loss": 2.9863,
      "theoretical_loss": 3.708310503714226,
      "tokens_seen": 846790656
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003754663991975928,
      "loss": 2.8624,
      "theoretical_loss": 3.7082823466221457,
      "tokens_seen": 846856192
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037545636910732194,
      "loss": 2.8989,
      "theoretical_loss": 3.708254192319053,
      "tokens_seen": 846921728
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003754463390170512,
      "loss": 2.925,
      "theoretical_loss": 3.7082260408044547,
      "tokens_seen": 846987264
    },
    {
      "epoch": 0.26,
      "objective/train/advantage_avg": 0.49368175864219666,
      "objective/train/docs_used": 487507,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.1468493938446045,
      "objective/train/original_loss": 3.1468496322631836,
      "objective/train/theoretical_loss": 3.7081978920778607,
      "objective/train/tokens_used": 867512800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2437974363565445,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050607442855835,
      "objective/train/weighted_lm_loss": 3.306027412414551,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0437887907028198,
      "theoretical_loss": 3.7081978920778607,
      "tokens_seen": 847052800
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003754363089267803,
      "loss": 3.0144,
      "theoretical_loss": 3.7081978920778607,
      "tokens_seen": 847052800
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037542627883650954,
      "loss": 2.7791,
      "theoretical_loss": 3.7081697461387777,
      "tokens_seen": 847118336
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003754162487462387,
      "loss": 2.8228,
      "theoretical_loss": 3.7081416029867147,
      "tokens_seen": 847183872
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003754062186559679,
      "loss": 3.1474,
      "theoretical_loss": 3.7081134626211796,
      "tokens_seen": 847249408
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003753961885656971,
      "loss": 3.0281,
      "theoretical_loss": 3.708085325041682,
      "tokens_seen": 847314944
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037538615847542626,
      "loss": 2.9705,
      "theoretical_loss": 3.7080571902477297,
      "tokens_seen": 847380480
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037537612838515544,
      "loss": 2.8172,
      "theoretical_loss": 3.7080290582388318,
      "tokens_seen": 847446016
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003753660982948847,
      "loss": 2.7872,
      "theoretical_loss": 3.7080009290144975,
      "tokens_seen": 847511552
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037535606820461386,
      "loss": 3.0256,
      "theoretical_loss": 3.7079728025742353,
      "tokens_seen": 847577088
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037534603811434304,
      "loss": 2.815,
      "theoretical_loss": 3.707944678917555,
      "tokens_seen": 847642624
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003753360080240723,
      "loss": 3.2116,
      "theoretical_loss": 3.7079165580439657,
      "tokens_seen": 847708160
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003753259779338014,
      "loss": 2.8839,
      "theoretical_loss": 3.707888439952977,
      "tokens_seen": 847773696
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037531594784353064,
      "loss": 2.7464,
      "theoretical_loss": 3.7078603246440984,
      "tokens_seen": 847839232
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037530591775325977,
      "loss": 3.0312,
      "theoretical_loss": 3.7078322121168394,
      "tokens_seen": 847904768
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.000375295887662989,
      "loss": 3.0956,
      "theoretical_loss": 3.70780410237071,
      "tokens_seen": 847970304
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003752858575727182,
      "loss": 3.0475,
      "theoretical_loss": 3.7077759954052203,
      "tokens_seen": 848035840
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037527582748244736,
      "loss": 2.9733,
      "theoretical_loss": 3.7077478912198805,
      "tokens_seen": 848101376
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037526579739217655,
      "loss": 2.8319,
      "theoretical_loss": 3.7077197898142,
      "tokens_seen": 848166912
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003752557673019057,
      "loss": 2.9112,
      "theoretical_loss": 3.70769169118769,
      "tokens_seen": 848232448
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003752457372116349,
      "loss": 2.8249,
      "theoretical_loss": 3.7076635953398607,
      "tokens_seen": 848297984
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037523570712136414,
      "loss": 2.8491,
      "theoretical_loss": 3.7076355022702225,
      "tokens_seen": 848363520
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037522567703109327,
      "loss": 2.8108,
      "theoretical_loss": 3.7076074119782865,
      "tokens_seen": 848429056
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003752156469408225,
      "loss": 2.9122,
      "theoretical_loss": 3.7075793244635626,
      "tokens_seen": 848494592
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037520561685055163,
      "loss": 3.0862,
      "theoretical_loss": 3.707551239725563,
      "tokens_seen": 848560128
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037519558676028087,
      "loss": 2.8977,
      "theoretical_loss": 3.707523157763798,
      "tokens_seen": 848625664
    },
    {
      "epoch": 0.26,
      "objective/train/advantage_avg": 0.49502864480018616,
      "objective/train/docs_used": 487917,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.808061122894287,
      "objective/train/original_loss": 2.808060646057129,
      "objective/train/theoretical_loss": 3.7074950785777787,
      "objective/train/tokens_used": 869151200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24511566758155823,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507489442825317,
      "objective/train/weighted_lm_loss": 2.950578212738037,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0253151655197144,
      "theoretical_loss": 3.7074950785777787,
      "tokens_seen": 848691200
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037518555667001005,
      "loss": 2.782,
      "theoretical_loss": 3.7074950785777787,
      "tokens_seen": 848691200
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037517552657973923,
      "loss": 3.0881,
      "theoretical_loss": 3.707467002167017,
      "tokens_seen": 848756736
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003751654964894684,
      "loss": 2.7193,
      "theoretical_loss": 3.7074389285310243,
      "tokens_seen": 848822272
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037515546639919765,
      "loss": 2.9112,
      "theoretical_loss": 3.707410857669312,
      "tokens_seen": 848887808
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003751454363089268,
      "loss": 2.8787,
      "theoretical_loss": 3.707382789581391,
      "tokens_seen": 848953344
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.000375135406218656,
      "loss": 3.0886,
      "theoretical_loss": 3.707354724266774,
      "tokens_seen": 849018880
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037512537612838514,
      "loss": 2.758,
      "theoretical_loss": 3.707326661724973,
      "tokens_seen": 849084416
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037511534603811437,
      "loss": 3.037,
      "theoretical_loss": 3.7072986019554994,
      "tokens_seen": 849149952
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037510531594784355,
      "loss": 3.1176,
      "theoretical_loss": 3.7072705449578667,
      "tokens_seen": 849215488
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037509528585757273,
      "loss": 3.0098,
      "theoretical_loss": 3.7072424907315855,
      "tokens_seen": 849281024
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003750852557673019,
      "loss": 2.7101,
      "theoretical_loss": 3.7072144392761697,
      "tokens_seen": 849346560
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003750752256770311,
      "loss": 2.9055,
      "theoretical_loss": 3.70718639059113,
      "tokens_seen": 849412096
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003750651955867603,
      "loss": 2.8692,
      "theoretical_loss": 3.7071583446759817,
      "tokens_seen": 849477632
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003750551654964895,
      "loss": 2.852,
      "theoretical_loss": 3.7071303015302357,
      "tokens_seen": 849543168
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037504513540621864,
      "loss": 2.9606,
      "theoretical_loss": 3.707102261153405,
      "tokens_seen": 849608704
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003750351053159479,
      "loss": 2.8559,
      "theoretical_loss": 3.7070742235450034,
      "tokens_seen": 849674240
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.000375025075225677,
      "loss": 2.8538,
      "theoretical_loss": 3.7070461887045436,
      "tokens_seen": 849739776
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037501504513540624,
      "loss": 3.0216,
      "theoretical_loss": 3.7070181566315394,
      "tokens_seen": 849805312
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003750050150451354,
      "loss": 3.1769,
      "theoretical_loss": 3.7069901273255037,
      "tokens_seen": 849870848
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003749949849548646,
      "loss": 3.014,
      "theoretical_loss": 3.70696210078595,
      "tokens_seen": 849936384
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003749849548645938,
      "loss": 3.0517,
      "theoretical_loss": 3.7069340770123924,
      "tokens_seen": 850001920
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.000374974924774323,
      "loss": 3.0514,
      "theoretical_loss": 3.7069060560043443,
      "tokens_seen": 850067456
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037496489468405214,
      "loss": 3.2051,
      "theoretical_loss": 3.70687803776132,
      "tokens_seen": 850132992
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003749548645937814,
      "loss": 2.7869,
      "theoretical_loss": 3.7068500222828327,
      "tokens_seen": 850198528
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003749448345035105,
      "loss": 2.7493,
      "theoretical_loss": 3.706822009568398,
      "tokens_seen": 850264064
    },
    {
      "epoch": 0.26,
      "objective/train/advantage_avg": 0.49221646785736084,
      "objective/train/docs_used": 489068,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.796395778656006,
      "objective/train/original_loss": 2.7963953018188477,
      "objective/train/theoretical_loss": 3.706793999617529,
      "objective/train/tokens_used": 870789600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24240517616271973,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050453782081604,
      "objective/train/weighted_lm_loss": 2.93764328956604,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0417574644088745,
      "theoretical_loss": 3.706793999617529,
      "tokens_seen": 850329600
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037493480441323974,
      "loss": 2.9294,
      "theoretical_loss": 3.706793999617529,
      "tokens_seen": 850329600
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003749247743229689,
      "loss": 3.1263,
      "theoretical_loss": 3.70676599242974,
      "tokens_seen": 850395136
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003749147442326981,
      "loss": 3.206,
      "theoretical_loss": 3.7067379880045466,
      "tokens_seen": 850460672
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003749047141424273,
      "loss": 3.169,
      "theoretical_loss": 3.7067099863414628,
      "tokens_seen": 850526208
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037489468405215646,
      "loss": 2.8577,
      "theoretical_loss": 3.7066819874400028,
      "tokens_seen": 850591744
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037488465396188564,
      "loss": 3.0168,
      "theoretical_loss": 3.7066539912996825,
      "tokens_seen": 850657280
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003748746238716149,
      "loss": 2.8854,
      "theoretical_loss": 3.7066259979200167,
      "tokens_seen": 850722816
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.000374864593781344,
      "loss": 3.0237,
      "theoretical_loss": 3.70659800730052,
      "tokens_seen": 850788352
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037485456369107324,
      "loss": 2.947,
      "theoretical_loss": 3.706570019440708,
      "tokens_seen": 850853888
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037484453360080237,
      "loss": 2.805,
      "theoretical_loss": 3.706542034340096,
      "tokens_seen": 850919424
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003748345035105316,
      "loss": 2.8728,
      "theoretical_loss": 3.7065140519982,
      "tokens_seen": 850984960
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003748244734202608,
      "loss": 2.8677,
      "theoretical_loss": 3.7064860724145348,
      "tokens_seen": 851050496
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037481444332998997,
      "loss": 3.0213,
      "theoretical_loss": 3.7064580955886166,
      "tokens_seen": 851116032
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037480441323971915,
      "loss": 3.1958,
      "theoretical_loss": 3.7064301215199613,
      "tokens_seen": 851181568
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003747943831494484,
      "loss": 2.9393,
      "theoretical_loss": 3.7064021502080853,
      "tokens_seen": 851247104
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003747843530591775,
      "loss": 2.9785,
      "theoretical_loss": 3.706374181652504,
      "tokens_seen": 851312640
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037477432296890675,
      "loss": 3.1691,
      "theoretical_loss": 3.706346215852734,
      "tokens_seen": 851378176
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037476429287863587,
      "loss": 2.8675,
      "theoretical_loss": 3.706318252808291,
      "tokens_seen": 851443712
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003747542627883651,
      "loss": 2.9329,
      "theoretical_loss": 3.706290292518692,
      "tokens_seen": 851509248
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003747442326980943,
      "loss": 2.8376,
      "theoretical_loss": 3.706262334983454,
      "tokens_seen": 851574784
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037473420260782347,
      "loss": 2.8226,
      "theoretical_loss": 3.7062343802020936,
      "tokens_seen": 851640320
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037472417251755265,
      "loss": 2.9974,
      "theoretical_loss": 3.706206428174127,
      "tokens_seen": 851705856
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037471414242728183,
      "loss": 2.9277,
      "theoretical_loss": 3.706178478899072,
      "tokens_seen": 851771392
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.000374704112337011,
      "loss": 2.9038,
      "theoretical_loss": 3.706150532376445,
      "tokens_seen": 851836928
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037469408224674025,
      "loss": 2.8696,
      "theoretical_loss": 3.706122588605764,
      "tokens_seen": 851902464
    },
    {
      "epoch": 0.26,
      "objective/train/advantage_avg": 0.4944770932197571,
      "objective/train/docs_used": 489617,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.6633026599884033,
      "objective/train/original_loss": 2.6633026599884033,
      "objective/train/theoretical_loss": 3.7060946475865455,
      "objective/train/tokens_used": 872428000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24457871913909912,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.05069100856781,
      "objective/train/weighted_lm_loss": 2.798090696334839,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0363596677780151,
      "theoretical_loss": 3.7060946475865455,
      "tokens_seen": 851968000
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003746840521564694,
      "loss": 2.9921,
      "theoretical_loss": 3.7060946475865455,
      "tokens_seen": 851968000
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003746740220661986,
      "loss": 2.9001,
      "theoretical_loss": 3.7060667093183075,
      "tokens_seen": 852033536
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037466399197592774,
      "loss": 3.1679,
      "theoretical_loss": 3.706038773800567,
      "tokens_seen": 852099072
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.000374653961885657,
      "loss": 2.9057,
      "theoretical_loss": 3.7060108410328425,
      "tokens_seen": 852164608
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037464393179538615,
      "loss": 2.907,
      "theoretical_loss": 3.705982911014651,
      "tokens_seen": 852230144
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037463390170511534,
      "loss": 2.9213,
      "theoretical_loss": 3.705954983745511,
      "tokens_seen": 852295680
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037462387161484457,
      "loss": 3.1564,
      "theoretical_loss": 3.7059270592249405,
      "tokens_seen": 852361216
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037461384152457375,
      "loss": 2.8622,
      "theoretical_loss": 3.705899137452458,
      "tokens_seen": 852426752
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037460381143430293,
      "loss": 2.9204,
      "theoretical_loss": 3.705871218427581,
      "tokens_seen": 852492288
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003745937813440321,
      "loss": 2.949,
      "theoretical_loss": 3.7058433021498285,
      "tokens_seen": 852557824
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003745837512537613,
      "loss": 2.9441,
      "theoretical_loss": 3.705815388618719,
      "tokens_seen": 852623360
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003745737211634905,
      "loss": 2.8543,
      "theoretical_loss": 3.7057874778337707,
      "tokens_seen": 852688896
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003745636910732197,
      "loss": 2.9345,
      "theoretical_loss": 3.7057595697945027,
      "tokens_seen": 852754432
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037455366098294884,
      "loss": 2.8742,
      "theoretical_loss": 3.7057316645004343,
      "tokens_seen": 852819968
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003745436308926781,
      "loss": 2.9037,
      "theoretical_loss": 3.705703761951084,
      "tokens_seen": 852885504
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003745336008024072,
      "loss": 2.8441,
      "theoretical_loss": 3.7056758621459718,
      "tokens_seen": 852951040
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037452357071213644,
      "loss": 3.0092,
      "theoretical_loss": 3.705647965084616,
      "tokens_seen": 853016576
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003745135406218656,
      "loss": 2.8956,
      "theoretical_loss": 3.7056200707665363,
      "tokens_seen": 853082112
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003745035105315948,
      "loss": 3.0283,
      "theoretical_loss": 3.7055921791912527,
      "tokens_seen": 853147648
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.000374493480441324,
      "loss": 2.8642,
      "theoretical_loss": 3.705564290358284,
      "tokens_seen": 853213184
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003744834503510532,
      "loss": 2.5733,
      "theoretical_loss": 3.705536404267151,
      "tokens_seen": 853278720
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037447342026078234,
      "loss": 3.1862,
      "theoretical_loss": 3.7055085209173724,
      "tokens_seen": 853344256
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003744633901705116,
      "loss": 3.0895,
      "theoretical_loss": 3.705480640308469,
      "tokens_seen": 853409792
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003744533600802407,
      "loss": 2.8902,
      "theoretical_loss": 3.705452762439961,
      "tokens_seen": 853475328
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037444332998996994,
      "loss": 3.0234,
      "theoretical_loss": 3.705424887311368,
      "tokens_seen": 853540864
    },
    {
      "epoch": 0.26,
      "objective/train/advantage_avg": 0.49216845631599426,
      "objective/train/docs_used": 491051,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.0019540786743164,
      "objective/train/original_loss": 3.0019543170928955,
      "objective/train/theoretical_loss": 3.705397014922211,
      "objective/train/tokens_used": 874066400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24230816960334778,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0504485368728638,
      "objective/train/weighted_lm_loss": 3.1532301902770996,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.044721245765686,
      "theoretical_loss": 3.705397014922211,
      "tokens_seen": 853606400
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003744332998996991,
      "loss": 3.3044,
      "theoretical_loss": 3.705397014922211,
      "tokens_seen": 853606400
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003744232698094283,
      "loss": 2.7832,
      "theoretical_loss": 3.7053691452720106,
      "tokens_seen": 853671936
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003744132397191575,
      "loss": 2.9024,
      "theoretical_loss": 3.7053412783602866,
      "tokens_seen": 853737472
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037440320962888666,
      "loss": 2.7671,
      "theoretical_loss": 3.70531341418656,
      "tokens_seen": 853803008
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037439317953861585,
      "loss": 3.0227,
      "theoretical_loss": 3.705285552750352,
      "tokens_seen": 853868544
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003743831494483451,
      "loss": 3.0813,
      "theoretical_loss": 3.7052576940511837,
      "tokens_seen": 853934080
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003743731193580742,
      "loss": 2.7913,
      "theoretical_loss": 3.7052298380885755,
      "tokens_seen": 853999616
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037436308926780344,
      "loss": 3.0498,
      "theoretical_loss": 3.7052019848620494,
      "tokens_seen": 854065152
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037435305917753257,
      "loss": 2.8458,
      "theoretical_loss": 3.7051741343711266,
      "tokens_seen": 854130688
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003743430290872618,
      "loss": 2.9417,
      "theoretical_loss": 3.705146286615328,
      "tokens_seen": 854196224
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.000374332998996991,
      "loss": 2.7316,
      "theoretical_loss": 3.705118441594175,
      "tokens_seen": 854261760
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037432296890672017,
      "loss": 3.2234,
      "theoretical_loss": 3.7050905993071908,
      "tokens_seen": 854327296
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037431293881644935,
      "loss": 2.8395,
      "theoretical_loss": 3.7050627597538957,
      "tokens_seen": 854392832
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003743029087261786,
      "loss": 2.9448,
      "theoretical_loss": 3.705034922933812,
      "tokens_seen": 854458368
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003742928786359077,
      "loss": 3.0606,
      "theoretical_loss": 3.705007088846462,
      "tokens_seen": 854523904
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037428284854563695,
      "loss": 3.0906,
      "theoretical_loss": 3.704979257491368,
      "tokens_seen": 854589440
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037427281845536607,
      "loss": 3.0973,
      "theoretical_loss": 3.7049514288680516,
      "tokens_seen": 854654976
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003742627883650953,
      "loss": 3.0314,
      "theoretical_loss": 3.704923602976036,
      "tokens_seen": 854720512
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003742527582748245,
      "loss": 2.9371,
      "theoretical_loss": 3.704895779814844,
      "tokens_seen": 854786048
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037424272818455367,
      "loss": 2.8788,
      "theoretical_loss": 3.704867959383997,
      "tokens_seen": 854851584
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037423269809428285,
      "loss": 3.1026,
      "theoretical_loss": 3.7048401416830186,
      "tokens_seen": 854917120
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037422266800401203,
      "loss": 3.1276,
      "theoretical_loss": 3.7048123267114317,
      "tokens_seen": 854982656
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003742126379137412,
      "loss": 3.0025,
      "theoretical_loss": 3.7047845144687592,
      "tokens_seen": 855048192
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037420260782347045,
      "loss": 3.0406,
      "theoretical_loss": 3.704756704954524,
      "tokens_seen": 855113728
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003741925777331996,
      "loss": 2.7932,
      "theoretical_loss": 3.7047288981682502,
      "tokens_seen": 855179264
    },
    {
      "epoch": 0.26,
      "objective/train/advantage_avg": 0.4930228590965271,
      "objective/train/docs_used": 491675,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.0573842525482178,
      "objective/train/original_loss": 3.057384490966797,
      "objective/train/theoretical_loss": 3.70470109410946,
      "objective/train/tokens_used": 875704800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2431517094373703,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050538182258606,
      "objective/train/weighted_lm_loss": 3.211825370788574,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0417574644088745,
      "theoretical_loss": 3.70470109410946,
      "tokens_seen": 855244800
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003741825476429288,
      "loss": 3.0879,
      "theoretical_loss": 3.70470109410946,
      "tokens_seen": 855244800
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037417251755265794,
      "loss": 2.8349,
      "theoretical_loss": 3.7046732927776773,
      "tokens_seen": 855310336
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003741624874623872,
      "loss": 2.8502,
      "theoretical_loss": 3.7046454941724263,
      "tokens_seen": 855375872
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037415245737211635,
      "loss": 3.0087,
      "theoretical_loss": 3.70461769829323,
      "tokens_seen": 855441408
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037414242728184554,
      "loss": 2.8602,
      "theoretical_loss": 3.7045899051396125,
      "tokens_seen": 855506944
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003741323971915747,
      "loss": 3.0113,
      "theoretical_loss": 3.7045621147110985,
      "tokens_seen": 855572480
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037412236710130395,
      "loss": 2.8641,
      "theoretical_loss": 3.704534327007211,
      "tokens_seen": 855638016
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003741123370110331,
      "loss": 3.1429,
      "theoretical_loss": 3.7045065420274748,
      "tokens_seen": 855703552
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003741023069207623,
      "loss": 3.0211,
      "theoretical_loss": 3.7044787597714137,
      "tokens_seen": 855769088
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037409227683049144,
      "loss": 2.9888,
      "theoretical_loss": 3.704450980238553,
      "tokens_seen": 855834624
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003740822467402207,
      "loss": 2.9714,
      "theoretical_loss": 3.7044232034284166,
      "tokens_seen": 855900160
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037407221664994986,
      "loss": 2.9648,
      "theoretical_loss": 3.7043954293405292,
      "tokens_seen": 855965696
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037406218655967904,
      "loss": 2.863,
      "theoretical_loss": 3.7043676579744163,
      "tokens_seen": 856031232
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003740521564694082,
      "loss": 3.0788,
      "theoretical_loss": 3.704339889329602,
      "tokens_seen": 856096768
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003740421263791374,
      "loss": 3.1091,
      "theoretical_loss": 3.704312123405612,
      "tokens_seen": 856162304
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003740320962888666,
      "loss": 2.8501,
      "theoretical_loss": 3.704284360201971,
      "tokens_seen": 856227840
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003740220661985958,
      "loss": 2.8779,
      "theoretical_loss": 3.704256599718205,
      "tokens_seen": 856293376
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037401203610832494,
      "loss": 3.0087,
      "theoretical_loss": 3.704228841953838,
      "tokens_seen": 856358912
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003740020060180542,
      "loss": 3.0346,
      "theoretical_loss": 3.704201086908397,
      "tokens_seen": 856424448
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037399197592778336,
      "loss": 3.157,
      "theoretical_loss": 3.704173334581407,
      "tokens_seen": 856489984
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037398194583751254,
      "loss": 2.9318,
      "theoretical_loss": 3.7041455849723937,
      "tokens_seen": 856555520
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003739719157472417,
      "loss": 2.9098,
      "theoretical_loss": 3.704117838080883,
      "tokens_seen": 856621056
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003739618856569709,
      "loss": 3.0931,
      "theoretical_loss": 3.7040900939064008,
      "tokens_seen": 856686592
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003739518555667001,
      "loss": 2.9612,
      "theoretical_loss": 3.704062352448474,
      "tokens_seen": 856752128
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003739418254764293,
      "loss": 3.0229,
      "theoretical_loss": 3.704034613706628,
      "tokens_seen": 856817664
    },
    {
      "epoch": 0.26,
      "objective/train/advantage_avg": 0.49166038632392883,
      "objective/train/docs_used": 492132,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.514518976211548,
      "objective/train/original_loss": 2.5145187377929688,
      "objective/train/theoretical_loss": 3.7040068776803894,
      "objective/train/tokens_used": 877343200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24181833863258362,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050395131111145,
      "objective/train/weighted_lm_loss": 2.6413304805755615,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0304546356201172,
      "theoretical_loss": 3.7040068776803894,
      "tokens_seen": 856883200
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037393179538615845,
      "loss": 2.6983,
      "theoretical_loss": 3.7040068776803894,
      "tokens_seen": 856883200
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003739217652958877,
      "loss": 3.2045,
      "theoretical_loss": 3.7039791443692844,
      "tokens_seen": 856948736
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003739117352056168,
      "loss": 3.0542,
      "theoretical_loss": 3.703951413772841,
      "tokens_seen": 857014272
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037390170511534605,
      "loss": 2.9729,
      "theoretical_loss": 3.7039236858905835,
      "tokens_seen": 857079808
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003738916750250752,
      "loss": 2.8744,
      "theoretical_loss": 3.7038959607220407,
      "tokens_seen": 857145344
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003738816449348044,
      "loss": 3.1234,
      "theoretical_loss": 3.7038682382667387,
      "tokens_seen": 857210880
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037387161484453364,
      "loss": 2.9165,
      "theoretical_loss": 3.7038405185242054,
      "tokens_seen": 857276416
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037386158475426277,
      "loss": 3.1265,
      "theoretical_loss": 3.7038128014939664,
      "tokens_seen": 857341952
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.000373851554663992,
      "loss": 3.0144,
      "theoretical_loss": 3.703785087175551,
      "tokens_seen": 857407488
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003738415245737212,
      "loss": 3.1,
      "theoretical_loss": 3.703757375568485,
      "tokens_seen": 857473024
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037383149448345037,
      "loss": 3.0311,
      "theoretical_loss": 3.703729666672297,
      "tokens_seen": 857538560
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037382146439317955,
      "loss": 2.9672,
      "theoretical_loss": 3.703701960486514,
      "tokens_seen": 857604096
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003738114343029088,
      "loss": 2.9743,
      "theoretical_loss": 3.7036742570106638,
      "tokens_seen": 857669632
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003738014042126379,
      "loss": 3.0594,
      "theoretical_loss": 3.7036465562442746,
      "tokens_seen": 857735168
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037379137412236715,
      "loss": 2.9419,
      "theoretical_loss": 3.703618858186875,
      "tokens_seen": 857800704
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037378134403209627,
      "loss": 3.1438,
      "theoretical_loss": 3.7035911628379914,
      "tokens_seen": 857866240
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003737713139418255,
      "loss": 2.9369,
      "theoretical_loss": 3.703563470197154,
      "tokens_seen": 857931776
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003737612838515547,
      "loss": 3.1026,
      "theoretical_loss": 3.7035357802638895,
      "tokens_seen": 857997312
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037375125376128387,
      "loss": 2.9586,
      "theoretical_loss": 3.703508093037728,
      "tokens_seen": 858062848
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037374122367101305,
      "loss": 3.0767,
      "theoretical_loss": 3.703480408518197,
      "tokens_seen": 858128384
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037373119358074223,
      "loss": 2.8997,
      "theoretical_loss": 3.7034527267048256,
      "tokens_seen": 858193920
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003737211634904714,
      "loss": 3.0005,
      "theoretical_loss": 3.7034250475971424,
      "tokens_seen": 858259456
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037371113340020065,
      "loss": 3.217,
      "theoretical_loss": 3.7033973711946766,
      "tokens_seen": 858324992
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003737011033099298,
      "loss": 3.1125,
      "theoretical_loss": 3.7033696974969574,
      "tokens_seen": 858390528
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.000373691073219659,
      "loss": 2.9267,
      "theoretical_loss": 3.703342026503514,
      "tokens_seen": 858456064
    },
    {
      "epoch": 0.26,
      "objective/train/advantage_avg": 0.49422621726989746,
      "objective/train/docs_used": 493247,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.9833905696868896,
      "objective/train/original_loss": 2.9833908081054688,
      "objective/train/theoretical_loss": 3.7033143582138752,
      "objective/train/tokens_used": 878981600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2443377524614334,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506646633148193,
      "objective/train/weighted_lm_loss": 3.1348800659179688,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0437887907028198,
      "theoretical_loss": 3.7033143582138752,
      "tokens_seen": 858521600
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037368104312938814,
      "loss": 2.8726,
      "theoretical_loss": 3.7033143582138752,
      "tokens_seen": 858521600
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003736710130391174,
      "loss": 3.1108,
      "theoretical_loss": 3.7032866926275707,
      "tokens_seen": 858587136
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037366098294884655,
      "loss": 3.0283,
      "theoretical_loss": 3.703259029744131,
      "tokens_seen": 858652672
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037365095285857574,
      "loss": 2.8145,
      "theoretical_loss": 3.703231369563084,
      "tokens_seen": 858718208
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003736409227683049,
      "loss": 3.0143,
      "theoretical_loss": 3.703203712083961,
      "tokens_seen": 858783744
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037363089267803415,
      "loss": 3.0182,
      "theoretical_loss": 3.703176057306291,
      "tokens_seen": 858849280
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003736208625877633,
      "loss": 3.165,
      "theoretical_loss": 3.703148405229604,
      "tokens_seen": 858914816
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003736108324974925,
      "loss": 2.5957,
      "theoretical_loss": 3.7031207558534316,
      "tokens_seen": 858980352
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037360080240722164,
      "loss": 3.2859,
      "theoretical_loss": 3.703093109177302,
      "tokens_seen": 859045888
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003735907723169509,
      "loss": 3.0823,
      "theoretical_loss": 3.703065465200747,
      "tokens_seen": 859111424
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037358074222668006,
      "loss": 2.8696,
      "theoretical_loss": 3.703037823923296,
      "tokens_seen": 859176960
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037357071213640924,
      "loss": 2.904,
      "theoretical_loss": 3.703010185344481,
      "tokens_seen": 859242496
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003735606820461384,
      "loss": 3.3076,
      "theoretical_loss": 3.7029825494638313,
      "tokens_seen": 859308032
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003735506519558676,
      "loss": 3.2718,
      "theoretical_loss": 3.702954916280879,
      "tokens_seen": 859373568
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003735406218655968,
      "loss": 2.9948,
      "theoretical_loss": 3.7029272857951545,
      "tokens_seen": 859439104
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.000373530591775326,
      "loss": 3.0486,
      "theoretical_loss": 3.7028996580061886,
      "tokens_seen": 859504640
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037352056168505514,
      "loss": 2.9021,
      "theoretical_loss": 3.7028720329135125,
      "tokens_seen": 859570176
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003735105315947844,
      "loss": 3.1349,
      "theoretical_loss": 3.702844410516658,
      "tokens_seen": 859635712
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037350050150451356,
      "loss": 2.8671,
      "theoretical_loss": 3.702816790815157,
      "tokens_seen": 859701248
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037349047141424274,
      "loss": 2.9509,
      "theoretical_loss": 3.7027891738085392,
      "tokens_seen": 859766784
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003734804413239719,
      "loss": 2.8908,
      "theoretical_loss": 3.702761559496338,
      "tokens_seen": 859832320
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003734704112337011,
      "loss": 2.83,
      "theoretical_loss": 3.7027339478780847,
      "tokens_seen": 859897856
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003734603811434303,
      "loss": 2.9674,
      "theoretical_loss": 3.7027063389533112,
      "tokens_seen": 859963392
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003734503510531595,
      "loss": 3.0512,
      "theoretical_loss": 3.7026787327215493,
      "tokens_seen": 860028928
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037344032096288865,
      "loss": 2.887,
      "theoretical_loss": 3.7026511291823314,
      "tokens_seen": 860094464
    },
    {
      "epoch": 0.26,
      "objective/train/advantage_avg": 0.4912792444229126,
      "objective/train/docs_used": 493898,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.766759157180786,
      "objective/train/original_loss": 2.766759157180786,
      "objective/train/theoretical_loss": 3.70262352833519,
      "objective/train/tokens_used": 880620000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24146176874637604,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050355315208435,
      "objective/train/weighted_lm_loss": 2.9059126377105713,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.044721245765686,
      "theoretical_loss": 3.70262352833519,
      "tokens_seen": 860160000
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003734302908726179,
      "loss": 2.9576,
      "theoretical_loss": 3.70262352833519,
      "tokens_seen": 860160000
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.000373420260782347,
      "loss": 2.7747,
      "theoretical_loss": 3.7025959301796565,
      "tokens_seen": 860225536
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037341023069207625,
      "loss": 3.1111,
      "theoretical_loss": 3.702568334715264,
      "tokens_seen": 860291072
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003734002006018054,
      "loss": 2.9425,
      "theoretical_loss": 3.7025407419415455,
      "tokens_seen": 860356608
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003733901705115346,
      "loss": 2.9258,
      "theoretical_loss": 3.7025131518580325,
      "tokens_seen": 860422144
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003733801404212638,
      "loss": 2.9579,
      "theoretical_loss": 3.7024855644642596,
      "tokens_seen": 860487680
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037337011033099297,
      "loss": 3.0847,
      "theoretical_loss": 3.7024579797597585,
      "tokens_seen": 860553216
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037336008024072215,
      "loss": 2.8825,
      "theoretical_loss": 3.7024303977440627,
      "tokens_seen": 860618752
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003733500501504514,
      "loss": 3.0994,
      "theoretical_loss": 3.7024028184167053,
      "tokens_seen": 860684288
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003733400200601805,
      "loss": 2.86,
      "theoretical_loss": 3.7023752417772195,
      "tokens_seen": 860749824
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037332998996990975,
      "loss": 2.8491,
      "theoretical_loss": 3.702347667825139,
      "tokens_seen": 860815360
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037331995987963893,
      "loss": 2.852,
      "theoretical_loss": 3.7023200965599967,
      "tokens_seen": 860880896
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003733099297893681,
      "loss": 2.9375,
      "theoretical_loss": 3.7022925279813266,
      "tokens_seen": 860946432
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003732998996990973,
      "loss": 2.8582,
      "theoretical_loss": 3.702264962088663,
      "tokens_seen": 861011968
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003732898696088265,
      "loss": 3.0417,
      "theoretical_loss": 3.702237398881539,
      "tokens_seen": 861077504
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037327983951855565,
      "loss": 3.0476,
      "theoretical_loss": 3.7022098383594892,
      "tokens_seen": 861143040
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003732698094282849,
      "loss": 3.1563,
      "theoretical_loss": 3.702182280522048,
      "tokens_seen": 861208576
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.000373259779338014,
      "loss": 3.1549,
      "theoretical_loss": 3.7021547253687483,
      "tokens_seen": 861274112
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037324974924774325,
      "loss": 3.0786,
      "theoretical_loss": 3.702127172899125,
      "tokens_seen": 861339648
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003732397191574724,
      "loss": 3.1164,
      "theoretical_loss": 3.702099623112714,
      "tokens_seen": 861405184
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003732296890672016,
      "loss": 2.6662,
      "theoretical_loss": 3.7020720760090478,
      "tokens_seen": 861470720
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003732196589769308,
      "loss": 2.896,
      "theoretical_loss": 3.7020445315876622,
      "tokens_seen": 861536256
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037320962888666,
      "loss": 2.949,
      "theoretical_loss": 3.702016989848092,
      "tokens_seen": 861601792
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037319959879638916,
      "loss": 2.9613,
      "theoretical_loss": 3.701989450789872,
      "tokens_seen": 861667328
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037318956870611834,
      "loss": 2.7892,
      "theoretical_loss": 3.701961914412537,
      "tokens_seen": 861732864
    },
    {
      "epoch": 0.26,
      "objective/train/advantage_avg": 0.49156513810157776,
      "objective/train/docs_used": 495406,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.0415921211242676,
      "objective/train/original_loss": 3.0415921211242676,
      "objective/train/theoretical_loss": 3.701934380715622,
      "objective/train/tokens_used": 882258400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24178960919380188,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050385594367981,
      "objective/train/weighted_lm_loss": 3.1950247287750244,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0352214574813843,
      "theoretical_loss": 3.701934380715622,
      "tokens_seen": 861798400
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003731795386158475,
      "loss": 2.9184,
      "theoretical_loss": 3.701934380715622,
      "tokens_seen": 861798400
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037316950852557675,
      "loss": 2.9004,
      "theoretical_loss": 3.7019068496986636,
      "tokens_seen": 861863936
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003731594784353059,
      "loss": 2.9927,
      "theoretical_loss": 3.7018793213611954,
      "tokens_seen": 861929472
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003731494483450351,
      "loss": 2.9272,
      "theoretical_loss": 3.7018517957027544,
      "tokens_seen": 861995008
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003731394182547643,
      "loss": 3.0055,
      "theoretical_loss": 3.7018242727228756,
      "tokens_seen": 862060544
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003731293881644935,
      "loss": 3.0013,
      "theoretical_loss": 3.701796752421094,
      "tokens_seen": 862126080
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003731193580742227,
      "loss": 2.8573,
      "theoretical_loss": 3.701769234796947,
      "tokens_seen": 862191616
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037310932798395184,
      "loss": 3.1371,
      "theoretical_loss": 3.7017417198499696,
      "tokens_seen": 862257152
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003730992978936811,
      "loss": 3.0718,
      "theoretical_loss": 3.701714207579698,
      "tokens_seen": 862322688
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037308926780341026,
      "loss": 3.0588,
      "theoretical_loss": 3.701686697985668,
      "tokens_seen": 862388224
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037307923771313944,
      "loss": 2.8821,
      "theoretical_loss": 3.7016591910674173,
      "tokens_seen": 862453760
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003730692076228686,
      "loss": 2.9303,
      "theoretical_loss": 3.7016316868244807,
      "tokens_seen": 862519296
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003730591775325978,
      "loss": 3.0713,
      "theoretical_loss": 3.7016041852563957,
      "tokens_seen": 862584832
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.000373049147442327,
      "loss": 2.9186,
      "theoretical_loss": 3.701576686362699,
      "tokens_seen": 862650368
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003730391173520562,
      "loss": 2.9698,
      "theoretical_loss": 3.7015491901429263,
      "tokens_seen": 862715904
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037302908726178534,
      "loss": 3.1671,
      "theoretical_loss": 3.7015216965966165,
      "tokens_seen": 862781440
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003730190571715146,
      "loss": 2.8958,
      "theoretical_loss": 3.7014942057233045,
      "tokens_seen": 862846976
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037300902708124376,
      "loss": 3.0211,
      "theoretical_loss": 3.701466717522529,
      "tokens_seen": 862912512
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037299899699097294,
      "loss": 3.1189,
      "theoretical_loss": 3.7014392319938265,
      "tokens_seen": 862978048
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003729889669007021,
      "loss": 3.1457,
      "theoretical_loss": 3.701411749136734,
      "tokens_seen": 863043584
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003729789368104313,
      "loss": 3.1074,
      "theoretical_loss": 3.7013842689507896,
      "tokens_seen": 863109120
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003729689067201605,
      "loss": 3.0446,
      "theoretical_loss": 3.701356791435532,
      "tokens_seen": 863174656
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003729588766298897,
      "loss": 3.0891,
      "theoretical_loss": 3.7013293165904964,
      "tokens_seen": 863240192
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037294884653961885,
      "loss": 2.9634,
      "theoretical_loss": 3.701301844415222,
      "tokens_seen": 863305728
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003729388164493481,
      "loss": 2.9245,
      "theoretical_loss": 3.7012743749092465,
      "tokens_seen": 863371264
    },
    {
      "epoch": 0.26,
      "objective/train/advantage_avg": 0.4950138330459595,
      "objective/train/docs_used": 496136,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.7012763023376465,
      "objective/train/original_loss": 2.701275587081909,
      "objective/train/theoretical_loss": 3.7012469080721084,
      "objective/train/tokens_used": 883896800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2451152205467224,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507473945617676,
      "objective/train/weighted_lm_loss": 2.8384015560150146,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0304546356201172,
      "theoretical_loss": 3.7012469080721084,
      "tokens_seen": 863436800
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003729287863590772,
      "loss": 2.9636,
      "theoretical_loss": 3.7012469080721084,
      "tokens_seen": 863436800
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037291875626880645,
      "loss": 3.1639,
      "theoretical_loss": 3.7012194439033457,
      "tokens_seen": 863502336
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003729087261785356,
      "loss": 2.8145,
      "theoretical_loss": 3.7011919824024964,
      "tokens_seen": 863567872
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003728986960882648,
      "loss": 2.8274,
      "theoretical_loss": 3.7011645235690986,
      "tokens_seen": 863633408
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.000372888665997994,
      "loss": 3.0175,
      "theoretical_loss": 3.701137067402692,
      "tokens_seen": 863698944
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037287863590772317,
      "loss": 3.1243,
      "theoretical_loss": 3.7011096139028137,
      "tokens_seen": 863764480
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037286860581745235,
      "loss": 2.9442,
      "theoretical_loss": 3.701082163069004,
      "tokens_seen": 863830016
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003728585757271816,
      "loss": 2.8906,
      "theoretical_loss": 3.7010547149008,
      "tokens_seen": 863895552
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003728485456369107,
      "loss": 2.8241,
      "theoretical_loss": 3.701027269397742,
      "tokens_seen": 863961088
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037283851554663995,
      "loss": 3.103,
      "theoretical_loss": 3.700999826559369,
      "tokens_seen": 864026624
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037282848545636913,
      "loss": 3.0128,
      "theoretical_loss": 3.70097238638522,
      "tokens_seen": 864092160
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003728184553660983,
      "loss": 3.0924,
      "theoretical_loss": 3.700944948874834,
      "tokens_seen": 864157696
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003728084252758275,
      "loss": 2.8856,
      "theoretical_loss": 3.7009175140277506,
      "tokens_seen": 864223232
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003727983951855567,
      "loss": 2.9015,
      "theoretical_loss": 3.700890081843509,
      "tokens_seen": 864288768
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037278836509528585,
      "loss": 3.0701,
      "theoretical_loss": 3.70086265232165,
      "tokens_seen": 864354304
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003727783350050151,
      "loss": 3.1799,
      "theoretical_loss": 3.700835225461712,
      "tokens_seen": 864419840
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003727683049147442,
      "loss": 3.0883,
      "theoretical_loss": 3.7008078012632364,
      "tokens_seen": 864485376
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037275827482447345,
      "loss": 2.851,
      "theoretical_loss": 3.700780379725761,
      "tokens_seen": 864550912
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003727482447342026,
      "loss": 3.1684,
      "theoretical_loss": 3.7007529608488285,
      "tokens_seen": 864616448
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003727382146439318,
      "loss": 2.9594,
      "theoretical_loss": 3.700725544631977,
      "tokens_seen": 864681984
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.000372728184553661,
      "loss": 3.0715,
      "theoretical_loss": 3.7006981310747484,
      "tokens_seen": 864747520
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003727181544633902,
      "loss": 3.0278,
      "theoretical_loss": 3.700670720176682,
      "tokens_seen": 864813056
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037270812437311936,
      "loss": 2.8439,
      "theoretical_loss": 3.700643311937319,
      "tokens_seen": 864878592
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037269809428284854,
      "loss": 2.8993,
      "theoretical_loss": 3.7006159063562,
      "tokens_seen": 864944128
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003726880641925777,
      "loss": 2.9401,
      "theoretical_loss": 3.700588503432866,
      "tokens_seen": 865009664
    },
    {
      "epoch": 0.26,
      "objective/train/advantage_avg": 0.49503597617149353,
      "objective/train/docs_used": 497267,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.6823549270629883,
      "objective/train/original_loss": 2.6823551654815674,
      "objective/train/theoretical_loss": 3.700561103166857,
      "objective/train/tokens_used": 885535200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24510851502418518,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507495403289795,
      "objective/train/weighted_lm_loss": 2.8182380199432373,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.044721245765686,
      "theoretical_loss": 3.700561103166857,
      "tokens_seen": 865075200
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037267803410230696,
      "loss": 2.9203,
      "theoretical_loss": 3.700561103166857,
      "tokens_seen": 865075200
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003726680040120361,
      "loss": 2.8324,
      "theoretical_loss": 3.7005337055577145,
      "tokens_seen": 865140736
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003726579739217653,
      "loss": 2.8823,
      "theoretical_loss": 3.7005063106049807,
      "tokens_seen": 865206272
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003726479438314945,
      "loss": 3.05,
      "theoretical_loss": 3.700478918308195,
      "tokens_seen": 865271808
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003726379137412237,
      "loss": 3.226,
      "theoretical_loss": 3.7004515286669006,
      "tokens_seen": 865337344
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037262788365095286,
      "loss": 3.2558,
      "theoretical_loss": 3.700424141680638,
      "tokens_seen": 865402880
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037261785356068204,
      "loss": 2.849,
      "theoretical_loss": 3.7003967573489485,
      "tokens_seen": 865468416
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003726078234704112,
      "loss": 2.7242,
      "theoretical_loss": 3.7003693756713747,
      "tokens_seen": 865533952
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037259779338014046,
      "loss": 2.7237,
      "theoretical_loss": 3.7003419966474578,
      "tokens_seen": 865599488
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003725877632898696,
      "loss": 3.0475,
      "theoretical_loss": 3.70031462027674,
      "tokens_seen": 865665024
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003725777331995988,
      "loss": 2.7143,
      "theoretical_loss": 3.700287246558763,
      "tokens_seen": 865730560
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037256770310932795,
      "loss": 2.7303,
      "theoretical_loss": 3.700259875493069,
      "tokens_seen": 865796096
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003725576730190572,
      "loss": 3.0861,
      "theoretical_loss": 3.7002325070792015,
      "tokens_seen": 865861632
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037254764292878636,
      "loss": 3.0594,
      "theoretical_loss": 3.700205141316701,
      "tokens_seen": 865927168
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037253761283851554,
      "loss": 2.9414,
      "theoretical_loss": 3.700177778205111,
      "tokens_seen": 865992704
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003725275827482447,
      "loss": 3.0075,
      "theoretical_loss": 3.7001504177439744,
      "tokens_seen": 866058240
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037251755265797396,
      "loss": 2.9856,
      "theoretical_loss": 3.7001230599328334,
      "tokens_seen": 866123776
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003725075225677031,
      "loss": 3.0326,
      "theoretical_loss": 3.7000957047712317,
      "tokens_seen": 866189312
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003724974924774323,
      "loss": 2.822,
      "theoretical_loss": 3.700068352258711,
      "tokens_seen": 866254848
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037248746238716145,
      "loss": 2.8832,
      "theoretical_loss": 3.7000410023948147,
      "tokens_seen": 866320384
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003724774322968907,
      "loss": 3.2649,
      "theoretical_loss": 3.7000136551790868,
      "tokens_seen": 866385920
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037246740220661987,
      "loss": 2.7415,
      "theoretical_loss": 3.6999863106110698,
      "tokens_seen": 866451456
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037245737211634905,
      "loss": 2.7892,
      "theoretical_loss": 3.699958968690307,
      "tokens_seen": 866516992
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037244734202607823,
      "loss": 3.0594,
      "theoretical_loss": 3.6999316294163433,
      "tokens_seen": 866582528
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003724373119358074,
      "loss": 2.8134,
      "theoretical_loss": 3.6999042927887205,
      "tokens_seen": 866648064
    },
    {
      "epoch": 0.26,
      "objective/train/advantage_avg": 0.49455276131629944,
      "objective/train/docs_used": 497986,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.8584694862365723,
      "objective/train/original_loss": 2.858468770980835,
      "objective/train/theoretical_loss": 3.6998769588069838,
      "objective/train/tokens_used": 887173600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24465540051460266,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506988763809204,
      "objective/train/weighted_lm_loss": 3.0034759044647217,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0408108234405518,
      "theoretical_loss": 3.6998769588069838,
      "tokens_seen": 866713600
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003724272818455366,
      "loss": 3.0643,
      "theoretical_loss": 3.6998769588069838,
      "tokens_seen": 866713600
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003724172517552658,
      "loss": 2.6995,
      "theoretical_loss": 3.699849627470676,
      "tokens_seen": 866779136
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037240722166499495,
      "loss": 2.8524,
      "theoretical_loss": 3.699822298779342,
      "tokens_seen": 866844672
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003723971915747242,
      "loss": 3.0163,
      "theoretical_loss": 3.699794972732525,
      "tokens_seen": 866910208
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003723871614844533,
      "loss": 2.9425,
      "theoretical_loss": 3.69976764932977,
      "tokens_seen": 866975744
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037237713139418255,
      "loss": 3.0792,
      "theoretical_loss": 3.6997403285706207,
      "tokens_seen": 867041280
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003723671013039118,
      "loss": 3.0625,
      "theoretical_loss": 3.699713010454622,
      "tokens_seen": 867106816
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003723570712136409,
      "loss": 3.2135,
      "theoretical_loss": 3.6996856949813184,
      "tokens_seen": 867172352
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037234704112337015,
      "loss": 2.9147,
      "theoretical_loss": 3.699658382150254,
      "tokens_seen": 867237888
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037233701103309933,
      "loss": 2.8772,
      "theoretical_loss": 3.699631071960974,
      "tokens_seen": 867303424
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003723269809428285,
      "loss": 3.245,
      "theoretical_loss": 3.699603764413024,
      "tokens_seen": 867368960
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003723169508525577,
      "loss": 2.9414,
      "theoretical_loss": 3.6995764595059475,
      "tokens_seen": 867434496
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003723069207622869,
      "loss": 2.9012,
      "theoretical_loss": 3.699549157239291,
      "tokens_seen": 867500032
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037229689067201605,
      "loss": 3.0552,
      "theoretical_loss": 3.699521857612599,
      "tokens_seen": 867565568
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003722868605817453,
      "loss": 3.1386,
      "theoretical_loss": 3.699494560625417,
      "tokens_seen": 867631104
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003722768304914744,
      "loss": 2.9699,
      "theoretical_loss": 3.6994672662772903,
      "tokens_seen": 867696640
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037226680040120365,
      "loss": 2.8989,
      "theoretical_loss": 3.699439974567764,
      "tokens_seen": 867762176
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003722567703109328,
      "loss": 2.8998,
      "theoretical_loss": 3.6994126854963856,
      "tokens_seen": 867827712
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.000372246740220662,
      "loss": 3.3831,
      "theoretical_loss": 3.699385399062699,
      "tokens_seen": 867893248
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003722367101303912,
      "loss": 2.8805,
      "theoretical_loss": 3.69935811526625,
      "tokens_seen": 867958784
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003722266800401204,
      "loss": 3.054,
      "theoretical_loss": 3.6993308341065863,
      "tokens_seen": 868024320
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037221664994984956,
      "loss": 3.107,
      "theoretical_loss": 3.6993035555832527,
      "tokens_seen": 868089856
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037220661985957874,
      "loss": 2.9784,
      "theoretical_loss": 3.699276279695796,
      "tokens_seen": 868155392
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003721965897693079,
      "loss": 3.145,
      "theoretical_loss": 3.6992490064437624,
      "tokens_seen": 868220928
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037218655967903716,
      "loss": 2.8627,
      "theoretical_loss": 3.6992217358266983,
      "tokens_seen": 868286464
    },
    {
      "epoch": 0.26,
      "objective/train/advantage_avg": 0.4935997724533081,
      "objective/train/docs_used": 499247,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.072469472885132,
      "objective/train/original_loss": 3.072469711303711,
      "objective/train/theoretical_loss": 3.6991944678441504,
      "objective/train/tokens_used": 888812000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24372528493404388,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505988597869873,
      "objective/train/weighted_lm_loss": 3.2278196811676025,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0396549701690674,
      "theoretical_loss": 3.6991944678441504,
      "tokens_seen": 868352000
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003721765295887663,
      "loss": 3.047,
      "theoretical_loss": 3.6991944678441504,
      "tokens_seen": 868352000
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003721664994984955,
      "loss": 2.6629,
      "theoretical_loss": 3.6991672024956648,
      "tokens_seen": 868417536
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003721564694082247,
      "loss": 2.8508,
      "theoretical_loss": 3.699139939780789,
      "tokens_seen": 868483072
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003721464393179539,
      "loss": 2.901,
      "theoretical_loss": 3.69911267969907,
      "tokens_seen": 868548608
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037213640922768306,
      "loss": 3.0083,
      "theoretical_loss": 3.699085422250054,
      "tokens_seen": 868614144
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037212637913741224,
      "loss": 2.986,
      "theoretical_loss": 3.6990581674332894,
      "tokens_seen": 868679680
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003721163490471414,
      "loss": 2.8911,
      "theoretical_loss": 3.699030915248322,
      "tokens_seen": 868745216
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037210631895687066,
      "loss": 3.1032,
      "theoretical_loss": 3.6990036656947005,
      "tokens_seen": 868810752
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003720962888665998,
      "loss": 2.913,
      "theoretical_loss": 3.698976418771971,
      "tokens_seen": 868876288
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.000372086258776329,
      "loss": 2.8891,
      "theoretical_loss": 3.698949174479682,
      "tokens_seen": 868941824
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037207622868605815,
      "loss": 2.9546,
      "theoretical_loss": 3.698921932817381,
      "tokens_seen": 869007360
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003720661985957874,
      "loss": 2.9227,
      "theoretical_loss": 3.6988946937846165,
      "tokens_seen": 869072896
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037205616850551656,
      "loss": 2.9411,
      "theoretical_loss": 3.698867457380935,
      "tokens_seen": 869138432
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037204613841524575,
      "loss": 3.2472,
      "theoretical_loss": 3.6988402236058855,
      "tokens_seen": 869203968
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003720361083249749,
      "loss": 2.8445,
      "theoretical_loss": 3.6988129924590156,
      "tokens_seen": 869269504
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037202607823470416,
      "loss": 3.0026,
      "theoretical_loss": 3.698785763939874,
      "tokens_seen": 869335040
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003720160481444333,
      "loss": 3.1296,
      "theoretical_loss": 3.698758538048009,
      "tokens_seen": 869400576
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003720060180541625,
      "loss": 3.2111,
      "theoretical_loss": 3.698731314782969,
      "tokens_seen": 869466112
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037199598796389165,
      "loss": 2.8957,
      "theoretical_loss": 3.698704094144303,
      "tokens_seen": 869531648
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003719859578736209,
      "loss": 2.9831,
      "theoretical_loss": 3.698676876131559,
      "tokens_seen": 869597184
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037197592778335007,
      "loss": 2.9501,
      "theoretical_loss": 3.6986496607442856,
      "tokens_seen": 869662720
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037196589769307925,
      "loss": 2.8903,
      "theoretical_loss": 3.698622447982033,
      "tokens_seen": 869728256
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037195586760280843,
      "loss": 3.0195,
      "theoretical_loss": 3.6985952378443487,
      "tokens_seen": 869793792
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003719458375125376,
      "loss": 2.8295,
      "theoretical_loss": 3.698568030330783,
      "tokens_seen": 869859328
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003719358074222668,
      "loss": 3.002,
      "theoretical_loss": 3.698540825440885,
      "tokens_seen": 869924864
    },
    {
      "epoch": 0.26,
      "objective/train/advantage_avg": 0.4940377175807953,
      "objective/train/docs_used": 499921,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.5911357402801514,
      "objective/train/original_loss": 2.5911355018615723,
      "objective/train/theoretical_loss": 3.6985136231742035,
      "objective/train/tokens_used": 890450400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24421533942222595,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506452322006226,
      "objective/train/weighted_lm_loss": 2.722421407699585,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0253151655197144,
      "theoretical_loss": 3.6985136231742035,
      "tokens_seen": 869990400
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.000371925777331996,
      "loss": 2.9368,
      "theoretical_loss": 3.6985136231742035,
      "tokens_seen": 869990400
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037191574724172515,
      "loss": 2.7618,
      "theoretical_loss": 3.6984864235302886,
      "tokens_seen": 870055936
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003719057171514544,
      "loss": 2.7977,
      "theoretical_loss": 3.6984592265086893,
      "tokens_seen": 870121472
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003718956870611835,
      "loss": 2.9132,
      "theoretical_loss": 3.698432032108956,
      "tokens_seen": 870187008
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037188565697091275,
      "loss": 2.6327,
      "theoretical_loss": 3.6984048403306375,
      "tokens_seen": 870252544
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037187562688064193,
      "loss": 2.9597,
      "theoretical_loss": 3.698377651173285,
      "tokens_seen": 870318080
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003718655967903711,
      "loss": 2.7397,
      "theoretical_loss": 3.698350464636447,
      "tokens_seen": 870383616
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003718555667001003,
      "loss": 2.7936,
      "theoretical_loss": 3.6983232807196753,
      "tokens_seen": 870449152
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037184553660982953,
      "loss": 3.1205,
      "theoretical_loss": 3.698296099422519,
      "tokens_seen": 870514688
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037183550651955866,
      "loss": 3.2295,
      "theoretical_loss": 3.698268920744529,
      "tokens_seen": 870580224
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003718254764292879,
      "loss": 3.1454,
      "theoretical_loss": 3.698241744685256,
      "tokens_seen": 870645760
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.000371815446339017,
      "loss": 3.1301,
      "theoretical_loss": 3.6982145712442502,
      "tokens_seen": 870711296
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037180541624874625,
      "loss": 2.8032,
      "theoretical_loss": 3.6981874004210615,
      "tokens_seen": 870776832
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037179538615847544,
      "loss": 3.0276,
      "theoretical_loss": 3.6981602322152423,
      "tokens_seen": 870842368
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003717853560682046,
      "loss": 2.8952,
      "theoretical_loss": 3.698133066626343,
      "tokens_seen": 870907904
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003717753259779338,
      "loss": 3.2167,
      "theoretical_loss": 3.6981059036539135,
      "tokens_seen": 870973440
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.000371765295887663,
      "loss": 2.8171,
      "theoretical_loss": 3.698078743297507,
      "tokens_seen": 871038976
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037175526579739216,
      "loss": 2.8972,
      "theoretical_loss": 3.698051585556673,
      "tokens_seen": 871104512
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003717452357071214,
      "loss": 2.7835,
      "theoretical_loss": 3.698024430430963,
      "tokens_seen": 871170048
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003717352056168505,
      "loss": 3.1419,
      "theoretical_loss": 3.6979972779199297,
      "tokens_seen": 871235584
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037172517552657976,
      "loss": 2.9609,
      "theoretical_loss": 3.6979701280231234,
      "tokens_seen": 871301120
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003717151454363089,
      "loss": 2.8272,
      "theoretical_loss": 3.6979429807400965,
      "tokens_seen": 871366656
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003717051153460381,
      "loss": 2.6648,
      "theoretical_loss": 3.697915836070401,
      "tokens_seen": 871432192
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003716950852557673,
      "loss": 2.7508,
      "theoretical_loss": 3.697888694013588,
      "tokens_seen": 871497728
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003716850551654965,
      "loss": 2.8917,
      "theoretical_loss": 3.6978615545692106,
      "tokens_seen": 871563264
    },
    {
      "epoch": 0.26,
      "objective/train/advantage_avg": 0.49236348271369934,
      "objective/train/docs_used": 501092,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.6943612098693848,
      "objective/train/original_loss": 2.6943612098693848,
      "objective/train/theoretical_loss": 3.69783441773682,
      "objective/train/tokens_used": 892088800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2425156980752945,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0504690408706665,
      "objective/train/weighted_lm_loss": 2.830291509628296,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0338950157165527,
      "theoretical_loss": 3.69783441773682,
      "tokens_seen": 871628800
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037167502507522566,
      "loss": 2.8016,
      "theoretical_loss": 3.69783441773682,
      "tokens_seen": 871628800
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003716649949849549,
      "loss": 3.0245,
      "theoretical_loss": 3.697807283515969,
      "tokens_seen": 871694336
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.000371654964894684,
      "loss": 3.0958,
      "theoretical_loss": 3.6977801519062092,
      "tokens_seen": 871759872
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037164493480441326,
      "loss": 2.8913,
      "theoretical_loss": 3.6977530229070945,
      "tokens_seen": 871825408
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003716349047141424,
      "loss": 3.0151,
      "theoretical_loss": 3.6977258965181763,
      "tokens_seen": 871890944
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003716248746238716,
      "loss": 2.9307,
      "theoretical_loss": 3.697698772739007,
      "tokens_seen": 871956480
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037161484453360086,
      "loss": 2.8632,
      "theoretical_loss": 3.6976716515691406,
      "tokens_seen": 872022016
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037160481444333,
      "loss": 2.9426,
      "theoretical_loss": 3.6976445330081296,
      "tokens_seen": 872087552
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003715947843530592,
      "loss": 2.7824,
      "theoretical_loss": 3.6976174170555263,
      "tokens_seen": 872153088
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037158475426278835,
      "loss": 2.7882,
      "theoretical_loss": 3.697590303710885,
      "tokens_seen": 872218624
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003715747241725176,
      "loss": 3.0397,
      "theoretical_loss": 3.6975631929737576,
      "tokens_seen": 872284160
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037156469408224676,
      "loss": 3.0053,
      "theoretical_loss": 3.6975360848436987,
      "tokens_seen": 872349696
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037155466399197595,
      "loss": 2.841,
      "theoretical_loss": 3.6975089793202613,
      "tokens_seen": 872415232
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003715446339017051,
      "loss": 2.8228,
      "theoretical_loss": 3.697481876402999,
      "tokens_seen": 872480768
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037153460381143436,
      "loss": 2.973,
      "theoretical_loss": 3.6974547760914653,
      "tokens_seen": 872546304
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003715245737211635,
      "loss": 2.8778,
      "theoretical_loss": 3.697427678385214,
      "tokens_seen": 872611840
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003715145436308927,
      "loss": 2.7768,
      "theoretical_loss": 3.697400583283799,
      "tokens_seen": 872677376
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037150451354062185,
      "loss": 2.9672,
      "theoretical_loss": 3.6973734907867746,
      "tokens_seen": 872742912
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003714944834503511,
      "loss": 2.7928,
      "theoretical_loss": 3.697346400893695,
      "tokens_seen": 872808448
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037148445336008027,
      "loss": 3.2422,
      "theoretical_loss": 3.697319313604114,
      "tokens_seen": 872873984
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037147442326980945,
      "loss": 2.9653,
      "theoretical_loss": 3.6972922289175862,
      "tokens_seen": 872939520
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037146439317953863,
      "loss": 2.8877,
      "theoretical_loss": 3.697265146833666,
      "tokens_seen": 873005056
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003714543630892678,
      "loss": 3.024,
      "theoretical_loss": 3.6972380673519076,
      "tokens_seen": 873070592
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.000371444332998997,
      "loss": 3.0884,
      "theoretical_loss": 3.6972109904718664,
      "tokens_seen": 873136128
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037143430290872623,
      "loss": 2.9644,
      "theoretical_loss": 3.6971839161930964,
      "tokens_seen": 873201664
    },
    {
      "epoch": 0.26,
      "objective/train/advantage_avg": 0.4922923147678375,
      "objective/train/docs_used": 501627,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.0124869346618652,
      "objective/train/original_loss": 3.0124874114990234,
      "objective/train/theoretical_loss": 3.6971568445151535,
      "objective/train/tokens_used": 893727200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24248440563678741,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050461769104004,
      "objective/train/weighted_lm_loss": 3.1643965244293213,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0417574644088745,
      "theoretical_loss": 3.6971568445151535,
      "tokens_seen": 873267200
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037142427281845535,
      "loss": 2.9921,
      "theoretical_loss": 3.6971568445151535,
      "tokens_seen": 873267200
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003714142427281846,
      "loss": 3.1184,
      "theoretical_loss": 3.697129775437592,
      "tokens_seen": 873332736
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003714042126379137,
      "loss": 2.8651,
      "theoretical_loss": 3.697102708959967,
      "tokens_seen": 873398272
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037139418254764295,
      "loss": 2.9219,
      "theoretical_loss": 3.697075645081833,
      "tokens_seen": 873463808
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037138415245737213,
      "loss": 3.0784,
      "theoretical_loss": 3.697048583802747,
      "tokens_seen": 873529344
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003713741223671013,
      "loss": 2.9862,
      "theoretical_loss": 3.697021525122264,
      "tokens_seen": 873594880
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003713640922768305,
      "loss": 3.11,
      "theoretical_loss": 3.6969944690399386,
      "tokens_seen": 873660416
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037135406218655973,
      "loss": 2.8958,
      "theoretical_loss": 3.696967415555327,
      "tokens_seen": 873725952
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037134403209628886,
      "loss": 3.0092,
      "theoretical_loss": 3.6969403646679853,
      "tokens_seen": 873791488
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003713340020060181,
      "loss": 3.1934,
      "theoretical_loss": 3.696913316377469,
      "tokens_seen": 873857024
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003713239719157472,
      "loss": 3.1181,
      "theoretical_loss": 3.6968862706833336,
      "tokens_seen": 873922560
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037131394182547645,
      "loss": 2.8902,
      "theoretical_loss": 3.6968592275851364,
      "tokens_seen": 873988096
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037130391173520564,
      "loss": 2.8199,
      "theoretical_loss": 3.6968321870824328,
      "tokens_seen": 874053632
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003712938816449348,
      "loss": 2.7883,
      "theoretical_loss": 3.696805149174779,
      "tokens_seen": 874119168
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.000371283851554664,
      "loss": 2.8147,
      "theoretical_loss": 3.6967781138617317,
      "tokens_seen": 874184704
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003712738214643932,
      "loss": 3.0195,
      "theoretical_loss": 3.696751081142848,
      "tokens_seen": 874250240
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.00037126379137412236,
      "loss": 3.0448,
      "theoretical_loss": 3.696724051017683,
      "tokens_seen": 874315776
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003712537612838516,
      "loss": 3.0244,
      "theoretical_loss": 3.696697023485795,
      "tokens_seen": 874381312
    },
    {
      "epoch": 0.26,
      "learning_rate": 0.0003712437311935807,
      "loss": 2.7068,
      "theoretical_loss": 3.6966699985467404,
      "tokens_seen": 874446848
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00037123370110330996,
      "loss": 3.2251,
      "theoretical_loss": 3.6966429762000756,
      "tokens_seen": 874512384
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003712236710130391,
      "loss": 2.8442,
      "theoretical_loss": 3.6966159564453585,
      "tokens_seen": 874577920
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003712136409227683,
      "loss": 3.0164,
      "theoretical_loss": 3.6965889392821456,
      "tokens_seen": 874643456
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003712036108324975,
      "loss": 2.8888,
      "theoretical_loss": 3.696561924709995,
      "tokens_seen": 874708992
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003711935807422267,
      "loss": 2.7823,
      "theoretical_loss": 3.6965349127284632,
      "tokens_seen": 874774528
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00037118355065195586,
      "loss": 3.1186,
      "theoretical_loss": 3.6965079033371078,
      "tokens_seen": 874840064
    },
    {
      "epoch": 0.27,
      "objective/train/advantage_avg": 0.4946102499961853,
      "objective/train/docs_used": 502963,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.748643636703491,
      "objective/train/original_loss": 2.7486438751220703,
      "objective/train/theoretical_loss": 3.696480896535487,
      "objective/train/tokens_used": 895365600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24467743933200836,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050704836845398,
      "objective/train/weighted_lm_loss": 2.8880460262298584,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0442860126495361,
      "theoretical_loss": 3.696480896535487,
      "tokens_seen": 874905600
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003711735205616851,
      "loss": 2.8908,
      "theoretical_loss": 3.696480896535487,
      "tokens_seen": 874905600
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003711634904714142,
      "loss": 2.9968,
      "theoretical_loss": 3.6964538923231585,
      "tokens_seen": 874971136
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00037115346038114346,
      "loss": 3.0013,
      "theoretical_loss": 3.696426890699679,
      "tokens_seen": 875036672
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003711434302908726,
      "loss": 3.1627,
      "theoretical_loss": 3.6963998916646084,
      "tokens_seen": 875102208
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003711334002006018,
      "loss": 3.0626,
      "theoretical_loss": 3.6963728952175035,
      "tokens_seen": 875167744
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.000371123370110331,
      "loss": 2.7765,
      "theoretical_loss": 3.6963459013579225,
      "tokens_seen": 875233280
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003711133400200602,
      "loss": 2.7893,
      "theoretical_loss": 3.6963189100854237,
      "tokens_seen": 875298816
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00037110330992978937,
      "loss": 2.9885,
      "theoretical_loss": 3.696291921399566,
      "tokens_seen": 875364352
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00037109327983951855,
      "loss": 2.8714,
      "theoretical_loss": 3.696264935299907,
      "tokens_seen": 875429888
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00037108324974924773,
      "loss": 3.0962,
      "theoretical_loss": 3.696237951786006,
      "tokens_seen": 875495424
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00037107321965897696,
      "loss": 3.0055,
      "theoretical_loss": 3.696210970857422,
      "tokens_seen": 875560960
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003710631895687061,
      "loss": 2.8386,
      "theoretical_loss": 3.6961839925137125,
      "tokens_seen": 875626496
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003710531594784353,
      "loss": 2.6906,
      "theoretical_loss": 3.6961570167544373,
      "tokens_seen": 875692032
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00037104312938816445,
      "loss": 2.7212,
      "theoretical_loss": 3.696130043579156,
      "tokens_seen": 875757568
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003710330992978937,
      "loss": 2.6783,
      "theoretical_loss": 3.6961030729874267,
      "tokens_seen": 875823104
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00037102306920762287,
      "loss": 3.1323,
      "theoretical_loss": 3.6960761049788093,
      "tokens_seen": 875888640
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00037101303911735205,
      "loss": 2.9872,
      "theoretical_loss": 3.696049139552863,
      "tokens_seen": 875954176
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00037100300902708123,
      "loss": 2.8677,
      "theoretical_loss": 3.6960221767091466,
      "tokens_seen": 876019712
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00037099297893681047,
      "loss": 2.8295,
      "theoretical_loss": 3.695995216447221,
      "tokens_seen": 876085248
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003709829488465396,
      "loss": 2.9825,
      "theoretical_loss": 3.6959682587666443,
      "tokens_seen": 876150784
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00037097291875626883,
      "loss": 3.3414,
      "theoretical_loss": 3.6959413036669777,
      "tokens_seen": 876216320
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00037096288866599796,
      "loss": 2.8463,
      "theoretical_loss": 3.6959143511477803,
      "tokens_seen": 876281856
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003709528585757272,
      "loss": 2.9221,
      "theoretical_loss": 3.6958874012086125,
      "tokens_seen": 876347392
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003709428284854564,
      "loss": 2.9685,
      "theoretical_loss": 3.695860453849034,
      "tokens_seen": 876412928
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00037093279839518555,
      "loss": 2.7036,
      "theoretical_loss": 3.6958335090686054,
      "tokens_seen": 876478464
    },
    {
      "epoch": 0.27,
      "objective/train/advantage_avg": 0.4933631122112274,
      "objective/train/docs_used": 503455,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.9829068183898926,
      "objective/train/original_loss": 2.9829063415527344,
      "objective/train/theoretical_loss": 3.695806566866887,
      "objective/train/tokens_used": 897004000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2435331493616104,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505743026733398,
      "objective/train/weighted_lm_loss": 3.133878707885742,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0408108234405518,
      "theoretical_loss": 3.695806566866887,
      "tokens_seen": 876544000
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00037092276830491474,
      "loss": 2.9343,
      "theoretical_loss": 3.695806566866887,
      "tokens_seen": 876544000
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003709127382146439,
      "loss": 2.5591,
      "theoretical_loss": 3.695779627243439,
      "tokens_seen": 876609536
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003709027081243731,
      "loss": 2.8415,
      "theoretical_loss": 3.695752690197822,
      "tokens_seen": 876675072
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00037089267803410233,
      "loss": 2.734,
      "theoretical_loss": 3.6957257557295966,
      "tokens_seen": 876740608
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00037088264794383146,
      "loss": 3.0547,
      "theoretical_loss": 3.695698823838324,
      "tokens_seen": 876806144
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003708726178535607,
      "loss": 3.0478,
      "theoretical_loss": 3.6956718945235645,
      "tokens_seen": 876871680
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00037086258776328993,
      "loss": 2.9657,
      "theoretical_loss": 3.6956449677848795,
      "tokens_seen": 876937216
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00037085255767301906,
      "loss": 2.9141,
      "theoretical_loss": 3.6956180436218293,
      "tokens_seen": 877002752
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003708425275827483,
      "loss": 3.1238,
      "theoretical_loss": 3.695591122033976,
      "tokens_seen": 877068288
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003708324974924774,
      "loss": 2.6161,
      "theoretical_loss": 3.695564203020881,
      "tokens_seen": 877133824
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00037082246740220665,
      "loss": 3.0754,
      "theoretical_loss": 3.695537286582105,
      "tokens_seen": 877199360
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00037081243731193584,
      "loss": 2.9956,
      "theoretical_loss": 3.6955103727172096,
      "tokens_seen": 877264896
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.000370802407221665,
      "loss": 3.0153,
      "theoretical_loss": 3.695483461425757,
      "tokens_seen": 877330432
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003707923771313942,
      "loss": 3.168,
      "theoretical_loss": 3.6954565527073084,
      "tokens_seen": 877395968
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003707823470411234,
      "loss": 2.7258,
      "theoretical_loss": 3.6954296465614256,
      "tokens_seen": 877461504
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00037077231695085256,
      "loss": 2.9737,
      "theoretical_loss": 3.695402742987671,
      "tokens_seen": 877527040
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003707622868605818,
      "loss": 2.7466,
      "theoretical_loss": 3.695375841985606,
      "tokens_seen": 877592576
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003707522567703109,
      "loss": 2.9859,
      "theoretical_loss": 3.695348943554793,
      "tokens_seen": 877658112
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00037074222668004016,
      "loss": 2.7539,
      "theoretical_loss": 3.6953220476947948,
      "tokens_seen": 877723648
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003707321965897693,
      "loss": 2.851,
      "theoretical_loss": 3.695295154405173,
      "tokens_seen": 877789184
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003707221664994985,
      "loss": 2.921,
      "theoretical_loss": 3.6952682636854908,
      "tokens_seen": 877854720
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003707121364092277,
      "loss": 2.9804,
      "theoretical_loss": 3.69524137553531,
      "tokens_seen": 877920256
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003707021063189569,
      "loss": 2.9744,
      "theoretical_loss": 3.695214489954194,
      "tokens_seen": 877985792
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00037069207622868606,
      "loss": 2.8848,
      "theoretical_loss": 3.6951876069417056,
      "tokens_seen": 878051328
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003706820461384153,
      "loss": 2.9915,
      "theoretical_loss": 3.6951607264974067,
      "tokens_seen": 878116864
    },
    {
      "epoch": 0.27,
      "objective/train/advantage_avg": 0.49528956413269043,
      "objective/train/docs_used": 504719,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.235717535018921,
      "objective/train/original_loss": 3.235717535018921,
      "objective/train/theoretical_loss": 3.695133848620862,
      "objective/train/tokens_used": 898642400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2453690469264984,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507762432098389,
      "objective/train/weighted_lm_loss": 3.4000658988952637,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0434160232543945,
      "theoretical_loss": 3.695133848620862,
      "tokens_seen": 878182400
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003706720160481444,
      "loss": 3.15,
      "theoretical_loss": 3.695133848620862,
      "tokens_seen": 878182400
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00037066198595787366,
      "loss": 3.1808,
      "theoretical_loss": 3.695106973311632,
      "tokens_seen": 878247936
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003706519558676028,
      "loss": 3.0491,
      "theoretical_loss": 3.695080100569283,
      "tokens_seen": 878313472
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.000370641925777332,
      "loss": 2.8171,
      "theoretical_loss": 3.695053230393376,
      "tokens_seen": 878379008
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003706318956870612,
      "loss": 2.8106,
      "theoretical_loss": 3.6950263627834765,
      "tokens_seen": 878444544
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003706218655967904,
      "loss": 3.0536,
      "theoretical_loss": 3.694999497739146,
      "tokens_seen": 878510080
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00037061183550651957,
      "loss": 3.0169,
      "theoretical_loss": 3.6949726352599495,
      "tokens_seen": 878575616
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00037060180541624875,
      "loss": 2.8802,
      "theoretical_loss": 3.6949457753454498,
      "tokens_seen": 878641152
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00037059177532597793,
      "loss": 2.9218,
      "theoretical_loss": 3.6949189179952113,
      "tokens_seen": 878706688
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00037058174523570716,
      "loss": 2.7601,
      "theoretical_loss": 3.694892063208798,
      "tokens_seen": 878772224
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003705717151454363,
      "loss": 2.9445,
      "theoretical_loss": 3.694865210985774,
      "tokens_seen": 878837760
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003705616850551655,
      "loss": 3.0788,
      "theoretical_loss": 3.6948383613257034,
      "tokens_seen": 878903296
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00037055165496489465,
      "loss": 2.8451,
      "theoretical_loss": 3.6948115142281504,
      "tokens_seen": 878968832
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003705416248746239,
      "loss": 2.7722,
      "theoretical_loss": 3.6947846696926803,
      "tokens_seen": 879034368
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00037053159478435307,
      "loss": 2.8346,
      "theoretical_loss": 3.6947578277188557,
      "tokens_seen": 879099904
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00037052156469408225,
      "loss": 2.9434,
      "theoretical_loss": 3.694730988306243,
      "tokens_seen": 879165440
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00037051153460381143,
      "loss": 3.0089,
      "theoretical_loss": 3.694704151454406,
      "tokens_seen": 879230976
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00037050150451354067,
      "loss": 2.9323,
      "theoretical_loss": 3.6946773171629097,
      "tokens_seen": 879296512
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003704914744232698,
      "loss": 2.5535,
      "theoretical_loss": 3.6946504854313194,
      "tokens_seen": 879362048
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00037048144433299903,
      "loss": 3.0903,
      "theoretical_loss": 3.6946236562591994,
      "tokens_seen": 879427584
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00037047141424272816,
      "loss": 3.1458,
      "theoretical_loss": 3.6945968296461156,
      "tokens_seen": 879493120
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003704613841524574,
      "loss": 2.9239,
      "theoretical_loss": 3.6945700055916326,
      "tokens_seen": 879558656
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003704513540621866,
      "loss": 2.911,
      "theoretical_loss": 3.6945431840953162,
      "tokens_seen": 879624192
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00037044132397191575,
      "loss": 2.7615,
      "theoretical_loss": 3.694516365156732,
      "tokens_seen": 879689728
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00037043129388164494,
      "loss": 2.9812,
      "theoretical_loss": 3.6944895487754454,
      "tokens_seen": 879755264
    },
    {
      "epoch": 0.27,
      "objective/train/advantage_avg": 0.4914601147174835,
      "objective/train/docs_used": 505361,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.032048463821411,
      "objective/train/original_loss": 3.0320487022399902,
      "objective/train/theoretical_loss": 3.694462734951022,
      "objective/train/tokens_used": 900280800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24170280992984772,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0503746271133423,
      "objective/train/weighted_lm_loss": 3.1840810775756836,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 3.694462734951022,
      "tokens_seen": 879820800
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003704212637913741,
      "loss": 3.0857,
      "theoretical_loss": 3.694462734951022,
      "tokens_seen": 879820800
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003704112337011033,
      "loss": 3.0262,
      "theoretical_loss": 3.6944359236830273,
      "tokens_seen": 879886336
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00037040120361083253,
      "loss": 2.8038,
      "theoretical_loss": 3.6944091149710276,
      "tokens_seen": 879951872
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00037039117352056166,
      "loss": 2.9796,
      "theoretical_loss": 3.694382308814589,
      "tokens_seen": 880017408
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003703811434302909,
      "loss": 2.962,
      "theoretical_loss": 3.6943555052132773,
      "tokens_seen": 880082944
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003703711133400201,
      "loss": 3.1228,
      "theoretical_loss": 3.6943287041666584,
      "tokens_seen": 880148480
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00037036108324974926,
      "loss": 3.164,
      "theoretical_loss": 3.6943019056742994,
      "tokens_seen": 880214016
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00037035105315947844,
      "loss": 3.1824,
      "theoretical_loss": 3.694275109735766,
      "tokens_seen": 880279552
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003703410230692076,
      "loss": 2.9985,
      "theoretical_loss": 3.694248316350625,
      "tokens_seen": 880345088
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003703309929789368,
      "loss": 2.9182,
      "theoretical_loss": 3.6942215255184436,
      "tokens_seen": 880410624
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00037032096288866604,
      "loss": 3.0293,
      "theoretical_loss": 3.6941947372387873,
      "tokens_seen": 880476160
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00037031093279839516,
      "loss": 3.0405,
      "theoretical_loss": 3.694167951511224,
      "tokens_seen": 880541696
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003703009027081244,
      "loss": 3.0111,
      "theoretical_loss": 3.69414116833532,
      "tokens_seen": 880607232
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003702908726178535,
      "loss": 2.9713,
      "theoretical_loss": 3.694114387710643,
      "tokens_seen": 880672768
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00037028084252758276,
      "loss": 3.0342,
      "theoretical_loss": 3.69408760963676,
      "tokens_seen": 880738304
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00037027081243731194,
      "loss": 2.9504,
      "theoretical_loss": 3.6940608341132375,
      "tokens_seen": 880803840
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003702607823470411,
      "loss": 2.9434,
      "theoretical_loss": 3.694034061139644,
      "tokens_seen": 880869376
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003702507522567703,
      "loss": 2.9629,
      "theoretical_loss": 3.694007290715546,
      "tokens_seen": 880934912
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003702407221664995,
      "loss": 2.7688,
      "theoretical_loss": 3.693980522840512,
      "tokens_seen": 881000448
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00037023069207622867,
      "loss": 2.9821,
      "theoretical_loss": 3.693953757514109,
      "tokens_seen": 881065984
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003702206619859579,
      "loss": 3.0744,
      "theoretical_loss": 3.693926994735904,
      "tokens_seen": 881131520
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00037021063189568703,
      "loss": 3.1138,
      "theoretical_loss": 3.693900234505467,
      "tokens_seen": 881197056
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00037020060180541626,
      "loss": 3.0195,
      "theoretical_loss": 3.6938734768223647,
      "tokens_seen": 881262592
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00037019057171514544,
      "loss": 2.9518,
      "theoretical_loss": 3.693846721686165,
      "tokens_seen": 881328128
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003701805416248746,
      "loss": 3.0893,
      "theoretical_loss": 3.693819969096437,
      "tokens_seen": 881393664
    },
    {
      "epoch": 0.27,
      "objective/train/advantage_avg": 0.4945959448814392,
      "objective/train/docs_used": 505856,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.7043116092681885,
      "objective/train/original_loss": 2.7043116092681885,
      "objective/train/theoretical_loss": 3.693793219052748,
      "objective/train/tokens_used": 901919200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24470101296901703,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507034063339233,
      "objective/train/weighted_lm_loss": 2.8414785861968994,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 3.693793219052748,
      "tokens_seen": 881459200
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003701705115346038,
      "loss": 2.9544,
      "theoretical_loss": 3.693793219052748,
      "tokens_seen": 881459200
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.000370160481444333,
      "loss": 2.8035,
      "theoretical_loss": 3.693766471554667,
      "tokens_seen": 881524736
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00037015045135406217,
      "loss": 3.043,
      "theoretical_loss": 3.6937397266017626,
      "tokens_seen": 881590272
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003701404212637914,
      "loss": 2.9554,
      "theoretical_loss": 3.693712984193603,
      "tokens_seen": 881655808
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00037013039117352053,
      "loss": 2.9213,
      "theoretical_loss": 3.693686244329758,
      "tokens_seen": 881721344
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00037012036108324977,
      "loss": 2.9786,
      "theoretical_loss": 3.6936595070097953,
      "tokens_seen": 881786880
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00037011033099297895,
      "loss": 2.7952,
      "theoretical_loss": 3.693632772233284,
      "tokens_seen": 881852416
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00037010030090270813,
      "loss": 2.9985,
      "theoretical_loss": 3.6936060399997936,
      "tokens_seen": 881917952
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00037009027081243736,
      "loss": 3.0119,
      "theoretical_loss": 3.693579310308893,
      "tokens_seen": 881983488
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003700802407221665,
      "loss": 2.9115,
      "theoretical_loss": 3.6935525831601517,
      "tokens_seen": 882049024
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003700702106318957,
      "loss": 2.9785,
      "theoretical_loss": 3.6935258585531385,
      "tokens_seen": 882114560
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00037006018054162485,
      "loss": 2.753,
      "theoretical_loss": 3.6934991364874232,
      "tokens_seen": 882180096
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003700501504513541,
      "loss": 2.8551,
      "theoretical_loss": 3.6934724169625763,
      "tokens_seen": 882245632
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00037004012036108327,
      "loss": 3.1019,
      "theoretical_loss": 3.6934456999781657,
      "tokens_seen": 882311168
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00037003009027081245,
      "loss": 2.9931,
      "theoretical_loss": 3.6934189855337625,
      "tokens_seen": 882376704
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00037002006018054163,
      "loss": 2.9272,
      "theoretical_loss": 3.693392273628936,
      "tokens_seen": 882442240
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00037001003009027087,
      "loss": 2.8205,
      "theoretical_loss": 3.6933655642632566,
      "tokens_seen": 882507776
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00037,
      "loss": 2.756,
      "theoretical_loss": 3.693338857436294,
      "tokens_seen": 882573312
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036998996990972923,
      "loss": 3.0837,
      "theoretical_loss": 3.693312153147619,
      "tokens_seen": 882638848
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036997993981945836,
      "loss": 2.729,
      "theoretical_loss": 3.693285451396801,
      "tokens_seen": 882704384
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003699699097291876,
      "loss": 2.6743,
      "theoretical_loss": 3.6932587521834117,
      "tokens_seen": 882769920
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003699598796389168,
      "loss": 2.8144,
      "theoretical_loss": 3.693232055507021,
      "tokens_seen": 882835456
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036994984954864595,
      "loss": 3.045,
      "theoretical_loss": 3.6932053613671982,
      "tokens_seen": 882900992
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036993981945837514,
      "loss": 2.8417,
      "theoretical_loss": 3.6931786697635163,
      "tokens_seen": 882966528
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003699297893681043,
      "loss": 2.9037,
      "theoretical_loss": 3.6931519806955446,
      "tokens_seen": 883032064
    },
    {
      "epoch": 0.27,
      "objective/train/advantage_avg": 0.49602365493774414,
      "objective/train/docs_used": 506939,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.677358627319336,
      "objective/train/original_loss": 2.677358627319336,
      "objective/train/theoretical_loss": 3.6931252941628547,
      "objective/train/tokens_used": 903557600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24607020616531372,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0508532524108887,
      "objective/train/weighted_lm_loss": 2.8134117126464844,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0472354888916016,
      "theoretical_loss": 3.6931252941628547,
      "tokens_seen": 883097600
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003699197592778335,
      "loss": 2.9825,
      "theoretical_loss": 3.6931252941628547,
      "tokens_seen": 883097600
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036990972918756273,
      "loss": 3.2138,
      "theoretical_loss": 3.693098610165017,
      "tokens_seen": 883163136
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036989969909729186,
      "loss": 3.2969,
      "theoretical_loss": 3.693071928701604,
      "tokens_seen": 883228672
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003698896690070211,
      "loss": 3.0442,
      "theoretical_loss": 3.6930452497721857,
      "tokens_seen": 883294208
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003698796389167503,
      "loss": 3.0118,
      "theoretical_loss": 3.6930185733763334,
      "tokens_seen": 883359744
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036986960882647946,
      "loss": 2.9399,
      "theoretical_loss": 3.6929918995136193,
      "tokens_seen": 883425280
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036985957873620864,
      "loss": 3.0823,
      "theoretical_loss": 3.6929652281836143,
      "tokens_seen": 883490816
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003698495486459378,
      "loss": 2.9348,
      "theoretical_loss": 3.692938559385891,
      "tokens_seen": 883556352
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.000369839518555667,
      "loss": 3.0963,
      "theoretical_loss": 3.69291189312002,
      "tokens_seen": 883621888
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036982948846539624,
      "loss": 2.5705,
      "theoretical_loss": 3.6928852293855745,
      "tokens_seen": 883687424
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036981945837512536,
      "loss": 2.9487,
      "theoretical_loss": 3.692858568182125,
      "tokens_seen": 883752960
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003698094282848546,
      "loss": 2.9384,
      "theoretical_loss": 3.692831909509245,
      "tokens_seen": 883818496
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003697993981945837,
      "loss": 2.7479,
      "theoretical_loss": 3.6928052533665054,
      "tokens_seen": 883884032
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036978936810431296,
      "loss": 2.8602,
      "theoretical_loss": 3.6927785997534794,
      "tokens_seen": 883949568
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036977933801404214,
      "loss": 3.0335,
      "theoretical_loss": 3.6927519486697395,
      "tokens_seen": 884015104
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003697693079237713,
      "loss": 3.1212,
      "theoretical_loss": 3.6927253001148577,
      "tokens_seen": 884080640
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003697592778335005,
      "loss": 2.8909,
      "theoretical_loss": 3.6926986540884066,
      "tokens_seen": 884146176
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003697492477432297,
      "loss": 3.1543,
      "theoretical_loss": 3.6926720105899586,
      "tokens_seen": 884211712
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036973921765295887,
      "loss": 2.8797,
      "theoretical_loss": 3.6926453696190875,
      "tokens_seen": 884277248
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003697291875626881,
      "loss": 3.1012,
      "theoretical_loss": 3.6926187311753655,
      "tokens_seen": 884342784
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036971915747241723,
      "loss": 2.9521,
      "theoretical_loss": 3.692592095258366,
      "tokens_seen": 884408320
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036970912738214646,
      "loss": 2.8208,
      "theoretical_loss": 3.692565461867662,
      "tokens_seen": 884473856
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036969909729187565,
      "loss": 3.1468,
      "theoretical_loss": 3.6925388310028264,
      "tokens_seen": 884539392
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003696890672016048,
      "loss": 2.8268,
      "theoretical_loss": 3.692512202663433,
      "tokens_seen": 884604928
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.000369679037111334,
      "loss": 3.073,
      "theoretical_loss": 3.692485576849055,
      "tokens_seen": 884670464
    },
    {
      "epoch": 0.27,
      "objective/train/advantage_avg": 0.49534934759140015,
      "objective/train/docs_used": 507654,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.5954573154449463,
      "objective/train/original_loss": 2.5954573154449463,
      "objective/train/theoretical_loss": 3.6924589535592656,
      "objective/train/tokens_used": 905196000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2454085797071457,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507824420928955,
      "objective/train/weighted_lm_loss": 2.7270560264587402,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0457526445388794,
      "theoretical_loss": 3.6924589535592656,
      "tokens_seen": 884736000
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003696690070210632,
      "loss": 2.8778,
      "theoretical_loss": 3.6924589535592656,
      "tokens_seen": 884736000
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036965897693079237,
      "loss": 2.7861,
      "theoretical_loss": 3.6924323327936395,
      "tokens_seen": 884801536
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003696489468405216,
      "loss": 2.8943,
      "theoretical_loss": 3.6924057145517493,
      "tokens_seen": 884867072
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036963891675025073,
      "loss": 2.9681,
      "theoretical_loss": 3.6923790988331695,
      "tokens_seen": 884932608
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036962888665997997,
      "loss": 2.9631,
      "theoretical_loss": 3.692352485637474,
      "tokens_seen": 884998144
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003696188565697091,
      "loss": 3.2253,
      "theoretical_loss": 3.6923258749642365,
      "tokens_seen": 885063680
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036960882647943833,
      "loss": 2.7017,
      "theoretical_loss": 3.6922992668130314,
      "tokens_seen": 885129216
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003695987963891675,
      "loss": 3.0747,
      "theoretical_loss": 3.6922726611834333,
      "tokens_seen": 885194752
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003695887662988967,
      "loss": 3.2079,
      "theoretical_loss": 3.692246058075016,
      "tokens_seen": 885260288
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036957873620862587,
      "loss": 2.9175,
      "theoretical_loss": 3.6922194574873544,
      "tokens_seen": 885325824
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036956870611835505,
      "loss": 3.2133,
      "theoretical_loss": 3.6921928594200226,
      "tokens_seen": 885391360
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036955867602808424,
      "loss": 2.985,
      "theoretical_loss": 3.6921662638725956,
      "tokens_seen": 885456896
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036954864593781347,
      "loss": 3.252,
      "theoretical_loss": 3.6921396708446483,
      "tokens_seen": 885522432
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003695386158475426,
      "loss": 3.15,
      "theoretical_loss": 3.692113080335755,
      "tokens_seen": 885587968
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036952858575727183,
      "loss": 2.8854,
      "theoretical_loss": 3.6920864923454917,
      "tokens_seen": 885653504
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.000369518555667001,
      "loss": 3.049,
      "theoretical_loss": 3.6920599068734328,
      "tokens_seen": 885719040
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003695085255767302,
      "loss": 2.8988,
      "theoretical_loss": 3.6920333239191536,
      "tokens_seen": 885784576
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003694984954864594,
      "loss": 2.9606,
      "theoretical_loss": 3.6920067434822297,
      "tokens_seen": 885850112
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036948846539618856,
      "loss": 2.9029,
      "theoretical_loss": 3.6919801655622355,
      "tokens_seen": 885915648
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036947843530591774,
      "loss": 2.7832,
      "theoretical_loss": 3.6919535901587475,
      "tokens_seen": 885981184
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.000369468405215647,
      "loss": 2.8829,
      "theoretical_loss": 3.6919270172713414,
      "tokens_seen": 886046720
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003694583751253761,
      "loss": 2.6638,
      "theoretical_loss": 3.6919004468995924,
      "tokens_seen": 886112256
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036944834503510534,
      "loss": 3.0863,
      "theoretical_loss": 3.6918738790430763,
      "tokens_seen": 886177792
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036943831494483446,
      "loss": 2.8947,
      "theoretical_loss": 3.691847313701369,
      "tokens_seen": 886243328
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003694282848545637,
      "loss": 3.0557,
      "theoretical_loss": 3.6918207508740473,
      "tokens_seen": 886308864
    },
    {
      "epoch": 0.27,
      "objective/train/advantage_avg": 0.49451756477355957,
      "objective/train/docs_used": 509172,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.002779245376587,
      "objective/train/original_loss": 3.0027787685394287,
      "objective/train/theoretical_loss": 3.691794190560686,
      "objective/train/tokens_used": 906834400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24471968412399292,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050695776939392,
      "objective/train/weighted_lm_loss": 3.1549060344696045,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 0.951229453086853,
      "theoretical_loss": 3.691794190560686,
      "tokens_seen": 886374400
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003694182547642929,
      "loss": 2.927,
      "theoretical_loss": 3.691794190560686,
      "tokens_seen": 886374400
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036940822467402206,
      "loss": 3.0597,
      "theoretical_loss": 3.691767632760863,
      "tokens_seen": 886439936
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036939819458375124,
      "loss": 2.8893,
      "theoretical_loss": 3.6917410774741524,
      "tokens_seen": 886505472
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003693881644934805,
      "loss": 2.9851,
      "theoretical_loss": 3.6917145247001324,
      "tokens_seen": 886571008
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003693781344032096,
      "loss": 2.9611,
      "theoretical_loss": 3.6916879744383793,
      "tokens_seen": 886636544
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036936810431293884,
      "loss": 3.07,
      "theoretical_loss": 3.6916614266884693,
      "tokens_seen": 886702080
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.000369358074222668,
      "loss": 3.1775,
      "theoretical_loss": 3.6916348814499793,
      "tokens_seen": 886767616
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003693480441323972,
      "loss": 3.0529,
      "theoretical_loss": 3.691608338722486,
      "tokens_seen": 886833152
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036933801404212644,
      "loss": 3.067,
      "theoretical_loss": 3.691581798505567,
      "tokens_seen": 886898688
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036932798395185556,
      "loss": 2.9261,
      "theoretical_loss": 3.6915552607987983,
      "tokens_seen": 886964224
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003693179538615848,
      "loss": 3.1403,
      "theoretical_loss": 3.691528725601758,
      "tokens_seen": 887029760
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003693079237713139,
      "loss": 2.7902,
      "theoretical_loss": 3.6915021929140224,
      "tokens_seen": 887095296
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036929789368104316,
      "loss": 3.1525,
      "theoretical_loss": 3.6914756627351695,
      "tokens_seen": 887160832
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036928786359077234,
      "loss": 3.0588,
      "theoretical_loss": 3.6914491350647767,
      "tokens_seen": 887226368
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003692778335005015,
      "loss": 2.7855,
      "theoretical_loss": 3.6914226099024217,
      "tokens_seen": 887291904
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003692678034102307,
      "loss": 3.2115,
      "theoretical_loss": 3.691396087247682,
      "tokens_seen": 887357440
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003692577733199599,
      "loss": 2.8642,
      "theoretical_loss": 3.691369567100135,
      "tokens_seen": 887422976
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036924774322968907,
      "loss": 2.4773,
      "theoretical_loss": 3.691343049459359,
      "tokens_seen": 887488512
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003692377131394183,
      "loss": 2.8789,
      "theoretical_loss": 3.6913165343249323,
      "tokens_seen": 887554048
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036922768304914743,
      "loss": 2.7387,
      "theoretical_loss": 3.6912900216964317,
      "tokens_seen": 887619584
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036921765295887666,
      "loss": 3.0029,
      "theoretical_loss": 3.6912635115734367,
      "tokens_seen": 887685120
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036920762286860585,
      "loss": 2.7336,
      "theoretical_loss": 3.691237003955525,
      "tokens_seen": 887750656
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.000369197592778335,
      "loss": 2.9752,
      "theoretical_loss": 3.691210498842275,
      "tokens_seen": 887816192
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003691875626880642,
      "loss": 2.8015,
      "theoretical_loss": 3.6911839962332653,
      "tokens_seen": 887881728
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003691775325977934,
      "loss": 2.8148,
      "theoretical_loss": 3.691157496128074,
      "tokens_seen": 887947264
    },
    {
      "epoch": 0.27,
      "objective/train/advantage_avg": 0.49514010548591614,
      "objective/train/docs_used": 509747,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.801408290863037,
      "objective/train/original_loss": 2.801408290863037,
      "objective/train/theoretical_loss": 3.691130998526281,
      "objective/train/tokens_used": 908472800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24524900317192078,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507607460021973,
      "objective/train/weighted_lm_loss": 2.9437036514282227,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 3.691130998526281,
      "tokens_seen": 888012800
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036916750250752257,
      "loss": 2.7345,
      "theoretical_loss": 3.691130998526281,
      "tokens_seen": 888012800
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003691574724172518,
      "loss": 2.6703,
      "theoretical_loss": 3.6911045034274634,
      "tokens_seen": 888078336
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036914744232698093,
      "loss": 3.0458,
      "theoretical_loss": 3.691078010831202,
      "tokens_seen": 888143872
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036913741223671017,
      "loss": 2.7205,
      "theoretical_loss": 3.6910515207370738,
      "tokens_seen": 888209408
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003691273821464393,
      "loss": 2.9366,
      "theoretical_loss": 3.6910250331446592,
      "tokens_seen": 888274944
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036911735205616853,
      "loss": 3.0614,
      "theoretical_loss": 3.6909985480535368,
      "tokens_seen": 888340480
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003691073219658977,
      "loss": 3.0631,
      "theoretical_loss": 3.6909720654632863,
      "tokens_seen": 888406016
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003690972918756269,
      "loss": 2.9872,
      "theoretical_loss": 3.6909455853734867,
      "tokens_seen": 888471552
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003690872617853561,
      "loss": 2.8744,
      "theoretical_loss": 3.690919107783718,
      "tokens_seen": 888537088
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036907723169508525,
      "loss": 3.1328,
      "theoretical_loss": 3.6908926326935596,
      "tokens_seen": 888602624
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036906720160481444,
      "loss": 2.8092,
      "theoretical_loss": 3.6908661601025905,
      "tokens_seen": 888668160
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036905717151454367,
      "loss": 2.8244,
      "theoretical_loss": 3.6908396900103915,
      "tokens_seen": 888733696
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003690471414242728,
      "loss": 2.9983,
      "theoretical_loss": 3.6908132224165424,
      "tokens_seen": 888799232
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036903711133400203,
      "loss": 2.9345,
      "theoretical_loss": 3.6907867573206223,
      "tokens_seen": 888864768
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003690270812437312,
      "loss": 2.9723,
      "theoretical_loss": 3.6907602947222125,
      "tokens_seen": 888930304
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003690170511534604,
      "loss": 3.0555,
      "theoretical_loss": 3.690733834620892,
      "tokens_seen": 888995840
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003690070210631896,
      "loss": 2.9401,
      "theoretical_loss": 3.6907073770162424,
      "tokens_seen": 889061376
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036899699097291876,
      "loss": 3.1359,
      "theoretical_loss": 3.690680921907843,
      "tokens_seen": 889126912
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036898696088264794,
      "loss": 2.8812,
      "theoretical_loss": 3.690654469295275,
      "tokens_seen": 889192448
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003689769307923772,
      "loss": 2.7411,
      "theoretical_loss": 3.690628019178119,
      "tokens_seen": 889257984
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003689669007021063,
      "loss": 3.0527,
      "theoretical_loss": 3.6906015715559555,
      "tokens_seen": 889323520
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036895687061183554,
      "loss": 2.8442,
      "theoretical_loss": 3.6905751264283646,
      "tokens_seen": 889389056
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036894684052156466,
      "loss": 2.9925,
      "theoretical_loss": 3.6905486837949284,
      "tokens_seen": 889454592
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003689368104312939,
      "loss": 2.9543,
      "theoretical_loss": 3.690522243655227,
      "tokens_seen": 889520128
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003689267803410231,
      "loss": 2.7227,
      "theoretical_loss": 3.690495806008843,
      "tokens_seen": 889585664
    },
    {
      "epoch": 0.27,
      "objective/train/advantage_avg": 0.49474120140075684,
      "objective/train/docs_used": 511200,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.821974277496338,
      "objective/train/original_loss": 2.821974992752075,
      "objective/train/theoretical_loss": 3.690469370855356,
      "objective/train/tokens_used": 910111200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24483764171600342,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507186651229858,
      "objective/train/weighted_lm_loss": 2.9650237560272217,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0417574644088745,
      "theoretical_loss": 3.690469370855356,
      "tokens_seen": 889651200
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036891675025075226,
      "loss": 2.9158,
      "theoretical_loss": 3.690469370855356,
      "tokens_seen": 889651200
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036890672016048144,
      "loss": 2.9946,
      "theoretical_loss": 3.690442938194348,
      "tokens_seen": 889716736
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003688966900702107,
      "loss": 2.8353,
      "theoretical_loss": 3.6904165080254003,
      "tokens_seen": 889782272
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003688866599799398,
      "loss": 2.8684,
      "theoretical_loss": 3.6903900803480942,
      "tokens_seen": 889847808
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036887662988966904,
      "loss": 2.9746,
      "theoretical_loss": 3.690363655162012,
      "tokens_seen": 889913344
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036886659979939817,
      "loss": 2.885,
      "theoretical_loss": 3.6903372324667347,
      "tokens_seen": 889978880
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003688565697091274,
      "loss": 2.8641,
      "theoretical_loss": 3.6903108122618447,
      "tokens_seen": 890044416
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003688465396188566,
      "loss": 3.0146,
      "theoretical_loss": 3.6902843945469233,
      "tokens_seen": 890109952
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036883650952858576,
      "loss": 2.9642,
      "theoretical_loss": 3.6902579793215535,
      "tokens_seen": 890175488
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036882647943831494,
      "loss": 3.0415,
      "theoretical_loss": 3.6902315665853163,
      "tokens_seen": 890241024
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003688164493480441,
      "loss": 3.0666,
      "theoretical_loss": 3.690205156337795,
      "tokens_seen": 890306560
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003688064192577733,
      "loss": 2.9706,
      "theoretical_loss": 3.6901787485785706,
      "tokens_seen": 890372096
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036879638916750254,
      "loss": 2.8046,
      "theoretical_loss": 3.690152343307227,
      "tokens_seen": 890437632
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036878635907723167,
      "loss": 2.9403,
      "theoretical_loss": 3.690125940523346,
      "tokens_seen": 890503168
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003687763289869609,
      "loss": 2.8262,
      "theoretical_loss": 3.6900995402265098,
      "tokens_seen": 890568704
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036876629889669003,
      "loss": 2.8447,
      "theoretical_loss": 3.6900731424163027,
      "tokens_seen": 890634240
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036875626880641927,
      "loss": 2.9335,
      "theoretical_loss": 3.690046747092305,
      "tokens_seen": 890699776
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036874623871614845,
      "loss": 3.1855,
      "theoretical_loss": 3.690020354254102,
      "tokens_seen": 890765312
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036873620862587763,
      "loss": 3.0105,
      "theoretical_loss": 3.6899939639012764,
      "tokens_seen": 890830848
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003687261785356068,
      "loss": 2.8932,
      "theoretical_loss": 3.68996757603341,
      "tokens_seen": 890896384
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036871614844533605,
      "loss": 2.664,
      "theoretical_loss": 3.689941190650087,
      "tokens_seen": 890961920
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036870611835506517,
      "loss": 2.8968,
      "theoretical_loss": 3.689914807750891,
      "tokens_seen": 891027456
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003686960882647944,
      "loss": 3.0381,
      "theoretical_loss": 3.6898884273354042,
      "tokens_seen": 891092992
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036868605817452353,
      "loss": 2.9589,
      "theoretical_loss": 3.689862049403212,
      "tokens_seen": 891158528
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036867602808425277,
      "loss": 2.9333,
      "theoretical_loss": 3.689835673953896,
      "tokens_seen": 891224064
    },
    {
      "epoch": 0.27,
      "objective/train/advantage_avg": 0.49371540546417236,
      "objective/train/docs_used": 511885,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.7282259464263916,
      "objective/train/original_loss": 2.7282257080078125,
      "objective/train/theoretical_loss": 3.689809300987042,
      "objective/train/tokens_used": 911749600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2438235580921173,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506110191345215,
      "objective/train/weighted_lm_loss": 2.8662567138671875,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0413066148757935,
      "theoretical_loss": 3.689809300987042,
      "tokens_seen": 891289600
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036866599799398195,
      "loss": 2.8769,
      "theoretical_loss": 3.689809300987042,
      "tokens_seen": 891289600
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036865596790371113,
      "loss": 2.8412,
      "theoretical_loss": 3.689782930502232,
      "tokens_seen": 891355136
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003686459378134403,
      "loss": 2.963,
      "theoretical_loss": 3.6897565624990505,
      "tokens_seen": 891420672
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003686359077231695,
      "loss": 3.1011,
      "theoretical_loss": 3.6897301969770817,
      "tokens_seen": 891486208
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003686258776328987,
      "loss": 3.2003,
      "theoretical_loss": 3.6897038339359107,
      "tokens_seen": 891551744
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003686158475426279,
      "loss": 3.1082,
      "theoretical_loss": 3.6896774733751205,
      "tokens_seen": 891617280
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003686058174523571,
      "loss": 3.1414,
      "theoretical_loss": 3.6896511152942955,
      "tokens_seen": 891682816
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003685957873620863,
      "loss": 3.0401,
      "theoretical_loss": 3.6896247596930207,
      "tokens_seen": 891748352
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036858575727181545,
      "loss": 3.0298,
      "theoretical_loss": 3.6895984065708807,
      "tokens_seen": 891813888
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036857572718154464,
      "loss": 3.049,
      "theoretical_loss": 3.689572055927459,
      "tokens_seen": 891879424
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036856569709127387,
      "loss": 3.0523,
      "theoretical_loss": 3.689545707762342,
      "tokens_seen": 891944960
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.000368555667001003,
      "loss": 2.8745,
      "theoretical_loss": 3.689519362075113,
      "tokens_seen": 892010496
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036854563691073223,
      "loss": 3.2373,
      "theoretical_loss": 3.6894930188653587,
      "tokens_seen": 892076032
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003685356068204614,
      "loss": 2.9797,
      "theoretical_loss": 3.6894666781326624,
      "tokens_seen": 892141568
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003685255767301906,
      "loss": 2.8351,
      "theoretical_loss": 3.68944033987661,
      "tokens_seen": 892207104
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003685155466399198,
      "loss": 2.6343,
      "theoretical_loss": 3.6894140040967867,
      "tokens_seen": 892272640
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036850551654964896,
      "loss": 2.7467,
      "theoretical_loss": 3.6893876707927777,
      "tokens_seen": 892338176
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036849548645937814,
      "loss": 2.965,
      "theoretical_loss": 3.6893613399641687,
      "tokens_seen": 892403712
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003684854563691074,
      "loss": 3.0186,
      "theoretical_loss": 3.6893350116105452,
      "tokens_seen": 892469248
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003684754262788365,
      "loss": 3.0198,
      "theoretical_loss": 3.689308685731493,
      "tokens_seen": 892534784
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036846539618856574,
      "loss": 2.9473,
      "theoretical_loss": 3.6892823623265976,
      "tokens_seen": 892600320
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036845536609829486,
      "loss": 2.8156,
      "theoretical_loss": 3.6892560413954447,
      "tokens_seen": 892665856
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003684453360080241,
      "loss": 2.7193,
      "theoretical_loss": 3.68922972293762,
      "tokens_seen": 892731392
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003684353059177533,
      "loss": 2.9783,
      "theoretical_loss": 3.6892034069527107,
      "tokens_seen": 892796928
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036842527582748246,
      "loss": 2.6665,
      "theoretical_loss": 3.6891770934403016,
      "tokens_seen": 892862464
    },
    {
      "epoch": 0.27,
      "objective/train/advantage_avg": 0.4956626892089844,
      "objective/train/docs_used": 512575,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.4313130378723145,
      "objective/train/original_loss": 2.4313130378723145,
      "objective/train/theoretical_loss": 3.68915078239998,
      "objective/train/tokens_used": 913388000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2457219958305359,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0508153438568115,
      "objective/train/weighted_lm_loss": 2.5549211502075195,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0470744371414185,
      "theoretical_loss": 3.68915078239998,
      "tokens_seen": 892928000
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036841524573721164,
      "loss": 2.847,
      "theoretical_loss": 3.68915078239998,
      "tokens_seen": 892928000
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003684052156469409,
      "loss": 3.0348,
      "theoretical_loss": 3.689124473831331,
      "tokens_seen": 892993536
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036839518555667,
      "loss": 2.9651,
      "theoretical_loss": 3.6890981677339427,
      "tokens_seen": 893059072
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036838515546639924,
      "loss": 3.0389,
      "theoretical_loss": 3.6890718641074005,
      "tokens_seen": 893124608
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036837512537612837,
      "loss": 2.9082,
      "theoretical_loss": 3.689045562951291,
      "tokens_seen": 893190144
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003683650952858576,
      "loss": 2.9648,
      "theoretical_loss": 3.6890192642652018,
      "tokens_seen": 893255680
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003683550651955868,
      "loss": 2.8746,
      "theoretical_loss": 3.688992968048719,
      "tokens_seen": 893321216
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036834503510531596,
      "loss": 2.855,
      "theoretical_loss": 3.6889666743014295,
      "tokens_seen": 893386752
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036833500501504514,
      "loss": 2.9518,
      "theoretical_loss": 3.6889403830229206,
      "tokens_seen": 893452288
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003683249749247743,
      "loss": 2.9328,
      "theoretical_loss": 3.6889140942127794,
      "tokens_seen": 893517824
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003683149448345035,
      "loss": 3.0877,
      "theoretical_loss": 3.688887807870593,
      "tokens_seen": 893583360
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036830491474423274,
      "loss": 3.0253,
      "theoretical_loss": 3.688861523995949,
      "tokens_seen": 893648896
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036829488465396187,
      "loss": 2.8138,
      "theoretical_loss": 3.6888352425884348,
      "tokens_seen": 893714432
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003682848545636911,
      "loss": 3.0495,
      "theoretical_loss": 3.6888089636476376,
      "tokens_seen": 893779968
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036827482447342023,
      "loss": 2.815,
      "theoretical_loss": 3.6887826871731453,
      "tokens_seen": 893845504
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036826479438314947,
      "loss": 3.0118,
      "theoretical_loss": 3.688756413164546,
      "tokens_seen": 893911040
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036825476429287865,
      "loss": 3.0493,
      "theoretical_loss": 3.688730141621427,
      "tokens_seen": 893976576
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036824473420260783,
      "loss": 3.019,
      "theoretical_loss": 3.688703872543376,
      "tokens_seen": 894042112
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.000368234704112337,
      "loss": 3.0039,
      "theoretical_loss": 3.6886776059299815,
      "tokens_seen": 894107648
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036822467402206625,
      "loss": 2.8026,
      "theoretical_loss": 3.6886513417808318,
      "tokens_seen": 894173184
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036821464393179537,
      "loss": 3.0984,
      "theoretical_loss": 3.6886250800955143,
      "tokens_seen": 894238720
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003682046138415246,
      "loss": 2.8399,
      "theoretical_loss": 3.6885988208736187,
      "tokens_seen": 894304256
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036819458375125373,
      "loss": 3.0648,
      "theoretical_loss": 3.6885725641147316,
      "tokens_seen": 894369792
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036818455366098297,
      "loss": 2.9342,
      "theoretical_loss": 3.6885463098184434,
      "tokens_seen": 894435328
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036817452357071215,
      "loss": 3.0271,
      "theoretical_loss": 3.6885200579843413,
      "tokens_seen": 894500864
    },
    {
      "epoch": 0.27,
      "objective/train/advantage_avg": 0.4927689731121063,
      "objective/train/docs_used": 513630,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.712967872619629,
      "objective/train/original_loss": 2.712967872619629,
      "objective/train/theoretical_loss": 3.688493808612015,
      "objective/train/tokens_used": 915026400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2428998202085495,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505115985870361,
      "objective/train/weighted_lm_loss": 2.8497402667999268,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 3.688493808612015,
      "tokens_seen": 894566400
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036816449348044133,
      "loss": 2.9487,
      "theoretical_loss": 3.688493808612015,
      "tokens_seen": 894566400
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003681544633901705,
      "loss": 2.8313,
      "theoretical_loss": 3.688467561701053,
      "tokens_seen": 894631936
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003681444332998997,
      "loss": 3.0517,
      "theoretical_loss": 3.6884413172510437,
      "tokens_seen": 894697472
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003681344032096289,
      "loss": 2.7169,
      "theoretical_loss": 3.6884150752615765,
      "tokens_seen": 894763008
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003681243731193581,
      "loss": 2.8047,
      "theoretical_loss": 3.688388835732241,
      "tokens_seen": 894828544
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036811434302908724,
      "loss": 3.0133,
      "theoretical_loss": 3.688362598662626,
      "tokens_seen": 894894080
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003681043129388165,
      "loss": 2.6101,
      "theoretical_loss": 3.6883363640523203,
      "tokens_seen": 894959616
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003680942828485456,
      "loss": 2.7652,
      "theoretical_loss": 3.6883101319009137,
      "tokens_seen": 895025152
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036808425275827484,
      "loss": 2.7118,
      "theoretical_loss": 3.6882839022079965,
      "tokens_seen": 895090688
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.000368074222668004,
      "loss": 2.9801,
      "theoretical_loss": 3.6882576749731575,
      "tokens_seen": 895156224
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003680641925777332,
      "loss": 2.8796,
      "theoretical_loss": 3.6882314501959863,
      "tokens_seen": 895221760
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003680541624874624,
      "loss": 3.1061,
      "theoretical_loss": 3.6882052278760726,
      "tokens_seen": 895287296
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003680441323971916,
      "loss": 2.9276,
      "theoretical_loss": 3.688179008013007,
      "tokens_seen": 895352832
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036803410230692074,
      "loss": 3.1316,
      "theoretical_loss": 3.688152790606379,
      "tokens_seen": 895418368
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036802407221665,
      "loss": 3.0489,
      "theoretical_loss": 3.6881265756557795,
      "tokens_seen": 895483904
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003680140421263791,
      "loss": 2.8952,
      "theoretical_loss": 3.688100363160797,
      "tokens_seen": 895549440
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036800401203610834,
      "loss": 2.9859,
      "theoretical_loss": 3.6880741531210237,
      "tokens_seen": 895614976
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003679939819458375,
      "loss": 2.8729,
      "theoretical_loss": 3.6880479455360486,
      "tokens_seen": 895680512
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003679839518555667,
      "loss": 2.9959,
      "theoretical_loss": 3.688021740405463,
      "tokens_seen": 895746048
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003679739217652959,
      "loss": 2.8307,
      "theoretical_loss": 3.687995537728858,
      "tokens_seen": 895811584
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036796389167502506,
      "loss": 2.9749,
      "theoretical_loss": 3.6879693375058222,
      "tokens_seen": 895877120
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036795386158475424,
      "loss": 2.9006,
      "theoretical_loss": 3.687943139735949,
      "tokens_seen": 895942656
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003679438314944835,
      "loss": 3.086,
      "theoretical_loss": 3.687916944418827,
      "tokens_seen": 896008192
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003679338014042126,
      "loss": 2.9908,
      "theoretical_loss": 3.6878907515540487,
      "tokens_seen": 896073728
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036792377131394184,
      "loss": 3.1075,
      "theoretical_loss": 3.687864561141205,
      "tokens_seen": 896139264
    },
    {
      "epoch": 0.27,
      "objective/train/advantage_avg": 0.4947876036167145,
      "objective/train/docs_used": 514378,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.545466899871826,
      "objective/train/original_loss": 2.545466423034668,
      "objective/train/theoretical_loss": 3.6878383731798863,
      "objective/train/tokens_used": 916664800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24487534165382385,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050723671913147,
      "objective/train/weighted_lm_loss": 2.674473524093628,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 3.6878383731798863,
      "tokens_seen": 896204800
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036791374122367097,
      "loss": 2.759,
      "theoretical_loss": 3.6878383731798863,
      "tokens_seen": 896204800
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003679037111334002,
      "loss": 3.0442,
      "theoretical_loss": 3.6878121876696843,
      "tokens_seen": 896270336
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003678936810431294,
      "loss": 3.1057,
      "theoretical_loss": 3.6877860046101905,
      "tokens_seen": 896335872
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036788365095285857,
      "loss": 3.2335,
      "theoretical_loss": 3.6877598240009966,
      "tokens_seen": 896401408
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036787362086258775,
      "loss": 3.124,
      "theoretical_loss": 3.6877336458416936,
      "tokens_seen": 896466944
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.000367863590772317,
      "loss": 2.9468,
      "theoretical_loss": 3.6877074701318735,
      "tokens_seen": 896532480
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036785356068204616,
      "loss": 2.8964,
      "theoretical_loss": 3.6876812968711286,
      "tokens_seen": 896598016
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036784353059177535,
      "loss": 2.9479,
      "theoretical_loss": 3.6876551260590498,
      "tokens_seen": 896663552
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003678335005015045,
      "loss": 3.0101,
      "theoretical_loss": 3.6876289576952295,
      "tokens_seen": 896729088
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003678234704112337,
      "loss": 2.7087,
      "theoretical_loss": 3.68760279177926,
      "tokens_seen": 896794624
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036781344032096294,
      "loss": 2.8869,
      "theoretical_loss": 3.6875766283107327,
      "tokens_seen": 896860160
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036780341023069207,
      "loss": 2.9924,
      "theoretical_loss": 3.6875504672892405,
      "tokens_seen": 896925696
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003677933801404213,
      "loss": 3.0237,
      "theoretical_loss": 3.6875243087143756,
      "tokens_seen": 896991232
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036778335005015043,
      "loss": 2.9288,
      "theoretical_loss": 3.687498152585731,
      "tokens_seen": 897056768
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036777331995987967,
      "loss": 3.0948,
      "theoretical_loss": 3.687471998902898,
      "tokens_seen": 897122304
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036776328986960885,
      "loss": 2.8288,
      "theoretical_loss": 3.6874458476654706,
      "tokens_seen": 897187840
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036775325977933803,
      "loss": 2.8552,
      "theoretical_loss": 3.6874196988730406,
      "tokens_seen": 897253376
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003677432296890672,
      "loss": 2.8827,
      "theoretical_loss": 3.6873935525252017,
      "tokens_seen": 897318912
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036773319959879645,
      "loss": 2.8997,
      "theoretical_loss": 3.6873674086215456,
      "tokens_seen": 897384448
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036772316950852557,
      "loss": 2.9356,
      "theoretical_loss": 3.687341267161667,
      "tokens_seen": 897449984
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003677131394182548,
      "loss": 3.1007,
      "theoretical_loss": 3.6873151281451575,
      "tokens_seen": 897515520
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036770310932798393,
      "loss": 2.9797,
      "theoretical_loss": 3.6872889915716107,
      "tokens_seen": 897581056
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036769307923771317,
      "loss": 3.1369,
      "theoretical_loss": 3.6872628574406208,
      "tokens_seen": 897646592
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036768304914744235,
      "loss": 2.756,
      "theoretical_loss": 3.68723672575178,
      "tokens_seen": 897712128
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036767301905717153,
      "loss": 3.195,
      "theoretical_loss": 3.687210596504683,
      "tokens_seen": 897777664
    },
    {
      "epoch": 0.27,
      "objective/train/advantage_avg": 0.494235634803772,
      "objective/train/docs_used": 515835,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.0064141750335693,
      "objective/train/original_loss": 3.0064146518707275,
      "objective/train/theoretical_loss": 3.6871844696989227,
      "objective/train/tokens_used": 918303200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24436835944652557,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506657361984253,
      "objective/train/weighted_lm_loss": 3.1586427688598633,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0363596677780151,
      "theoretical_loss": 3.6871844696989227,
      "tokens_seen": 897843200
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003676629889669007,
      "loss": 3.0208,
      "theoretical_loss": 3.6871844696989227,
      "tokens_seen": 897843200
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003676529588766299,
      "loss": 2.888,
      "theoretical_loss": 3.687158345334093,
      "tokens_seen": 897908736
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003676429287863591,
      "loss": 2.8789,
      "theoretical_loss": 3.6871322234097876,
      "tokens_seen": 897974272
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003676328986960883,
      "loss": 3.2821,
      "theoretical_loss": 3.6871061039256006,
      "tokens_seen": 898039808
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036762286860581744,
      "loss": 2.9308,
      "theoretical_loss": 3.687079986881126,
      "tokens_seen": 898105344
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003676128385155467,
      "loss": 3.0588,
      "theoretical_loss": 3.687053872275958,
      "tokens_seen": 898170880
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003676028084252758,
      "loss": 2.9126,
      "theoretical_loss": 3.6870277601096904,
      "tokens_seen": 898236416
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036759277833500504,
      "loss": 2.7559,
      "theoretical_loss": 3.6870016503819176,
      "tokens_seen": 898301952
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003675827482447342,
      "loss": 2.9109,
      "theoretical_loss": 3.6869755430922346,
      "tokens_seen": 898367488
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003675727181544634,
      "loss": 2.8487,
      "theoretical_loss": 3.6869494382402355,
      "tokens_seen": 898433024
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003675626880641926,
      "loss": 2.9935,
      "theoretical_loss": 3.6869233358255147,
      "tokens_seen": 898498560
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003675526579739218,
      "loss": 2.8553,
      "theoretical_loss": 3.686897235847667,
      "tokens_seen": 898564096
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036754262788365094,
      "loss": 2.9852,
      "theoretical_loss": 3.6868711383062873,
      "tokens_seen": 898629632
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003675325977933802,
      "loss": 2.925,
      "theoretical_loss": 3.6868450432009707,
      "tokens_seen": 898695168
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003675225677031093,
      "loss": 2.806,
      "theoretical_loss": 3.6868189505313116,
      "tokens_seen": 898760704
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036751253761283854,
      "loss": 2.6761,
      "theoretical_loss": 3.686792860296906,
      "tokens_seen": 898826240
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003675025075225677,
      "loss": 2.8025,
      "theoretical_loss": 3.686766772497348,
      "tokens_seen": 898891776
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003674924774322969,
      "loss": 3.0253,
      "theoretical_loss": 3.6867406871322332,
      "tokens_seen": 898957312
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003674824473420261,
      "loss": 2.9346,
      "theoretical_loss": 3.686714604201158,
      "tokens_seen": 899022848
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036747241725175526,
      "loss": 3.0259,
      "theoretical_loss": 3.686688523703716,
      "tokens_seen": 899088384
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036746238716148444,
      "loss": 2.8034,
      "theoretical_loss": 3.686662445639504,
      "tokens_seen": 899153920
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003674523570712137,
      "loss": 2.8469,
      "theoretical_loss": 3.686636370008118,
      "tokens_seen": 899219456
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003674423269809428,
      "loss": 3.0735,
      "theoretical_loss": 3.686610296809153,
      "tokens_seen": 899284992
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036743229689067204,
      "loss": 3.0069,
      "theoretical_loss": 3.6865842260422053,
      "tokens_seen": 899350528
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036742226680040117,
      "loss": 2.826,
      "theoretical_loss": 3.68655815770687,
      "tokens_seen": 899416064
    },
    {
      "epoch": 0.27,
      "objective/train/advantage_avg": 0.4945004880428314,
      "objective/train/docs_used": 516375,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.016789197921753,
      "objective/train/original_loss": 3.016789436340332,
      "objective/train/theoretical_loss": 3.6865320918027438,
      "objective/train/tokens_used": 919941600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24459737539291382,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050693392753601,
      "objective/train/weighted_lm_loss": 3.1695141792297363,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0442860126495361,
      "theoretical_loss": 3.6865320918027438,
      "tokens_seen": 899481600
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003674122367101304,
      "loss": 2.9959,
      "theoretical_loss": 3.6865320918027438,
      "tokens_seen": 899481600
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003674022066198596,
      "loss": 2.8623,
      "theoretical_loss": 3.6865060283294238,
      "tokens_seen": 899547136
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036739217652958877,
      "loss": 2.9589,
      "theoretical_loss": 3.686479967286504,
      "tokens_seen": 899612672
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036738214643931795,
      "loss": 2.7366,
      "theoretical_loss": 3.686453908673583,
      "tokens_seen": 899678208
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003673721163490472,
      "loss": 2.7622,
      "theoretical_loss": 3.6864278524902554,
      "tokens_seen": 899743744
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003673620862587763,
      "loss": 2.8608,
      "theoretical_loss": 3.686401798736119,
      "tokens_seen": 899809280
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036735205616850555,
      "loss": 2.8276,
      "theoretical_loss": 3.6863757474107697,
      "tokens_seen": 899874816
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036734202607823467,
      "loss": 3.0254,
      "theoretical_loss": 3.686349698513805,
      "tokens_seen": 899940352
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003673319959879639,
      "loss": 2.7703,
      "theoretical_loss": 3.686323652044821,
      "tokens_seen": 900005888
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003673219658976931,
      "loss": 2.6564,
      "theoretical_loss": 3.686297608003415,
      "tokens_seen": 900071424
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036731193580742227,
      "loss": 3.1279,
      "theoretical_loss": 3.6862715663891836,
      "tokens_seen": 900136960
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036730190571715145,
      "loss": 3.1208,
      "theoretical_loss": 3.6862455272017245,
      "tokens_seen": 900202496
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036729187562688063,
      "loss": 2.8249,
      "theoretical_loss": 3.6862194904406342,
      "tokens_seen": 900268032
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003672818455366098,
      "loss": 2.8107,
      "theoretical_loss": 3.6861934561055105,
      "tokens_seen": 900333568
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036727181544633905,
      "loss": 2.9536,
      "theoretical_loss": 3.6861674241959506,
      "tokens_seen": 900399104
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003672617853560682,
      "loss": 2.763,
      "theoretical_loss": 3.6861413947115524,
      "tokens_seen": 900464640
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003672517552657974,
      "loss": 3.0663,
      "theoretical_loss": 3.686115367651913,
      "tokens_seen": 900530176
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036724172517552654,
      "loss": 2.7867,
      "theoretical_loss": 3.68608934301663,
      "tokens_seen": 900595712
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036723169508525577,
      "loss": 2.8117,
      "theoretical_loss": 3.686063320805302,
      "tokens_seen": 900661248
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036722166499498495,
      "loss": 2.9051,
      "theoretical_loss": 3.6860373010175262,
      "tokens_seen": 900726784
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036721163490471414,
      "loss": 2.7744,
      "theoretical_loss": 3.6860112836529,
      "tokens_seen": 900792320
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003672016048144433,
      "loss": 2.9139,
      "theoretical_loss": 3.685985268711023,
      "tokens_seen": 900857856
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036719157472417255,
      "loss": 2.8895,
      "theoretical_loss": 3.685959256191492,
      "tokens_seen": 900923392
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003671815446339017,
      "loss": 3.0633,
      "theoretical_loss": 3.685933246093906,
      "tokens_seen": 900988928
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003671715145436309,
      "loss": 2.7923,
      "theoretical_loss": 3.685907238417863,
      "tokens_seen": 901054464
    },
    {
      "epoch": 0.27,
      "objective/train/advantage_avg": 0.4954161047935486,
      "objective/train/docs_used": 517599,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.1963560581207275,
      "objective/train/original_loss": 3.1963558197021484,
      "objective/train/theoretical_loss": 3.685881233162962,
      "objective/train/tokens_used": 921580000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24553503096103668,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050789713859558,
      "objective/train/weighted_lm_loss": 3.3587193489074707,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0253151655197144,
      "theoretical_loss": 3.685881233162962,
      "tokens_seen": 901120000
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036716148445336004,
      "loss": 3.1374,
      "theoretical_loss": 3.685881233162962,
      "tokens_seen": 901120000
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003671514543630893,
      "loss": 2.9498,
      "theoretical_loss": 3.6858552303288006,
      "tokens_seen": 901185536
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036714142427281846,
      "loss": 3.0601,
      "theoretical_loss": 3.6858292299149786,
      "tokens_seen": 901251072
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036713139418254764,
      "loss": 2.8023,
      "theoretical_loss": 3.685803231921094,
      "tokens_seen": 901316608
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003671213640922768,
      "loss": 3.0195,
      "theoretical_loss": 3.6857772363467447,
      "tokens_seen": 901382144
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.000367111334002006,
      "loss": 3.1878,
      "theoretical_loss": 3.685751243191532,
      "tokens_seen": 901447680
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036710130391173524,
      "loss": 3.205,
      "theoretical_loss": 3.685725252455053,
      "tokens_seen": 901513216
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003670912738214644,
      "loss": 3.0835,
      "theoretical_loss": 3.6856992641369075,
      "tokens_seen": 901578752
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003670812437311936,
      "loss": 2.9534,
      "theoretical_loss": 3.685673278236695,
      "tokens_seen": 901644288
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003670712136409228,
      "loss": 2.9988,
      "theoretical_loss": 3.6856472947540144,
      "tokens_seen": 901709824
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.000367061183550652,
      "loss": 2.8214,
      "theoretical_loss": 3.685621313688465,
      "tokens_seen": 901775360
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036705115346038114,
      "loss": 3.0777,
      "theoretical_loss": 3.685595335039647,
      "tokens_seen": 901840896
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003670411233701104,
      "loss": 3.0579,
      "theoretical_loss": 3.6855693588071587,
      "tokens_seen": 901906432
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003670310932798395,
      "loss": 2.7666,
      "theoretical_loss": 3.685543384990601,
      "tokens_seen": 901971968
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036702106318956874,
      "loss": 2.8287,
      "theoretical_loss": 3.685517413589573,
      "tokens_seen": 902037504
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003670110330992979,
      "loss": 2.8524,
      "theoretical_loss": 3.6854914446036755,
      "tokens_seen": 902103040
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003670010030090271,
      "loss": 2.9008,
      "theoretical_loss": 3.6854654780325076,
      "tokens_seen": 902168576
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003669909729187563,
      "loss": 2.8228,
      "theoretical_loss": 3.6854395138756693,
      "tokens_seen": 902234112
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036698094282848546,
      "loss": 2.8224,
      "theoretical_loss": 3.6854135521327613,
      "tokens_seen": 902299648
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036697091273821464,
      "loss": 2.965,
      "theoretical_loss": 3.6853875928033837,
      "tokens_seen": 902365184
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003669608826479439,
      "loss": 3.0272,
      "theoretical_loss": 3.685361635887137,
      "tokens_seen": 902430720
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.000366950852557673,
      "loss": 2.7054,
      "theoretical_loss": 3.6853356813836218,
      "tokens_seen": 902496256
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036694082246740224,
      "loss": 2.9569,
      "theoretical_loss": 3.6853097292924377,
      "tokens_seen": 902561792
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036693079237713137,
      "loss": 3.1137,
      "theoretical_loss": 3.6852837796131865,
      "tokens_seen": 902627328
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003669207622868606,
      "loss": 3.1587,
      "theoretical_loss": 3.685257832345468,
      "tokens_seen": 902692864
    },
    {
      "epoch": 0.27,
      "objective/train/advantage_avg": 0.4948800802230835,
      "objective/train/docs_used": 518319,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.923860549926758,
      "objective/train/original_loss": 2.923860788345337,
      "objective/train/theoretical_loss": 3.6852318874888836,
      "objective/train/tokens_used": 923218400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24496930837631226,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507333278656006,
      "objective/train/weighted_lm_loss": 3.072321653366089,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0338950157165527,
      "theoretical_loss": 3.6852318874888836,
      "tokens_seen": 902758400
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003669107321965898,
      "loss": 2.854,
      "theoretical_loss": 3.6852318874888836,
      "tokens_seen": 902758400
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036690070210631897,
      "loss": 2.7281,
      "theoretical_loss": 3.6852059450430343,
      "tokens_seen": 902823936
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036689067201604815,
      "loss": 2.9651,
      "theoretical_loss": 3.685180005007521,
      "tokens_seen": 902889472
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003668806419257774,
      "loss": 3.0037,
      "theoretical_loss": 3.6851540673819443,
      "tokens_seen": 902955008
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003668706118355065,
      "loss": 2.9494,
      "theoretical_loss": 3.6851281321659064,
      "tokens_seen": 903020544
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036686058174523575,
      "loss": 2.8016,
      "theoretical_loss": 3.685102199359008,
      "tokens_seen": 903086080
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036685055165496487,
      "loss": 2.9475,
      "theoretical_loss": 3.6850762689608505,
      "tokens_seen": 903151616
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003668405215646941,
      "loss": 2.9574,
      "theoretical_loss": 3.685050340971036,
      "tokens_seen": 903217152
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003668304914744233,
      "loss": 2.9397,
      "theoretical_loss": 3.685024415389165,
      "tokens_seen": 903282688
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036682046138415247,
      "loss": 2.9669,
      "theoretical_loss": 3.6849984922148398,
      "tokens_seen": 903348224
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036681043129388165,
      "loss": 2.7748,
      "theoretical_loss": 3.6849725714476627,
      "tokens_seen": 903413760
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036680040120361083,
      "loss": 2.7447,
      "theoretical_loss": 3.684946653087235,
      "tokens_seen": 903479296
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036679037111334,
      "loss": 2.8398,
      "theoretical_loss": 3.6849207371331585,
      "tokens_seen": 903544832
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036678034102306925,
      "loss": 2.913,
      "theoretical_loss": 3.6848948235850356,
      "tokens_seen": 903610368
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003667703109327984,
      "loss": 3.0157,
      "theoretical_loss": 3.6848689124424685,
      "tokens_seen": 903675904
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003667602808425276,
      "loss": 2.9666,
      "theoretical_loss": 3.6848430037050592,
      "tokens_seen": 903741440
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036675025075225674,
      "loss": 3.0225,
      "theoretical_loss": 3.6848170973724104,
      "tokens_seen": 903806976
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.000366740220661986,
      "loss": 2.9014,
      "theoretical_loss": 3.6847911934441244,
      "tokens_seen": 903872512
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036673019057171515,
      "loss": 2.8875,
      "theoretical_loss": 3.6847652919198035,
      "tokens_seen": 903938048
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036672016048144434,
      "loss": 3.2122,
      "theoretical_loss": 3.6847393927990506,
      "tokens_seen": 904003584
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003667101303911735,
      "loss": 2.9299,
      "theoretical_loss": 3.6847134960814687,
      "tokens_seen": 904069120
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036670010030090275,
      "loss": 2.9891,
      "theoretical_loss": 3.6846876017666603,
      "tokens_seen": 904134656
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003666900702106319,
      "loss": 2.6011,
      "theoretical_loss": 3.684661709854228,
      "tokens_seen": 904200192
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003666800401203611,
      "loss": 3.0032,
      "theoretical_loss": 3.6846358203437752,
      "tokens_seen": 904265728
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036667001003009024,
      "loss": 3.0514,
      "theoretical_loss": 3.6846099332349054,
      "tokens_seen": 904331264
    },
    {
      "epoch": 0.27,
      "objective/train/advantage_avg": 0.4941151440143585,
      "objective/train/docs_used": 519704,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.818493604660034,
      "objective/train/original_loss": 2.818493366241455,
      "objective/train/theoretical_loss": 3.6845840485272205,
      "objective/train/tokens_used": 924856800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2442319542169571,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050653100013733,
      "objective/train/weighted_lm_loss": 2.9612598419189453,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0304546356201172,
      "theoretical_loss": 3.6845840485272205,
      "tokens_seen": 904396800
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003666599799398195,
      "loss": 2.8773,
      "theoretical_loss": 3.6845840485272205,
      "tokens_seen": 904396800
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036664994984954866,
      "loss": 2.778,
      "theoretical_loss": 3.6845581662203255,
      "tokens_seen": 904462336
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036663991975927784,
      "loss": 2.9597,
      "theoretical_loss": 3.6845322863138232,
      "tokens_seen": 904527872
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.000366629889669007,
      "loss": 3.0385,
      "theoretical_loss": 3.6845064088073163,
      "tokens_seen": 904593408
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003666198595787362,
      "loss": 2.829,
      "theoretical_loss": 3.684480533700409,
      "tokens_seen": 904658944
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003666098294884654,
      "loss": 2.8292,
      "theoretical_loss": 3.6844546609927047,
      "tokens_seen": 904724480
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003665997993981946,
      "loss": 3.2381,
      "theoretical_loss": 3.684428790683808,
      "tokens_seen": 904790016
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036658976930792374,
      "loss": 2.8693,
      "theoretical_loss": 3.684402922773322,
      "tokens_seen": 904855552
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.000366579739217653,
      "loss": 2.7366,
      "theoretical_loss": 3.6843770572608507,
      "tokens_seen": 904921088
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036656970912738216,
      "loss": 3.0211,
      "theoretical_loss": 3.6843511941459983,
      "tokens_seen": 904986624
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036655967903711134,
      "loss": 3.028,
      "theoretical_loss": 3.6843253334283688,
      "tokens_seen": 905052160
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003665496489468405,
      "loss": 2.7183,
      "theoretical_loss": 3.684299475107567,
      "tokens_seen": 905117696
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003665396188565697,
      "loss": 2.824,
      "theoretical_loss": 3.684273619183196,
      "tokens_seen": 905183232
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003665295887662989,
      "loss": 3.2726,
      "theoretical_loss": 3.684247765654862,
      "tokens_seen": 905248768
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003665195586760281,
      "loss": 2.9994,
      "theoretical_loss": 3.6842219145221677,
      "tokens_seen": 905314304
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036650952858575725,
      "loss": 3.0568,
      "theoretical_loss": 3.684196065784719,
      "tokens_seen": 905379840
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003664994984954865,
      "loss": 3.2388,
      "theoretical_loss": 3.6841702194421204,
      "tokens_seen": 905445376
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003664894684052156,
      "loss": 3.0009,
      "theoretical_loss": 3.684144375493976,
      "tokens_seen": 905510912
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036647943831494484,
      "loss": 2.9816,
      "theoretical_loss": 3.684118533939891,
      "tokens_seen": 905576448
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.000366469408224674,
      "loss": 2.8529,
      "theoretical_loss": 3.6840926947794714,
      "tokens_seen": 905641984
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003664593781344032,
      "loss": 2.9171,
      "theoretical_loss": 3.6840668580123213,
      "tokens_seen": 905707520
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003664493480441324,
      "loss": 2.864,
      "theoretical_loss": 3.684041023638046,
      "tokens_seen": 905773056
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036643931795386157,
      "loss": 2.971,
      "theoretical_loss": 3.6840151916562505,
      "tokens_seen": 905838592
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036642928786359075,
      "loss": 2.749,
      "theoretical_loss": 3.6839893620665407,
      "tokens_seen": 905904128
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036641925777332,
      "loss": 3.1281,
      "theoretical_loss": 3.6839635348685222,
      "tokens_seen": 905969664
    },
    {
      "epoch": 0.27,
      "objective/train/advantage_avg": 0.49396416544914246,
      "objective/train/docs_used": 520483,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.6379759311676025,
      "objective/train/original_loss": 2.6379759311676025,
      "objective/train/theoretical_loss": 3.6839377100618,
      "objective/train/tokens_used": 926495200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2440721094608307,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050637125968933,
      "objective/train/weighted_lm_loss": 2.771681547164917,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0396549701690674,
      "theoretical_loss": 3.6839377100618,
      "tokens_seen": 906035200
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003664092276830491,
      "loss": 2.9129,
      "theoretical_loss": 3.6839377100618,
      "tokens_seen": 906035200
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036639919759277835,
      "loss": 2.7482,
      "theoretical_loss": 3.6839118876459795,
      "tokens_seen": 906100736
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036638916750250753,
      "loss": 3.0346,
      "theoretical_loss": 3.6838860676206675,
      "tokens_seen": 906166272
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003663791374122367,
      "loss": 2.7129,
      "theoretical_loss": 3.683860249985469,
      "tokens_seen": 906231808
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003663691073219659,
      "loss": 2.733,
      "theoretical_loss": 3.68383443473999,
      "tokens_seen": 906297344
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036635907723169507,
      "loss": 2.8906,
      "theoretical_loss": 3.683808621883837,
      "tokens_seen": 906362880
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003663490471414243,
      "loss": 3.0867,
      "theoretical_loss": 3.6837828114166156,
      "tokens_seen": 906428416
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003663390170511535,
      "loss": 2.8448,
      "theoretical_loss": 3.683757003337932,
      "tokens_seen": 906493952
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036632898696088267,
      "loss": 2.7204,
      "theoretical_loss": 3.6837311976473934,
      "tokens_seen": 906559488
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036631895687061185,
      "loss": 2.9664,
      "theoretical_loss": 3.6837053943446048,
      "tokens_seen": 906625024
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036630892678034103,
      "loss": 3.0296,
      "theoretical_loss": 3.683679593429174,
      "tokens_seen": 906690560
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003662988966900702,
      "loss": 2.772,
      "theoretical_loss": 3.6836537949007067,
      "tokens_seen": 906756096
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036628886659979945,
      "loss": 2.9459,
      "theoretical_loss": 3.6836279987588103,
      "tokens_seen": 906821632
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003662788365095286,
      "loss": 2.7373,
      "theoretical_loss": 3.6836022050030905,
      "tokens_seen": 906887168
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003662688064192578,
      "loss": 2.9369,
      "theoretical_loss": 3.683576413633155,
      "tokens_seen": 906952704
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036625877632898694,
      "loss": 2.8237,
      "theoretical_loss": 3.6835506246486105,
      "tokens_seen": 907018240
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003662487462387162,
      "loss": 2.8447,
      "theoretical_loss": 3.6835248380490646,
      "tokens_seen": 907083776
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036623871614844535,
      "loss": 2.7584,
      "theoretical_loss": 3.6834990538341237,
      "tokens_seen": 907149312
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036622868605817454,
      "loss": 2.7349,
      "theoretical_loss": 3.683473272003395,
      "tokens_seen": 907214848
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003662186559679037,
      "loss": 2.8381,
      "theoretical_loss": 3.6834474925564864,
      "tokens_seen": 907280384
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.00036620862587763295,
      "loss": 2.8703,
      "theoretical_loss": 3.683421715493005,
      "tokens_seen": 907345920
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003661985957873621,
      "loss": 2.8002,
      "theoretical_loss": 3.683395940812558,
      "tokens_seen": 907411456
    },
    {
      "epoch": 0.27,
      "learning_rate": 0.0003661885656970913,
      "loss": 2.8786,
      "theoretical_loss": 3.683370168514754,
      "tokens_seen": 907476992
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036617853560682044,
      "loss": 3.0149,
      "theoretical_loss": 3.6833443985992,
      "tokens_seen": 907542528
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003661685055165497,
      "loss": 2.7716,
      "theoretical_loss": 3.6833186310655033,
      "tokens_seen": 907608064
    },
    {
      "epoch": 0.28,
      "objective/train/advantage_avg": 0.49532827734947205,
      "objective/train/docs_used": 521740,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.7446799278259277,
      "objective/train/original_loss": 2.744680643081665,
      "objective/train/theoretical_loss": 3.6832928659132724,
      "objective/train/tokens_used": 928133600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24542145431041718,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507804155349731,
      "objective/train/weighted_lm_loss": 2.8839433193206787,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0338950157165527,
      "theoretical_loss": 3.6832928659132724,
      "tokens_seen": 907673600
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036615847542627886,
      "loss": 2.8738,
      "theoretical_loss": 3.6832928659132724,
      "tokens_seen": 907673600
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036614844533600804,
      "loss": 2.7178,
      "theoretical_loss": 3.6832671031421156,
      "tokens_seen": 907739136
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003661384152457372,
      "loss": 2.8614,
      "theoretical_loss": 3.6832413427516406,
      "tokens_seen": 907804672
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003661283851554664,
      "loss": 3.2621,
      "theoretical_loss": 3.683215584741455,
      "tokens_seen": 907870208
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003661183550651956,
      "loss": 2.922,
      "theoretical_loss": 3.683189829111168,
      "tokens_seen": 907935744
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003661083249749248,
      "loss": 3.1028,
      "theoretical_loss": 3.683164075860388,
      "tokens_seen": 908001280
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036609829488465394,
      "loss": 2.6785,
      "theoretical_loss": 3.6831383249887226,
      "tokens_seen": 908066816
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003660882647943832,
      "loss": 2.8526,
      "theoretical_loss": 3.6831125764957804,
      "tokens_seen": 908132352
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036607823470411236,
      "loss": 2.84,
      "theoretical_loss": 3.683086830381171,
      "tokens_seen": 908197888
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036606820461384154,
      "loss": 2.7749,
      "theoretical_loss": 3.6830610866445017,
      "tokens_seen": 908263424
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003660581745235707,
      "loss": 2.7119,
      "theoretical_loss": 3.683035345285383,
      "tokens_seen": 908328960
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003660481444332999,
      "loss": 2.7928,
      "theoretical_loss": 3.6830096063034223,
      "tokens_seen": 908394496
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003660381143430291,
      "loss": 2.8589,
      "theoretical_loss": 3.682983869698229,
      "tokens_seen": 908460032
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003660280842527583,
      "loss": 3.0166,
      "theoretical_loss": 3.6829581354694128,
      "tokens_seen": 908525568
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036601805416248745,
      "loss": 2.9128,
      "theoretical_loss": 3.682932403616583,
      "tokens_seen": 908591104
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003660080240722167,
      "loss": 2.5992,
      "theoretical_loss": 3.6829066741393475,
      "tokens_seen": 908656640
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003659979939819458,
      "loss": 3.0184,
      "theoretical_loss": 3.682880947037317,
      "tokens_seen": 908722176
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036598796389167504,
      "loss": 2.7675,
      "theoretical_loss": 3.6828552223101,
      "tokens_seen": 908787712
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003659779338014042,
      "loss": 2.8915,
      "theoretical_loss": 3.6828294999573075,
      "tokens_seen": 908853248
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003659679037111334,
      "loss": 2.8513,
      "theoretical_loss": 3.6828037799785474,
      "tokens_seen": 908918784
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003659578736208626,
      "loss": 2.9526,
      "theoretical_loss": 3.68277806237343,
      "tokens_seen": 908984320
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036594784353059177,
      "loss": 2.907,
      "theoretical_loss": 3.682752347141566,
      "tokens_seen": 909049856
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036593781344032095,
      "loss": 2.7538,
      "theoretical_loss": 3.682726634282564,
      "tokens_seen": 909115392
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003659277833500502,
      "loss": 3.096,
      "theoretical_loss": 3.6827009237960353,
      "tokens_seen": 909180928
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003659177532597793,
      "loss": 2.8225,
      "theoretical_loss": 3.682675215681589,
      "tokens_seen": 909246464
    },
    {
      "epoch": 0.28,
      "objective/train/advantage_avg": 0.4935477077960968,
      "objective/train/docs_used": 522301,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.795099973678589,
      "objective/train/original_loss": 2.795099973678589,
      "objective/train/theoretical_loss": 3.682649509938836,
      "objective/train/tokens_used": 929772000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24365054070949554,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050593376159668,
      "objective/train/weighted_lm_loss": 2.936554431915283,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0425469875335693,
      "theoretical_loss": 3.682649509938836,
      "tokens_seen": 909312000
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036590772316950855,
      "loss": 2.886,
      "theoretical_loss": 3.682649509938836,
      "tokens_seen": 909312000
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036589769307923773,
      "loss": 2.9664,
      "theoretical_loss": 3.6826238065673866,
      "tokens_seen": 909377536
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003658876629889669,
      "loss": 2.8433,
      "theoretical_loss": 3.68259810556685,
      "tokens_seen": 909443072
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003658776328986961,
      "loss": 3.0158,
      "theoretical_loss": 3.6825724069368384,
      "tokens_seen": 909508608
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036586760280842527,
      "loss": 2.8747,
      "theoretical_loss": 3.6825467106769607,
      "tokens_seen": 909574144
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036585757271815445,
      "loss": 2.8604,
      "theoretical_loss": 3.682521016786829,
      "tokens_seen": 909639680
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003658475426278837,
      "loss": 3.0134,
      "theoretical_loss": 3.682495325266053,
      "tokens_seen": 909705216
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003658375125376128,
      "loss": 2.7367,
      "theoretical_loss": 3.682469636114244,
      "tokens_seen": 909770752
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036582748244734205,
      "loss": 2.6925,
      "theoretical_loss": 3.6824439493310135,
      "tokens_seen": 909836288
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003658174523570712,
      "loss": 2.7928,
      "theoretical_loss": 3.682418264915971,
      "tokens_seen": 909901824
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003658074222668004,
      "loss": 3.0097,
      "theoretical_loss": 3.6823925828687294,
      "tokens_seen": 909967360
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003657973921765296,
      "loss": 3.1045,
      "theoretical_loss": 3.682366903188899,
      "tokens_seen": 910032896
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003657873620862588,
      "loss": 2.9382,
      "theoretical_loss": 3.6823412258760904,
      "tokens_seen": 910098432
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036577733199598796,
      "loss": 2.8612,
      "theoretical_loss": 3.682315550929917,
      "tokens_seen": 910163968
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036576730190571714,
      "loss": 2.8877,
      "theoretical_loss": 3.6822898783499873,
      "tokens_seen": 910229504
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003657572718154463,
      "loss": 2.7726,
      "theoretical_loss": 3.682264208135916,
      "tokens_seen": 910295040
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036574724172517555,
      "loss": 3.2201,
      "theoretical_loss": 3.682238540287313,
      "tokens_seen": 910360576
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003657372116349047,
      "loss": 3.0155,
      "theoretical_loss": 3.6822128748037906,
      "tokens_seen": 910426112
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003657271815446339,
      "loss": 2.7924,
      "theoretical_loss": 3.68218721168496,
      "tokens_seen": 910491648
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003657171514543631,
      "loss": 2.7148,
      "theoretical_loss": 3.6821615509304344,
      "tokens_seen": 910557184
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003657071213640923,
      "loss": 2.8941,
      "theoretical_loss": 3.682135892539825,
      "tokens_seen": 910622720
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036569709127382146,
      "loss": 3.0391,
      "theoretical_loss": 3.6821102365127434,
      "tokens_seen": 910688256
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036568706118355064,
      "loss": 3.1275,
      "theoretical_loss": 3.6820845828488027,
      "tokens_seen": 910753792
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003656770310932798,
      "loss": 2.7933,
      "theoretical_loss": 3.6820589315476147,
      "tokens_seen": 910819328
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036566700100300906,
      "loss": 2.8248,
      "theoretical_loss": 3.6820332826087925,
      "tokens_seen": 910884864
    },
    {
      "epoch": 0.28,
      "objective/train/advantage_avg": 0.49525147676467896,
      "objective/train/docs_used": 523753,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.577423334121704,
      "objective/train/original_loss": 2.577423572540283,
      "objective/train/theoretical_loss": 3.6820076360319485,
      "objective/train/tokens_used": 931410400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2454495131969452,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507729053497314,
      "objective/train/weighted_lm_loss": 2.708021640777588,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 0.951229453086853,
      "theoretical_loss": 3.6820076360319485,
      "tokens_seen": 910950400
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003656569709127382,
      "loss": 2.8834,
      "theoretical_loss": 3.6820076360319485,
      "tokens_seen": 910950400
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003656469408224674,
      "loss": 2.9341,
      "theoretical_loss": 3.681981991816694,
      "tokens_seen": 911015936
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036563691073219655,
      "loss": 2.8771,
      "theoretical_loss": 3.6819563499626433,
      "tokens_seen": 911081472
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003656268806419258,
      "loss": 2.865,
      "theoretical_loss": 3.681930710469408,
      "tokens_seen": 911147008
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036561685055165496,
      "loss": 2.7801,
      "theoretical_loss": 3.6819050733366017,
      "tokens_seen": 911212544
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036560682046138414,
      "loss": 2.753,
      "theoretical_loss": 3.681879438563837,
      "tokens_seen": 911278080
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003655967903711134,
      "loss": 2.8546,
      "theoretical_loss": 3.681853806150727,
      "tokens_seen": 911343616
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036558676028084256,
      "loss": 2.943,
      "theoretical_loss": 3.681828176096885,
      "tokens_seen": 911409152
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036557673019057174,
      "loss": 2.7386,
      "theoretical_loss": 3.681802548401924,
      "tokens_seen": 911474688
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003655667001003009,
      "loss": 2.7205,
      "theoretical_loss": 3.681776923065457,
      "tokens_seen": 911540224
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003655566700100301,
      "loss": 2.8349,
      "theoretical_loss": 3.6817513000870985,
      "tokens_seen": 911605760
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003655466399197593,
      "loss": 2.8168,
      "theoretical_loss": 3.681725679466461,
      "tokens_seen": 911671296
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003655366098294885,
      "loss": 2.8309,
      "theoretical_loss": 3.6817000612031583,
      "tokens_seen": 911736832
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036552657973921765,
      "loss": 2.7968,
      "theoretical_loss": 3.681674445296804,
      "tokens_seen": 911802368
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003655165496489469,
      "loss": 3.1401,
      "theoretical_loss": 3.6816488317470126,
      "tokens_seen": 911867904
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.000365506519558676,
      "loss": 2.8901,
      "theoretical_loss": 3.681623220553397,
      "tokens_seen": 911933440
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036549648946840525,
      "loss": 3.0189,
      "theoretical_loss": 3.6815976117155715,
      "tokens_seen": 911998976
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003654864593781344,
      "loss": 3.0331,
      "theoretical_loss": 3.681572005233151,
      "tokens_seen": 912064512
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003654764292878636,
      "loss": 2.8577,
      "theoretical_loss": 3.6815464011057477,
      "tokens_seen": 912130048
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003654663991975928,
      "loss": 2.9644,
      "theoretical_loss": 3.681520799332978,
      "tokens_seen": 912195584
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036545636910732197,
      "loss": 2.9463,
      "theoretical_loss": 3.6814951999144547,
      "tokens_seen": 912261120
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036544633901705115,
      "loss": 3.0716,
      "theoretical_loss": 3.681469602849793,
      "tokens_seen": 912326656
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003654363089267804,
      "loss": 3.1596,
      "theoretical_loss": 3.681444008138606,
      "tokens_seen": 912392192
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003654262788365095,
      "loss": 2.9219,
      "theoretical_loss": 3.6814184157805108,
      "tokens_seen": 912457728
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036541624874623875,
      "loss": 3.0341,
      "theoretical_loss": 3.68139282577512,
      "tokens_seen": 912523264
    },
    {
      "epoch": 0.28,
      "objective/train/advantage_avg": 0.4947524070739746,
      "objective/train/docs_used": 524575,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.6747376918792725,
      "objective/train/original_loss": 2.6747379302978516,
      "objective/train/theoretical_loss": 3.681367238122049,
      "objective/train/tokens_used": 933048800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24483996629714966,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507198572158813,
      "objective/train/weighted_lm_loss": 2.8103768825531006,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 3.681367238122049,
      "tokens_seen": 912588800
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036540621865596793,
      "loss": 2.8167,
      "theoretical_loss": 3.681367238122049,
      "tokens_seen": 912588800
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003653961885656971,
      "loss": 2.9913,
      "theoretical_loss": 3.6813416528209126,
      "tokens_seen": 912654336
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003653861584754263,
      "loss": 3.0299,
      "theoretical_loss": 3.6813160698713263,
      "tokens_seen": 912719872
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036537612838515547,
      "loss": 2.7733,
      "theoretical_loss": 3.6812904892729046,
      "tokens_seen": 912785408
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036536609829488465,
      "loss": 2.651,
      "theoretical_loss": 3.681264911025263,
      "tokens_seen": 912850944
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003653560682046139,
      "loss": 2.74,
      "theoretical_loss": 3.6812393351280157,
      "tokens_seen": 912916480
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.000365346038114343,
      "loss": 2.9678,
      "theoretical_loss": 3.681213761580779,
      "tokens_seen": 912982016
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036533600802407225,
      "loss": 2.8335,
      "theoretical_loss": 3.6811881903831685,
      "tokens_seen": 913047552
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003653259779338014,
      "loss": 2.9302,
      "theoretical_loss": 3.681162621534799,
      "tokens_seen": 913113088
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003653159478435306,
      "loss": 2.9476,
      "theoretical_loss": 3.681137055035286,
      "tokens_seen": 913178624
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003653059177532598,
      "loss": 2.7713,
      "theoretical_loss": 3.6811114908842457,
      "tokens_seen": 913244160
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.000365295887662989,
      "loss": 3.0209,
      "theoretical_loss": 3.681085929081294,
      "tokens_seen": 913309696
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036528585757271816,
      "loss": 3.009,
      "theoretical_loss": 3.6810603696260467,
      "tokens_seen": 913375232
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036527582748244734,
      "loss": 3.0254,
      "theoretical_loss": 3.681034812518119,
      "tokens_seen": 913440768
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003652657973921765,
      "loss": 2.9853,
      "theoretical_loss": 3.6810092577571276,
      "tokens_seen": 913506304
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036525576730190575,
      "loss": 3.0564,
      "theoretical_loss": 3.6809837053426886,
      "tokens_seen": 913571840
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003652457372116349,
      "loss": 3.1085,
      "theoretical_loss": 3.6809581552744177,
      "tokens_seen": 913637376
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003652357071213641,
      "loss": 2.8759,
      "theoretical_loss": 3.680932607551932,
      "tokens_seen": 913702912
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003652256770310933,
      "loss": 2.8917,
      "theoretical_loss": 3.6809070621748474,
      "tokens_seen": 913768448
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003652156469408225,
      "loss": 2.6902,
      "theoretical_loss": 3.6808815191427806,
      "tokens_seen": 913833984
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036520561685055166,
      "loss": 3.0307,
      "theoretical_loss": 3.680855978455348,
      "tokens_seen": 913899520
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036519558676028084,
      "loss": 2.804,
      "theoretical_loss": 3.6808304401121665,
      "tokens_seen": 913965056
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036518555667001,
      "loss": 2.8137,
      "theoretical_loss": 3.6808049041128523,
      "tokens_seen": 914030592
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036517552657973926,
      "loss": 2.8519,
      "theoretical_loss": 3.6807793704570235,
      "tokens_seen": 914096128
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003651654964894684,
      "loss": 2.6445,
      "theoretical_loss": 3.6807538391442955,
      "tokens_seen": 914161664
    },
    {
      "epoch": 0.28,
      "objective/train/advantage_avg": 0.49475592374801636,
      "objective/train/docs_used": 525722,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.731079339981079,
      "objective/train/original_loss": 2.731079339981079,
      "objective/train/theoretical_loss": 3.6807283101742865,
      "objective/train/tokens_used": 934687200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24484093487262726,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507200956344604,
      "objective/train/weighted_lm_loss": 2.8694708347320557,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0396549701690674,
      "theoretical_loss": 3.6807283101742865,
      "tokens_seen": 914227200
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003651554663991976,
      "loss": 2.8642,
      "theoretical_loss": 3.6807283101742865,
      "tokens_seen": 914227200
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036514543630892675,
      "loss": 2.8474,
      "theoretical_loss": 3.6807027835466126,
      "tokens_seen": 914292736
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.000365135406218656,
      "loss": 2.8514,
      "theoretical_loss": 3.680677259260892,
      "tokens_seen": 914358272
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036512537612838516,
      "loss": 2.8883,
      "theoretical_loss": 3.680651737316741,
      "tokens_seen": 914423808
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036511534603811434,
      "loss": 2.8382,
      "theoretical_loss": 3.6806262177137787,
      "tokens_seen": 914489344
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003651053159478435,
      "loss": 2.9693,
      "theoretical_loss": 3.680600700451621,
      "tokens_seen": 914554880
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036509528585757276,
      "loss": 2.8695,
      "theoretical_loss": 3.6805751855298854,
      "tokens_seen": 914620416
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003650852557673019,
      "loss": 2.9812,
      "theoretical_loss": 3.6805496729481906,
      "tokens_seen": 914685952
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003650752256770311,
      "loss": 2.946,
      "theoretical_loss": 3.6805241627061536,
      "tokens_seen": 914751488
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036506519558676025,
      "loss": 2.887,
      "theoretical_loss": 3.6804986548033924,
      "tokens_seen": 914817024
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003650551654964895,
      "loss": 2.6518,
      "theoretical_loss": 3.680473149239525,
      "tokens_seen": 914882560
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036504513540621867,
      "loss": 3.0666,
      "theoretical_loss": 3.6804476460141697,
      "tokens_seen": 914948096
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036503510531594785,
      "loss": 2.7478,
      "theoretical_loss": 3.6804221451269443,
      "tokens_seen": 915013632
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036502507522567703,
      "loss": 2.574,
      "theoretical_loss": 3.680396646577467,
      "tokens_seen": 915079168
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003650150451354062,
      "loss": 2.8127,
      "theoretical_loss": 3.680371150365356,
      "tokens_seen": 915144704
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003650050150451354,
      "loss": 2.6887,
      "theoretical_loss": 3.6803456564902297,
      "tokens_seen": 915210240
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003649949849548646,
      "loss": 2.8694,
      "theoretical_loss": 3.680320164951707,
      "tokens_seen": 915275776
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036498495486459375,
      "loss": 2.7953,
      "theoretical_loss": 3.6802946757494057,
      "tokens_seen": 915341312
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.000364974924774323,
      "loss": 2.9514,
      "theoretical_loss": 3.6802691888829453,
      "tokens_seen": 915406848
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003649648946840521,
      "loss": 2.8604,
      "theoretical_loss": 3.680243704351944,
      "tokens_seen": 915472384
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036495486459378135,
      "loss": 2.9338,
      "theoretical_loss": 3.6802182221560207,
      "tokens_seen": 915537920
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036494483450351053,
      "loss": 3.1862,
      "theoretical_loss": 3.6801927422947944,
      "tokens_seen": 915603456
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003649348044132397,
      "loss": 2.8319,
      "theoretical_loss": 3.680167264767884,
      "tokens_seen": 915668992
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003649247743229689,
      "loss": 2.8874,
      "theoretical_loss": 3.680141789574909,
      "tokens_seen": 915734528
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036491474423269813,
      "loss": 2.8148,
      "theoretical_loss": 3.680116316715488,
      "tokens_seen": 915800064
    },
    {
      "epoch": 0.28,
      "objective/train/advantage_avg": 0.49382081627845764,
      "objective/train/docs_used": 526330,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.444495439529419,
      "objective/train/original_loss": 2.44449520111084,
      "objective/train/theoretical_loss": 3.6800908461892403,
      "objective/train/tokens_used": 936325600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24393555521965027,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506221055984497,
      "objective/train/weighted_lm_loss": 2.5681865215301514,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 3.6800908461892403,
      "tokens_seen": 915865600
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036490471414242726,
      "loss": 2.8015,
      "theoretical_loss": 3.6800908461892403,
      "tokens_seen": 915865600
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003648946840521565,
      "loss": 3.0142,
      "theoretical_loss": 3.6800653779957857,
      "tokens_seen": 915931136
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003648846539618856,
      "loss": 3.0812,
      "theoretical_loss": 3.6800399121347436,
      "tokens_seen": 915996672
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036487462387161485,
      "loss": 2.8837,
      "theoretical_loss": 3.680014448605733,
      "tokens_seen": 916062208
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003648645937813441,
      "loss": 2.9149,
      "theoretical_loss": 3.6799889874083744,
      "tokens_seen": 916127744
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003648545636910732,
      "loss": 3.0495,
      "theoretical_loss": 3.6799635285422867,
      "tokens_seen": 916193280
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036484453360080245,
      "loss": 2.6777,
      "theoretical_loss": 3.67993807200709,
      "tokens_seen": 916258816
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003648345035105316,
      "loss": 2.9064,
      "theoretical_loss": 3.6799126178024046,
      "tokens_seen": 916324352
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003648244734202608,
      "loss": 2.9802,
      "theoretical_loss": 3.67988716592785,
      "tokens_seen": 916389888
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036481444332999,
      "loss": 2.9679,
      "theoretical_loss": 3.679861716383046,
      "tokens_seen": 916455424
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003648044132397192,
      "loss": 2.917,
      "theoretical_loss": 3.6798362691676134,
      "tokens_seen": 916520960
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036479438314944836,
      "loss": 3.099,
      "theoretical_loss": 3.6798108242811725,
      "tokens_seen": 916586496
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036478435305917754,
      "loss": 3.0394,
      "theoretical_loss": 3.6797853817233435,
      "tokens_seen": 916652032
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003647743229689067,
      "loss": 2.932,
      "theoretical_loss": 3.6797599414937463,
      "tokens_seen": 916717568
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036476429287863595,
      "loss": 3.1354,
      "theoretical_loss": 3.6797345035920017,
      "tokens_seen": 916783104
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003647542627883651,
      "loss": 3.1101,
      "theoretical_loss": 3.679709068017731,
      "tokens_seen": 916848640
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003647442326980943,
      "loss": 2.9773,
      "theoretical_loss": 3.679683634770554,
      "tokens_seen": 916914176
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003647342026078235,
      "loss": 2.8551,
      "theoretical_loss": 3.6796582038500922,
      "tokens_seen": 916979712
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003647241725175527,
      "loss": 2.8429,
      "theoretical_loss": 3.6796327752559654,
      "tokens_seen": 917045248
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036471414242728186,
      "loss": 2.8137,
      "theoretical_loss": 3.679607348987796,
      "tokens_seen": 917110784
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036470411233701104,
      "loss": 2.8703,
      "theoretical_loss": 3.6795819250452038,
      "tokens_seen": 917176320
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003646940822467402,
      "loss": 2.9524,
      "theoretical_loss": 3.679556503427811,
      "tokens_seen": 917241856
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036468405215646946,
      "loss": 3.0322,
      "theoretical_loss": 3.6795310841352378,
      "tokens_seen": 917307392
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003646740220661986,
      "loss": 3.0926,
      "theoretical_loss": 3.6795056671671063,
      "tokens_seen": 917372928
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003646639919759278,
      "loss": 3.0118,
      "theoretical_loss": 3.679480252523038,
      "tokens_seen": 917438464
    },
    {
      "epoch": 0.28,
      "objective/train/advantage_avg": 0.4939126670360565,
      "objective/train/docs_used": 527026,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.369124412536621,
      "objective/train/original_loss": 2.369124412536621,
      "objective/train/theoretical_loss": 3.6794548402026535,
      "objective/train/tokens_used": 937964000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24405470490455627,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506318807601929,
      "objective/train/weighted_lm_loss": 2.489412546157837,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0413066148757935,
      "theoretical_loss": 3.6794548402026535,
      "tokens_seen": 917504000
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036465396188565695,
      "loss": 2.8547,
      "theoretical_loss": 3.6794548402026535,
      "tokens_seen": 917504000
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003646439317953862,
      "loss": 2.8182,
      "theoretical_loss": 3.679429430205575,
      "tokens_seen": 917569536
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036463390170511536,
      "loss": 2.8643,
      "theoretical_loss": 3.6794040225314246,
      "tokens_seen": 917635072
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036462387161484454,
      "loss": 2.8302,
      "theoretical_loss": 3.679378617179823,
      "tokens_seen": 917700608
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003646138415245737,
      "loss": 2.6768,
      "theoretical_loss": 3.679353214150393,
      "tokens_seen": 917766144
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036460381143430296,
      "loss": 2.9009,
      "theoretical_loss": 3.679327813442756,
      "tokens_seen": 917831680
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003645937813440321,
      "loss": 2.7658,
      "theoretical_loss": 3.6793024150565343,
      "tokens_seen": 917897216
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003645837512537613,
      "loss": 2.8921,
      "theoretical_loss": 3.6792770189913497,
      "tokens_seen": 917962752
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036457372116349045,
      "loss": 2.9393,
      "theoretical_loss": 3.679251625246825,
      "tokens_seen": 918028288
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003645636910732197,
      "loss": 2.8632,
      "theoretical_loss": 3.6792262338225825,
      "tokens_seen": 918093824
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036455366098294887,
      "loss": 2.7927,
      "theoretical_loss": 3.6792008447182436,
      "tokens_seen": 918159360
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036454363089267805,
      "loss": 2.9361,
      "theoretical_loss": 3.6791754579334315,
      "tokens_seen": 918224896
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036453360080240723,
      "loss": 3.0804,
      "theoretical_loss": 3.6791500734677687,
      "tokens_seen": 918290432
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003645235707121364,
      "loss": 2.9364,
      "theoretical_loss": 3.6791246913208777,
      "tokens_seen": 918355968
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003645135406218656,
      "loss": 2.7911,
      "theoretical_loss": 3.6790993114923816,
      "tokens_seen": 918421504
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003645035105315948,
      "loss": 3.04,
      "theoretical_loss": 3.6790739339819027,
      "tokens_seen": 918487040
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036449348044132395,
      "loss": 2.9129,
      "theoretical_loss": 3.6790485587890642,
      "tokens_seen": 918552576
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003644834503510532,
      "loss": 3.0175,
      "theoretical_loss": 3.6790231859134894,
      "tokens_seen": 918618112
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003644734202607823,
      "loss": 2.9528,
      "theoretical_loss": 3.6789978153548013,
      "tokens_seen": 918683648
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036446339017051155,
      "loss": 2.7563,
      "theoretical_loss": 3.6789724471126224,
      "tokens_seen": 918749184
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036445336008024073,
      "loss": 3.0354,
      "theoretical_loss": 3.6789470811865765,
      "tokens_seen": 918814720
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003644433299899699,
      "loss": 2.8205,
      "theoretical_loss": 3.6789217175762867,
      "tokens_seen": 918880256
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003644332998996991,
      "loss": 3.1656,
      "theoretical_loss": 3.678896356281377,
      "tokens_seen": 918945792
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036442326980942833,
      "loss": 3.0113,
      "theoretical_loss": 3.6788709973014706,
      "tokens_seen": 919011328
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036441323971915746,
      "loss": 2.7346,
      "theoretical_loss": 3.678845640636191,
      "tokens_seen": 919076864
    },
    {
      "epoch": 0.28,
      "objective/train/advantage_avg": 0.4956441819667816,
      "objective/train/docs_used": 528220,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.729145050048828,
      "objective/train/original_loss": 2.729145050048828,
      "objective/train/theoretical_loss": 3.678820286285162,
      "objective/train/tokens_used": 939602400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2457113415002823,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0508134365081787,
      "objective/train/weighted_lm_loss": 2.8677659034729004,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0417574644088745,
      "theoretical_loss": 3.678820286285162,
      "tokens_seen": 919142400
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003644032096288867,
      "loss": 2.9601,
      "theoretical_loss": 3.678820286285162,
      "tokens_seen": 919142400
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003643931795386158,
      "loss": 2.9827,
      "theoretical_loss": 3.6787949342480073,
      "tokens_seen": 919207936
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036438314944834505,
      "loss": 2.9773,
      "theoretical_loss": 3.6787695845243507,
      "tokens_seen": 919273472
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036437311935807424,
      "loss": 2.9578,
      "theoretical_loss": 3.6787442371138166,
      "tokens_seen": 919339008
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003643630892678034,
      "loss": 2.8847,
      "theoretical_loss": 3.678718892016029,
      "tokens_seen": 919404544
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003643530591775326,
      "loss": 2.6878,
      "theoretical_loss": 3.6786935492306116,
      "tokens_seen": 919470080
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003643430290872618,
      "loss": 2.9889,
      "theoretical_loss": 3.6786682087571894,
      "tokens_seen": 919535616
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036433299899699096,
      "loss": 2.9042,
      "theoretical_loss": 3.6786428705953855,
      "tokens_seen": 919601152
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003643229689067202,
      "loss": 3.2433,
      "theoretical_loss": 3.6786175347448253,
      "tokens_seen": 919666688
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003643129388164493,
      "loss": 2.9422,
      "theoretical_loss": 3.678592201205133,
      "tokens_seen": 919732224
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036430290872617856,
      "loss": 2.8321,
      "theoretical_loss": 3.6785668699759335,
      "tokens_seen": 919797760
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003642928786359077,
      "loss": 2.8073,
      "theoretical_loss": 3.678541541056851,
      "tokens_seen": 919863296
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003642828485456369,
      "loss": 2.9287,
      "theoretical_loss": 3.6785162144475105,
      "tokens_seen": 919928832
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003642728184553661,
      "loss": 2.7458,
      "theoretical_loss": 3.6784908901475366,
      "tokens_seen": 919994368
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003642627883650953,
      "loss": 2.868,
      "theoretical_loss": 3.678465568156555,
      "tokens_seen": 920059904
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036425275827482446,
      "loss": 2.8982,
      "theoretical_loss": 3.67844024847419,
      "tokens_seen": 920125440
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003642427281845537,
      "loss": 3.0157,
      "theoretical_loss": 3.6784149311000665,
      "tokens_seen": 920190976
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003642326980942828,
      "loss": 3.1927,
      "theoretical_loss": 3.67838961603381,
      "tokens_seen": 920256512
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036422266800401206,
      "loss": 3.0534,
      "theoretical_loss": 3.678364303275046,
      "tokens_seen": 920322048
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003642126379137412,
      "loss": 2.596,
      "theoretical_loss": 3.6783389928234005,
      "tokens_seen": 920387584
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003642026078234704,
      "loss": 2.9955,
      "theoretical_loss": 3.6783136846784976,
      "tokens_seen": 920453120
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003641925777331996,
      "loss": 2.7994,
      "theoretical_loss": 3.678288378839963,
      "tokens_seen": 920518656
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003641825476429288,
      "loss": 3.042,
      "theoretical_loss": 3.6782630753074237,
      "tokens_seen": 920584192
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036417251755265797,
      "loss": 2.5648,
      "theoretical_loss": 3.6782377740805043,
      "tokens_seen": 920649728
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036416248746238715,
      "loss": 2.8538,
      "theoretical_loss": 3.6782124751588308,
      "tokens_seen": 920715264
    },
    {
      "epoch": 0.28,
      "objective/train/advantage_avg": 0.4951143264770508,
      "objective/train/docs_used": 528886,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.2411110401153564,
      "objective/train/original_loss": 2.2411108016967773,
      "objective/train/theoretical_loss": 3.678187178542029,
      "objective/train/tokens_used": 941240800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2451726794242859,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050757884979248,
      "objective/train/weighted_lm_loss": 2.3547983169555664,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0454468727111816,
      "theoretical_loss": 3.678187178542029,
      "tokens_seen": 920780800
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036415245737211633,
      "loss": 2.8006,
      "theoretical_loss": 3.678187178542029,
      "tokens_seen": 920780800
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036414242728184556,
      "loss": 2.6828,
      "theoretical_loss": 3.6781618842297252,
      "tokens_seen": 920846336
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003641323971915747,
      "loss": 2.968,
      "theoretical_loss": 3.6781365922215454,
      "tokens_seen": 920911872
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003641223671013039,
      "loss": 2.8975,
      "theoretical_loss": 3.6781113025171157,
      "tokens_seen": 920977408
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036411233701103316,
      "loss": 2.9607,
      "theoretical_loss": 3.678086015116062,
      "tokens_seen": 921042944
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003641023069207623,
      "loss": 2.9527,
      "theoretical_loss": 3.6780607300180113,
      "tokens_seen": 921108480
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003640922768304915,
      "loss": 3.0201,
      "theoretical_loss": 3.6780354472225896,
      "tokens_seen": 921174016
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036408224674022065,
      "loss": 2.7915,
      "theoretical_loss": 3.678010166729423,
      "tokens_seen": 921239552
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003640722166499499,
      "loss": 3.0467,
      "theoretical_loss": 3.6779848885381394,
      "tokens_seen": 921305088
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036406218655967907,
      "loss": 2.7508,
      "theoretical_loss": 3.6779596126483645,
      "tokens_seen": 921370624
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036405215646940825,
      "loss": 2.7365,
      "theoretical_loss": 3.6779343390597248,
      "tokens_seen": 921436160
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036404212637913743,
      "loss": 2.7848,
      "theoretical_loss": 3.677909067771848,
      "tokens_seen": 921501696
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003640320962888666,
      "loss": 2.9281,
      "theoretical_loss": 3.6778837987843604,
      "tokens_seen": 921567232
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003640220661985958,
      "loss": 2.9908,
      "theoretical_loss": 3.6778585320968893,
      "tokens_seen": 921632768
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.000364012036108325,
      "loss": 2.9881,
      "theoretical_loss": 3.6778332677090617,
      "tokens_seen": 921698304
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036400200601805415,
      "loss": 2.9178,
      "theoretical_loss": 3.6778080056205056,
      "tokens_seen": 921763840
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003639919759277834,
      "loss": 2.8476,
      "theoretical_loss": 3.6777827458308474,
      "tokens_seen": 921829376
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003639819458375125,
      "loss": 2.8672,
      "theoretical_loss": 3.677757488339714,
      "tokens_seen": 921894912
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036397191574724175,
      "loss": 2.9899,
      "theoretical_loss": 3.6777322331467346,
      "tokens_seen": 921960448
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036396188565697093,
      "loss": 2.6888,
      "theoretical_loss": 3.677706980251535,
      "tokens_seen": 922025984
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003639518555667001,
      "loss": 2.942,
      "theoretical_loss": 3.677681729653744,
      "tokens_seen": 922091520
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003639418254764293,
      "loss": 2.9733,
      "theoretical_loss": 3.6776564813529884,
      "tokens_seen": 922157056
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036393179538615853,
      "loss": 2.7049,
      "theoretical_loss": 3.6776312353488967,
      "tokens_seen": 922222592
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036392176529588766,
      "loss": 2.8723,
      "theoretical_loss": 3.677605991641097,
      "tokens_seen": 922288128
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003639117352056169,
      "loss": 2.739,
      "theoretical_loss": 3.6775807502292164,
      "tokens_seen": 922353664
    },
    {
      "epoch": 0.28,
      "objective/train/advantage_avg": 0.4935733377933502,
      "objective/train/docs_used": 530064,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.6247262954711914,
      "objective/train/original_loss": 2.6247262954711914,
      "objective/train/theoretical_loss": 3.6775555111128835,
      "objective/train/tokens_used": 942879200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2436830848455429,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050595998764038,
      "objective/train/weighted_lm_loss": 2.7573447227478027,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0396549701690674,
      "theoretical_loss": 3.6775555111128835,
      "tokens_seen": 922419200
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.000363901705115346,
      "loss": 2.6485,
      "theoretical_loss": 3.6775555111128835,
      "tokens_seen": 922419200
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036389167502507525,
      "loss": 2.8744,
      "theoretical_loss": 3.677530274291727,
      "tokens_seen": 922484736
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036388164493480444,
      "loss": 3.0516,
      "theoretical_loss": 3.677505039765374,
      "tokens_seen": 922550272
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003638716148445336,
      "loss": 2.9521,
      "theoretical_loss": 3.6774798075334543,
      "tokens_seen": 922615808
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003638615847542628,
      "loss": 2.9394,
      "theoretical_loss": 3.6774545775955945,
      "tokens_seen": 922681344
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.000363851554663992,
      "loss": 2.8814,
      "theoretical_loss": 3.6774293499514243,
      "tokens_seen": 922746880
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036384152457372116,
      "loss": 2.7559,
      "theoretical_loss": 3.6774041246005726,
      "tokens_seen": 922812416
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003638314944834504,
      "loss": 3.0075,
      "theoretical_loss": 3.677378901542667,
      "tokens_seen": 922877952
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003638214643931795,
      "loss": 2.772,
      "theoretical_loss": 3.677353680777337,
      "tokens_seen": 922943488
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036381143430290876,
      "loss": 2.9348,
      "theoretical_loss": 3.677328462304211,
      "tokens_seen": 923009024
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003638014042126379,
      "loss": 2.6856,
      "theoretical_loss": 3.6773032461229183,
      "tokens_seen": 923074560
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003637913741223671,
      "loss": 2.8362,
      "theoretical_loss": 3.6772780322330876,
      "tokens_seen": 923140096
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003637813440320963,
      "loss": 3.0608,
      "theoretical_loss": 3.677252820634349,
      "tokens_seen": 923205632
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003637713139418255,
      "loss": 2.722,
      "theoretical_loss": 3.6772276113263302,
      "tokens_seen": 923271168
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036376128385155466,
      "loss": 2.7776,
      "theoretical_loss": 3.6772024043086615,
      "tokens_seen": 923336704
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003637512537612839,
      "loss": 2.8891,
      "theoretical_loss": 3.677177199580972,
      "tokens_seen": 923402240
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.000363741223671013,
      "loss": 3.0349,
      "theoretical_loss": 3.677151997142891,
      "tokens_seen": 923467776
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036373119358074226,
      "loss": 2.9791,
      "theoretical_loss": 3.677126796994049,
      "tokens_seen": 923533312
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003637211634904714,
      "loss": 2.6572,
      "theoretical_loss": 3.677101599134074,
      "tokens_seen": 923598848
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003637111334002006,
      "loss": 2.612,
      "theoretical_loss": 3.6770764035625967,
      "tokens_seen": 923664384
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003637011033099298,
      "loss": 2.941,
      "theoretical_loss": 3.677051210279247,
      "tokens_seen": 923729920
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.000363691073219659,
      "loss": 2.892,
      "theoretical_loss": 3.6770260192836544,
      "tokens_seen": 923795456
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036368104312938817,
      "loss": 2.9942,
      "theoretical_loss": 3.677000830575449,
      "tokens_seen": 923860992
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036367101303911735,
      "loss": 2.7547,
      "theoretical_loss": 3.676975644154261,
      "tokens_seen": 923926528
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036366098294884653,
      "loss": 2.9772,
      "theoretical_loss": 3.6769504600197203,
      "tokens_seen": 923992064
    },
    {
      "epoch": 0.28,
      "objective/train/advantage_avg": 0.493282675743103,
      "objective/train/docs_used": 530632,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.3169541358947754,
      "objective/train/original_loss": 3.3169543743133545,
      "objective/train/theoretical_loss": 3.6769252781714576,
      "objective/train/tokens_used": 944517600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24340273439884186,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050565481185913,
      "objective/train/weighted_lm_loss": 3.4846675395965576,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0408108234405518,
      "theoretical_loss": 3.6769252781714576,
      "tokens_seen": 924057600
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036365095285857576,
      "loss": 3.3007,
      "theoretical_loss": 3.6769252781714576,
      "tokens_seen": 924057600
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003636409227683049,
      "loss": 2.7515,
      "theoretical_loss": 3.6769000986091025,
      "tokens_seen": 924123136
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003636308926780341,
      "loss": 2.8537,
      "theoretical_loss": 3.6768749213322858,
      "tokens_seen": 924188672
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036362086258776325,
      "loss": 2.965,
      "theoretical_loss": 3.676849746340639,
      "tokens_seen": 924254208
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003636108324974925,
      "loss": 2.9084,
      "theoretical_loss": 3.67682457363379,
      "tokens_seen": 924319744
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036360080240722167,
      "loss": 2.8848,
      "theoretical_loss": 3.6767994032113727,
      "tokens_seen": 924385280
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036359077231695085,
      "loss": 2.9899,
      "theoretical_loss": 3.6767742350730153,
      "tokens_seen": 924450816
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036358074222668003,
      "loss": 2.6668,
      "theoretical_loss": 3.67674906921835,
      "tokens_seen": 924516352
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036357071213640927,
      "loss": 2.8329,
      "theoretical_loss": 3.676723905647007,
      "tokens_seen": 924581888
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003635606820461384,
      "loss": 2.7467,
      "theoretical_loss": 3.6766987443586183,
      "tokens_seen": 924647424
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036355065195586763,
      "loss": 3.038,
      "theoretical_loss": 3.6766735853528143,
      "tokens_seen": 924712960
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036354062186559676,
      "loss": 3.0563,
      "theoretical_loss": 3.6766484286292256,
      "tokens_seen": 924778496
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.000363530591775326,
      "loss": 2.8072,
      "theoretical_loss": 3.6766232741874845,
      "tokens_seen": 924844032
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036352056168505517,
      "loss": 2.7746,
      "theoretical_loss": 3.676598122027222,
      "tokens_seen": 924909568
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036351053159478435,
      "loss": 2.9157,
      "theoretical_loss": 3.6765729721480693,
      "tokens_seen": 924975104
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036350050150451353,
      "loss": 2.8188,
      "theoretical_loss": 3.676547824549658,
      "tokens_seen": 925040640
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003634904714142427,
      "loss": 2.9579,
      "theoretical_loss": 3.6765226792316206,
      "tokens_seen": 925106176
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003634804413239719,
      "loss": 2.9469,
      "theoretical_loss": 3.676497536193587,
      "tokens_seen": 925171712
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036347041123370113,
      "loss": 2.9552,
      "theoretical_loss": 3.6764723954351908,
      "tokens_seen": 925237248
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036346038114343026,
      "loss": 2.8716,
      "theoretical_loss": 3.6764472569560622,
      "tokens_seen": 925302784
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003634503510531595,
      "loss": 2.8345,
      "theoretical_loss": 3.6764221207558343,
      "tokens_seen": 925368320
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003634403209628886,
      "loss": 2.8512,
      "theoretical_loss": 3.6763969868341393,
      "tokens_seen": 925433856
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036343029087261786,
      "loss": 2.858,
      "theoretical_loss": 3.6763718551906086,
      "tokens_seen": 925499392
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036342026078234704,
      "loss": 2.7665,
      "theoretical_loss": 3.6763467258248745,
      "tokens_seen": 925564928
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003634102306920762,
      "loss": 2.7937,
      "theoretical_loss": 3.67632159873657,
      "tokens_seen": 925630464
    },
    {
      "epoch": 0.28,
      "objective/train/advantage_avg": 0.4944971203804016,
      "objective/train/docs_used": 532017,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.4862895011901855,
      "objective/train/original_loss": 2.4862895011901855,
      "objective/train/theoretical_loss": 3.6762964739253263,
      "objective/train/tokens_used": 946156000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24457329511642456,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506929159164429,
      "objective/train/weighted_lm_loss": 2.6121864318847656,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0437887907028198,
      "theoretical_loss": 3.6762964739253263,
      "tokens_seen": 925696000
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003634002006018054,
      "loss": 2.7534,
      "theoretical_loss": 3.6762964739253263,
      "tokens_seen": 925696000
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036339017051153464,
      "loss": 2.6108,
      "theoretical_loss": 3.6762713513907768,
      "tokens_seen": 925761536
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036338014042126376,
      "loss": 2.9007,
      "theoretical_loss": 3.676246231132554,
      "tokens_seen": 925827072
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.000363370110330993,
      "loss": 3.0693,
      "theoretical_loss": 3.67622111315029,
      "tokens_seen": 925892608
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003633600802407222,
      "loss": 2.6756,
      "theoretical_loss": 3.6761959974436174,
      "tokens_seen": 925958144
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036335005015045136,
      "loss": 2.7607,
      "theoretical_loss": 3.67617088401217,
      "tokens_seen": 926023680
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003633400200601806,
      "loss": 2.574,
      "theoretical_loss": 3.6761457728555795,
      "tokens_seen": 926089216
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003633299899699097,
      "loss": 2.9785,
      "theoretical_loss": 3.67612066397348,
      "tokens_seen": 926154752
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036331995987963896,
      "loss": 2.8886,
      "theoretical_loss": 3.676095557365504,
      "tokens_seen": 926220288
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003633099297893681,
      "loss": 3.1116,
      "theoretical_loss": 3.6760704530312847,
      "tokens_seen": 926285824
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003632998996990973,
      "loss": 2.836,
      "theoretical_loss": 3.6760453509704547,
      "tokens_seen": 926351360
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003632898696088265,
      "loss": 2.9562,
      "theoretical_loss": 3.6760202511826483,
      "tokens_seen": 926416896
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003632798395185557,
      "loss": 2.7757,
      "theoretical_loss": 3.6759951536674986,
      "tokens_seen": 926482432
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036326980942828486,
      "loss": 2.8913,
      "theoretical_loss": 3.6759700584246393,
      "tokens_seen": 926547968
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003632597793380141,
      "loss": 2.6583,
      "theoretical_loss": 3.675944965453703,
      "tokens_seen": 926613504
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003632497492477432,
      "loss": 2.9518,
      "theoretical_loss": 3.6759198747543245,
      "tokens_seen": 926679040
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036323971915747246,
      "loss": 3.0914,
      "theoretical_loss": 3.675894786326137,
      "tokens_seen": 926744576
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003632296890672016,
      "loss": 2.6381,
      "theoretical_loss": 3.675869700168774,
      "tokens_seen": 926810112
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003632196589769308,
      "loss": 3.0061,
      "theoretical_loss": 3.6758446162818705,
      "tokens_seen": 926875648
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036320962888666,
      "loss": 2.7776,
      "theoretical_loss": 3.6758195346650595,
      "tokens_seen": 926941184
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003631995987963892,
      "loss": 2.8612,
      "theoretical_loss": 3.675794455317975,
      "tokens_seen": 927006720
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036318956870611837,
      "loss": 2.9136,
      "theoretical_loss": 3.675769378240252,
      "tokens_seen": 927072256
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036317953861584755,
      "loss": 3.0831,
      "theoretical_loss": 3.6757443034315243,
      "tokens_seen": 927137792
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036316950852557673,
      "loss": 3.0091,
      "theoretical_loss": 3.675719230891426,
      "tokens_seen": 927203328
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036315947843530596,
      "loss": 2.926,
      "theoretical_loss": 3.6756941606195914,
      "tokens_seen": 927268864
    },
    {
      "epoch": 0.28,
      "objective/train/advantage_avg": 0.4947701394557953,
      "objective/train/docs_used": 532753,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.7250068187713623,
      "objective/train/original_loss": 2.725006580352783,
      "objective/train/theoretical_loss": 3.6756690926156557,
      "objective/train/tokens_used": 947794400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24485142529010773,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507218837738037,
      "objective/train/weighted_lm_loss": 2.863236665725708,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0363596677780151,
      "theoretical_loss": 3.6756690926156557,
      "tokens_seen": 927334400
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003631494483450351,
      "loss": 2.6739,
      "theoretical_loss": 3.6756690926156557,
      "tokens_seen": 927334400
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003631394182547643,
      "loss": 2.8652,
      "theoretical_loss": 3.6756440268792536,
      "tokens_seen": 927399936
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036312938816449345,
      "loss": 3.0107,
      "theoretical_loss": 3.6756189634100185,
      "tokens_seen": 927465472
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003631193580742227,
      "loss": 2.8032,
      "theoretical_loss": 3.6755939022075865,
      "tokens_seen": 927531008
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036310932798395187,
      "loss": 2.7543,
      "theoretical_loss": 3.6755688432715914,
      "tokens_seen": 927596544
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036309929789368105,
      "loss": 2.9565,
      "theoretical_loss": 3.6755437866016694,
      "tokens_seen": 927662080
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036308926780341023,
      "loss": 2.7568,
      "theoretical_loss": 3.675518732197454,
      "tokens_seen": 927727616
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036307923771313947,
      "loss": 2.7663,
      "theoretical_loss": 3.6754936800585813,
      "tokens_seen": 927793152
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003630692076228686,
      "loss": 2.9387,
      "theoretical_loss": 3.675468630184686,
      "tokens_seen": 927858688
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036305917753259783,
      "loss": 2.6537,
      "theoretical_loss": 3.6754435825754044,
      "tokens_seen": 927924224
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036304914744232696,
      "loss": 2.7712,
      "theoretical_loss": 3.6754185372303705,
      "tokens_seen": 927989760
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003630391173520562,
      "loss": 2.8181,
      "theoretical_loss": 3.6753934941492203,
      "tokens_seen": 928055296
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036302908726178537,
      "loss": 2.6783,
      "theoretical_loss": 3.6753684533315893,
      "tokens_seen": 928120832
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036301905717151455,
      "loss": 2.8142,
      "theoretical_loss": 3.675343414777113,
      "tokens_seen": 928186368
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036300902708124374,
      "loss": 2.7761,
      "theoretical_loss": 3.675318378485427,
      "tokens_seen": 928251904
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003629989969909729,
      "loss": 3.1295,
      "theoretical_loss": 3.6752933444561675,
      "tokens_seen": 928317440
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003629889669007021,
      "loss": 2.9398,
      "theoretical_loss": 3.6752683126889703,
      "tokens_seen": 928382976
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036297893681043133,
      "loss": 2.8864,
      "theoretical_loss": 3.675243283183471,
      "tokens_seen": 928448512
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036296890672016046,
      "loss": 2.9171,
      "theoretical_loss": 3.6752182559393054,
      "tokens_seen": 928514048
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003629588766298897,
      "loss": 2.8388,
      "theoretical_loss": 3.6751932309561104,
      "tokens_seen": 928579584
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003629488465396188,
      "loss": 2.9076,
      "theoretical_loss": 3.6751682082335217,
      "tokens_seen": 928645120
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036293881644934806,
      "loss": 2.8537,
      "theoretical_loss": 3.6751431877711758,
      "tokens_seen": 928710656
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036292878635907724,
      "loss": 2.8673,
      "theoretical_loss": 3.6751181695687087,
      "tokens_seen": 928776192
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003629187562688064,
      "loss": 2.7718,
      "theoretical_loss": 3.675093153625757,
      "tokens_seen": 928841728
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003629087261785356,
      "loss": 2.7023,
      "theoretical_loss": 3.675068139941957,
      "tokens_seen": 928907264
    },
    {
      "epoch": 0.28,
      "objective/train/advantage_avg": 0.49563974142074585,
      "objective/train/docs_used": 534166,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.6936721801757812,
      "objective/train/original_loss": 2.6936721801757812,
      "objective/train/theoretical_loss": 3.6750431285169465,
      "objective/train/tokens_used": 949432800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2457052320241928,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0508129596710205,
      "objective/train/weighted_lm_loss": 2.83058762550354,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0442860126495361,
      "theoretical_loss": 3.6750431285169465,
      "tokens_seen": 928972800
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036289869608826484,
      "loss": 2.8757,
      "theoretical_loss": 3.6750431285169465,
      "tokens_seen": 928972800
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036288866599799396,
      "loss": 2.9124,
      "theoretical_loss": 3.6750181193503604,
      "tokens_seen": 929038336
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003628786359077232,
      "loss": 2.6939,
      "theoretical_loss": 3.674993112441837,
      "tokens_seen": 929103872
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003628686058174523,
      "loss": 2.6961,
      "theoretical_loss": 3.6749681077910124,
      "tokens_seen": 929169408
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036285857572718156,
      "loss": 2.9898,
      "theoretical_loss": 3.6749431053975234,
      "tokens_seen": 929234944
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036284854563691074,
      "loss": 2.6551,
      "theoretical_loss": 3.674918105261008,
      "tokens_seen": 929300480
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003628385155466399,
      "loss": 3.0196,
      "theoretical_loss": 3.674893107381102,
      "tokens_seen": 929366016
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003628284854563691,
      "loss": 2.8186,
      "theoretical_loss": 3.6748681117574433,
      "tokens_seen": 929431552
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003628184553660983,
      "loss": 2.924,
      "theoretical_loss": 3.6748431183896693,
      "tokens_seen": 929497088
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036280842527582747,
      "loss": 2.8521,
      "theoretical_loss": 3.674818127277417,
      "tokens_seen": 929562624
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003627983951855567,
      "loss": 2.8094,
      "theoretical_loss": 3.6747931384203243,
      "tokens_seen": 929628160
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036278836509528583,
      "loss": 2.7268,
      "theoretical_loss": 3.6747681518180286,
      "tokens_seen": 929693696
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036277833500501506,
      "loss": 2.8458,
      "theoretical_loss": 3.6747431674701674,
      "tokens_seen": 929759232
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036276830491474424,
      "loss": 2.7912,
      "theoretical_loss": 3.6747181853763786,
      "tokens_seen": 929824768
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003627582748244734,
      "loss": 3.011,
      "theoretical_loss": 3.6746932055362995,
      "tokens_seen": 929890304
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003627482447342026,
      "loss": 2.9553,
      "theoretical_loss": 3.674668227949568,
      "tokens_seen": 929955840
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003627382146439318,
      "loss": 2.84,
      "theoretical_loss": 3.674643252615823,
      "tokens_seen": 930021376
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036272818455366097,
      "loss": 3.0145,
      "theoretical_loss": 3.6746182795347013,
      "tokens_seen": 930086912
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003627181544633902,
      "loss": 2.7545,
      "theoretical_loss": 3.674593308705842,
      "tokens_seen": 930152448
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036270812437311933,
      "loss": 3.1807,
      "theoretical_loss": 3.6745683401288827,
      "tokens_seen": 930217984
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036269809428284857,
      "loss": 2.9266,
      "theoretical_loss": 3.674543373803462,
      "tokens_seen": 930283520
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003626880641925777,
      "loss": 2.7749,
      "theoretical_loss": 3.674518409729218,
      "tokens_seen": 930349056
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036267803410230693,
      "loss": 2.7678,
      "theoretical_loss": 3.67449344790579,
      "tokens_seen": 930414592
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003626680040120361,
      "loss": 2.9767,
      "theoretical_loss": 3.674468488332815,
      "tokens_seen": 930480128
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003626579739217653,
      "loss": 2.8473,
      "theoretical_loss": 3.674443531009933,
      "tokens_seen": 930545664
    },
    {
      "epoch": 0.28,
      "objective/train/advantage_avg": 0.49492645263671875,
      "objective/train/docs_used": 534638,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.886765718460083,
      "objective/train/original_loss": 2.886765718460083,
      "objective/train/theoretical_loss": 3.674418575936782,
      "objective/train/tokens_used": 951071200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24499695003032684,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507380962371826,
      "objective/train/weighted_lm_loss": 3.033230781555176,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0465034246444702,
      "theoretical_loss": 3.674418575936782,
      "tokens_seen": 930611200
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036264794383149447,
      "loss": 2.9595,
      "theoretical_loss": 3.674418575936782,
      "tokens_seen": 930611200
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036263791374122365,
      "loss": 2.8514,
      "theoretical_loss": 3.674393623113001,
      "tokens_seen": 930676736
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036262788365095283,
      "loss": 2.6374,
      "theoretical_loss": 3.674368672538229,
      "tokens_seen": 930742272
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036261785356068207,
      "loss": 2.9241,
      "theoretical_loss": 3.674343724212105,
      "tokens_seen": 930807808
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036260782347041125,
      "loss": 3.1427,
      "theoretical_loss": 3.6743187781342677,
      "tokens_seen": 930873344
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036259779338014043,
      "loss": 3.1438,
      "theoretical_loss": 3.674293834304356,
      "tokens_seen": 930938880
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036258776328986967,
      "loss": 2.802,
      "theoretical_loss": 3.67426889272201,
      "tokens_seen": 931004416
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003625777331995988,
      "loss": 3.0262,
      "theoretical_loss": 3.6742439533868687,
      "tokens_seen": 931069952
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036256770310932803,
      "loss": 2.9381,
      "theoretical_loss": 3.674219016298571,
      "tokens_seen": 931135488
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036255767301905716,
      "loss": 2.9719,
      "theoretical_loss": 3.6741940814567564,
      "tokens_seen": 931201024
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003625476429287864,
      "loss": 2.9061,
      "theoretical_loss": 3.674169148861065,
      "tokens_seen": 931266560
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003625376128385156,
      "loss": 2.8243,
      "theoretical_loss": 3.674144218511136,
      "tokens_seen": 931332096
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036252758274824475,
      "loss": 2.804,
      "theoretical_loss": 3.674119290406609,
      "tokens_seen": 931397632
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036251755265797394,
      "loss": 2.8751,
      "theoretical_loss": 3.6740943645471242,
      "tokens_seen": 931463168
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003625075225677031,
      "loss": 2.8295,
      "theoretical_loss": 3.6740694409323216,
      "tokens_seen": 931528704
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003624974924774323,
      "loss": 2.8595,
      "theoretical_loss": 3.6740445195618405,
      "tokens_seen": 931594240
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036248746238716153,
      "loss": 2.887,
      "theoretical_loss": 3.6740196004353214,
      "tokens_seen": 931659776
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036247743229689066,
      "loss": 2.921,
      "theoretical_loss": 3.6739946835524044,
      "tokens_seen": 931725312
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003624674022066199,
      "loss": 2.9906,
      "theoretical_loss": 3.6739697689127295,
      "tokens_seen": 931790848
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.000362457372116349,
      "loss": 2.6256,
      "theoretical_loss": 3.673944856515937,
      "tokens_seen": 931856384
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036244734202607826,
      "loss": 3.1491,
      "theoretical_loss": 3.673919946361667,
      "tokens_seen": 931921920
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036243731193580744,
      "loss": 2.7087,
      "theoretical_loss": 3.673895038449561,
      "tokens_seen": 931987456
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003624272818455366,
      "loss": 2.6763,
      "theoretical_loss": 3.6738701327792587,
      "tokens_seen": 932052992
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003624172517552658,
      "loss": 2.8092,
      "theoretical_loss": 3.673845229350401,
      "tokens_seen": 932118528
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036240722166499504,
      "loss": 2.7935,
      "theoretical_loss": 3.673820328162628,
      "tokens_seen": 932184064
    },
    {
      "epoch": 0.28,
      "objective/train/advantage_avg": 0.4953204095363617,
      "objective/train/docs_used": 535899,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.877894878387451,
      "objective/train/original_loss": 2.8778953552246094,
      "objective/train/theoretical_loss": 3.6737954292155814,
      "objective/train/tokens_used": 952709600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24541819095611572,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507795810699463,
      "objective/train/weighted_lm_loss": 3.023832321166992,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0253151655197144,
      "theoretical_loss": 3.6737954292155814,
      "tokens_seen": 932249600
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036239719157472416,
      "loss": 2.79,
      "theoretical_loss": 3.6737954292155814,
      "tokens_seen": 932249600
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003623871614844534,
      "loss": 2.8114,
      "theoretical_loss": 3.673770532508901,
      "tokens_seen": 932315136
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003623771313941825,
      "loss": 2.9563,
      "theoretical_loss": 3.673745638042229,
      "tokens_seen": 932380672
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036236710130391176,
      "loss": 3.0132,
      "theoretical_loss": 3.673720745815206,
      "tokens_seen": 932446208
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036235707121364094,
      "loss": 2.9704,
      "theoretical_loss": 3.673695855827473,
      "tokens_seen": 932511744
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003623470411233701,
      "loss": 2.9942,
      "theoretical_loss": 3.6736709680786706,
      "tokens_seen": 932577280
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003623370110330993,
      "loss": 3.2739,
      "theoretical_loss": 3.6736460825684407,
      "tokens_seen": 932642816
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003623269809428285,
      "loss": 2.6712,
      "theoretical_loss": 3.6736211992964254,
      "tokens_seen": 932708352
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036231695085255767,
      "loss": 2.7824,
      "theoretical_loss": 3.6735963182622644,
      "tokens_seen": 932773888
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003623069207622869,
      "loss": 3.0554,
      "theoretical_loss": 3.6735714394656007,
      "tokens_seen": 932839424
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036229689067201603,
      "loss": 2.5743,
      "theoretical_loss": 3.673546562906075,
      "tokens_seen": 932904960
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036228686058174526,
      "loss": 2.8545,
      "theoretical_loss": 3.67352168858333,
      "tokens_seen": 932970496
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036227683049147444,
      "loss": 2.9085,
      "theoretical_loss": 3.673496816497006,
      "tokens_seen": 933036032
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003622668004012036,
      "loss": 3.1804,
      "theoretical_loss": 3.6734719466467465,
      "tokens_seen": 933101568
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003622567703109328,
      "loss": 2.9679,
      "theoretical_loss": 3.6734470790321927,
      "tokens_seen": 933167104
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.000362246740220662,
      "loss": 2.7823,
      "theoretical_loss": 3.673422213652986,
      "tokens_seen": 933232640
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036223671013039117,
      "loss": 2.9388,
      "theoretical_loss": 3.6733973505087696,
      "tokens_seen": 933298176
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003622266800401204,
      "loss": 2.9991,
      "theoretical_loss": 3.673372489599185,
      "tokens_seen": 933363712
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036221664994984953,
      "loss": 2.8591,
      "theoretical_loss": 3.6733476309238746,
      "tokens_seen": 933429248
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036220661985957877,
      "loss": 2.9457,
      "theoretical_loss": 3.6733227744824815,
      "tokens_seen": 933494784
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003621965897693079,
      "loss": 2.803,
      "theoretical_loss": 3.6732979202746465,
      "tokens_seen": 933560320
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036218655967903713,
      "loss": 2.9959,
      "theoretical_loss": 3.6732730683000137,
      "tokens_seen": 933625856
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003621765295887663,
      "loss": 2.5503,
      "theoretical_loss": 3.673248218558225,
      "tokens_seen": 933691392
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003621664994984955,
      "loss": 2.8644,
      "theoretical_loss": 3.6732233710489233,
      "tokens_seen": 933756928
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036215646940822467,
      "loss": 2.8197,
      "theoretical_loss": 3.6731985257717508,
      "tokens_seen": 933822464
    },
    {
      "epoch": 0.28,
      "objective/train/advantage_avg": 0.49425435066223145,
      "objective/train/docs_used": 536516,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.8062825202941895,
      "objective/train/original_loss": 2.8062820434570312,
      "objective/train/theoretical_loss": 3.6731736827263513,
      "objective/train/tokens_used": 954348000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2443319410085678,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050667405128479,
      "objective/train/weighted_lm_loss": 2.9485347270965576,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0465034246444702,
      "theoretical_loss": 3.6731736827263513,
      "tokens_seen": 933888000
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036214643931795385,
      "loss": 2.8356,
      "theoretical_loss": 3.6731736827263513,
      "tokens_seen": 933888000
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036213640922768303,
      "loss": 2.8609,
      "theoretical_loss": 3.6731488419123672,
      "tokens_seen": 933953536
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036212637913741227,
      "loss": 2.7906,
      "theoretical_loss": 3.6731240033294412,
      "tokens_seen": 934019072
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003621163490471414,
      "loss": 3.1439,
      "theoretical_loss": 3.673099166977217,
      "tokens_seen": 934084608
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036210631895687063,
      "loss": 2.7855,
      "theoretical_loss": 3.6730743328553377,
      "tokens_seen": 934150144
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003620962888665998,
      "loss": 2.9678,
      "theoretical_loss": 3.6730495009634465,
      "tokens_seen": 934215680
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.000362086258776329,
      "loss": 2.9871,
      "theoretical_loss": 3.673024671301186,
      "tokens_seen": 934281216
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003620762286860582,
      "loss": 2.6872,
      "theoretical_loss": 3.6729998438682006,
      "tokens_seen": 934346752
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036206619859578736,
      "loss": 2.9686,
      "theoretical_loss": 3.6729750186641335,
      "tokens_seen": 934412288
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036205616850551654,
      "loss": 2.8078,
      "theoretical_loss": 3.672950195688628,
      "tokens_seen": 934477824
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003620461384152458,
      "loss": 2.7694,
      "theoretical_loss": 3.6729253749413284,
      "tokens_seen": 934543360
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003620361083249749,
      "loss": 3.0724,
      "theoretical_loss": 3.6729005564218777,
      "tokens_seen": 934608896
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036202607823470414,
      "loss": 2.727,
      "theoretical_loss": 3.6728757401299204,
      "tokens_seen": 934674432
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036201604814443326,
      "loss": 2.6455,
      "theoretical_loss": 3.6728509260651006,
      "tokens_seen": 934739968
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003620060180541625,
      "loss": 2.9586,
      "theoretical_loss": 3.6728261142270613,
      "tokens_seen": 934805504
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003619959879638917,
      "loss": 2.8624,
      "theoretical_loss": 3.672801304615447,
      "tokens_seen": 934871040
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036198595787362086,
      "loss": 3.089,
      "theoretical_loss": 3.672776497229902,
      "tokens_seen": 934936576
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036197592778335004,
      "loss": 2.9986,
      "theoretical_loss": 3.6727516920700705,
      "tokens_seen": 935002112
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003619658976930792,
      "loss": 2.8869,
      "theoretical_loss": 3.672726889135597,
      "tokens_seen": 935067648
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003619558676028084,
      "loss": 2.704,
      "theoretical_loss": 3.6727020884261257,
      "tokens_seen": 935133184
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036194583751253764,
      "loss": 2.7659,
      "theoretical_loss": 3.6726772899413014,
      "tokens_seen": 935198720
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036193580742226677,
      "loss": 2.6931,
      "theoretical_loss": 3.6726524936807676,
      "tokens_seen": 935264256
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.000361925777331996,
      "loss": 2.8477,
      "theoretical_loss": 3.6726276996441705,
      "tokens_seen": 935329792
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003619157472417252,
      "loss": 2.6839,
      "theoretical_loss": 3.672602907831154,
      "tokens_seen": 935395328
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036190571715145436,
      "loss": 2.7247,
      "theoretical_loss": 3.672578118241363,
      "tokens_seen": 935460864
    },
    {
      "epoch": 0.28,
      "objective/train/advantage_avg": 0.4944755733013153,
      "objective/train/docs_used": 537462,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.673536777496338,
      "objective/train/original_loss": 2.673537015914917,
      "objective/train/theoretical_loss": 3.6725533308744422,
      "objective/train/tokens_used": 955986400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24458403885364532,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050690770149231,
      "objective/train/weighted_lm_loss": 2.809121608734131,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 3.6725533308744422,
      "tokens_seen": 935526400
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036189568706118354,
      "loss": 2.8459,
      "theoretical_loss": 3.6725533308744422,
      "tokens_seen": 935526400
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003618856569709127,
      "loss": 2.9334,
      "theoretical_loss": 3.6725285457300374,
      "tokens_seen": 935591936
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003618756268806419,
      "loss": 2.8771,
      "theoretical_loss": 3.672503762807793,
      "tokens_seen": 935657472
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036186559679037114,
      "loss": 2.7552,
      "theoretical_loss": 3.6724789821073536,
      "tokens_seen": 935723008
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003618555667001003,
      "loss": 3.0693,
      "theoretical_loss": 3.672454203628365,
      "tokens_seen": 935788544
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003618455366098295,
      "loss": 2.8586,
      "theoretical_loss": 3.6724294273704725,
      "tokens_seen": 935854080
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003618355065195587,
      "loss": 2.7809,
      "theoretical_loss": 3.672404653333322,
      "tokens_seen": 935919616
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036182547642928787,
      "loss": 3.1065,
      "theoretical_loss": 3.6723798815165583,
      "tokens_seen": 935985152
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003618154463390171,
      "loss": 3.0155,
      "theoretical_loss": 3.672355111919827,
      "tokens_seen": 936050688
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036180541624874623,
      "loss": 2.801,
      "theoretical_loss": 3.672330344542774,
      "tokens_seen": 936116224
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036179538615847546,
      "loss": 2.8443,
      "theoretical_loss": 3.672305579385045,
      "tokens_seen": 936181760
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036178535606820464,
      "loss": 2.8018,
      "theoretical_loss": 3.6722808164462855,
      "tokens_seen": 936247296
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003617753259779338,
      "loss": 3.069,
      "theoretical_loss": 3.672256055726142,
      "tokens_seen": 936312832
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.000361765295887663,
      "loss": 2.9681,
      "theoretical_loss": 3.6722312972242594,
      "tokens_seen": 936378368
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003617552657973922,
      "loss": 2.8807,
      "theoretical_loss": 3.672206540940284,
      "tokens_seen": 936443904
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036174523570712137,
      "loss": 2.765,
      "theoretical_loss": 3.672181786873863,
      "tokens_seen": 936509440
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003617352056168506,
      "loss": 2.9878,
      "theoretical_loss": 3.6721570350246417,
      "tokens_seen": 936574976
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036172517552657973,
      "loss": 2.8684,
      "theoretical_loss": 3.6721322853922667,
      "tokens_seen": 936640512
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036171514543630897,
      "loss": 2.8275,
      "theoretical_loss": 3.672107537976384,
      "tokens_seen": 936706048
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003617051153460381,
      "loss": 3.0285,
      "theoretical_loss": 3.67208279277664,
      "tokens_seen": 936771584
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036169508525576733,
      "loss": 3.0426,
      "theoretical_loss": 3.672058049792682,
      "tokens_seen": 936837120
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003616850551654965,
      "loss": 3.0636,
      "theoretical_loss": 3.6720333090241555,
      "tokens_seen": 936902656
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003616750250752257,
      "loss": 2.7115,
      "theoretical_loss": 3.672008570470708,
      "tokens_seen": 936968192
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036166499498495487,
      "loss": 2.9089,
      "theoretical_loss": 3.6719838341319866,
      "tokens_seen": 937033728
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036165496489468405,
      "loss": 2.8859,
      "theoretical_loss": 3.671959100007637,
      "tokens_seen": 937099264
    },
    {
      "epoch": 0.28,
      "objective/train/advantage_avg": 0.4939897954463959,
      "objective/train/docs_used": 538123,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.981351375579834,
      "objective/train/original_loss": 2.981350898742676,
      "objective/train/theoretical_loss": 3.6719343680973067,
      "objective/train/tokens_used": 957624800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24408851563930511,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506397485733032,
      "objective/train/weighted_lm_loss": 3.132274627685547,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0417574644088745,
      "theoretical_loss": 3.6719343680973067,
      "tokens_seen": 937164800
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036164493480441323,
      "loss": 3.088,
      "theoretical_loss": 3.6719343680973067,
      "tokens_seen": 937164800
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036163490471414247,
      "loss": 2.8825,
      "theoretical_loss": 3.671909638400643,
      "tokens_seen": 937230336
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003616248746238716,
      "loss": 2.9158,
      "theoretical_loss": 3.671884910917292,
      "tokens_seen": 937295872
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036161484453360083,
      "loss": 3.1223,
      "theoretical_loss": 3.6718601856469024,
      "tokens_seen": 937361408
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036160481444333,
      "loss": 2.7772,
      "theoretical_loss": 3.6718354625891205,
      "tokens_seen": 937426944
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003615947843530592,
      "loss": 2.8309,
      "theoretical_loss": 3.671810741743594,
      "tokens_seen": 937492480
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003615847542627884,
      "loss": 2.996,
      "theoretical_loss": 3.6717860231099695,
      "tokens_seen": 937558016
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036157472417251756,
      "loss": 2.746,
      "theoretical_loss": 3.6717613066878956,
      "tokens_seen": 937623552
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036156469408224674,
      "loss": 2.9696,
      "theoretical_loss": 3.671736592477019,
      "tokens_seen": 937689088
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.000361554663991976,
      "loss": 2.9877,
      "theoretical_loss": 3.6717118804769875,
      "tokens_seen": 937754624
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003615446339017051,
      "loss": 2.9039,
      "theoretical_loss": 3.67168717068745,
      "tokens_seen": 937820160
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036153460381143434,
      "loss": 2.855,
      "theoretical_loss": 3.6716624631080528,
      "tokens_seen": 937885696
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036152457372116346,
      "loss": 2.679,
      "theoretical_loss": 3.671637757738444,
      "tokens_seen": 937951232
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003615145436308927,
      "loss": 2.9766,
      "theoretical_loss": 3.671613054578273,
      "tokens_seen": 938016768
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003615045135406219,
      "loss": 3.1712,
      "theoretical_loss": 3.671588353627186,
      "tokens_seen": 938082304
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036149448345035106,
      "loss": 2.664,
      "theoretical_loss": 3.6715636548848325,
      "tokens_seen": 938147840
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036148445336008024,
      "loss": 2.6672,
      "theoretical_loss": 3.6715389583508595,
      "tokens_seen": 938213376
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003614744232698094,
      "loss": 3.1893,
      "theoretical_loss": 3.6715142640249168,
      "tokens_seen": 938278912
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003614643931795386,
      "loss": 2.6902,
      "theoretical_loss": 3.6714895719066516,
      "tokens_seen": 938344448
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036145436308926784,
      "loss": 3.1657,
      "theoretical_loss": 3.6714648819957123,
      "tokens_seen": 938409984
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036144433299899697,
      "loss": 2.8316,
      "theoretical_loss": 3.6714401942917485,
      "tokens_seen": 938475520
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003614343029087262,
      "loss": 2.727,
      "theoretical_loss": 3.6714155087944085,
      "tokens_seen": 938541056
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003614242728184554,
      "loss": 2.9503,
      "theoretical_loss": 3.67139082550334,
      "tokens_seen": 938606592
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036141424272818456,
      "loss": 2.9838,
      "theoretical_loss": 3.671366144418193,
      "tokens_seen": 938672128
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036140421263791374,
      "loss": 2.7244,
      "theoretical_loss": 3.6713414655386156,
      "tokens_seen": 938737664
    },
    {
      "epoch": 0.28,
      "objective/train/advantage_avg": 0.4926060438156128,
      "objective/train/docs_used": 539283,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.9779231548309326,
      "objective/train/original_loss": 2.9779233932495117,
      "objective/train/theoretical_loss": 3.6713167888642575,
      "objective/train/tokens_used": 959263200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24277104437351227,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.05049467086792,
      "objective/train/weighted_lm_loss": 3.12852144241333,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0408108234405518,
      "theoretical_loss": 3.6713167888642575,
      "tokens_seen": 938803200
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003613941825476429,
      "loss": 3.0112,
      "theoretical_loss": 3.6713167888642575,
      "tokens_seen": 938803200
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003613841524573721,
      "loss": 2.9318,
      "theoretical_loss": 3.6712921143947668,
      "tokens_seen": 938868736
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036137412236710134,
      "loss": 2.8343,
      "theoretical_loss": 3.671267442129793,
      "tokens_seen": 938934272
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036136409227683047,
      "loss": 3.049,
      "theoretical_loss": 3.6712427720689855,
      "tokens_seen": 938999808
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003613540621865597,
      "loss": 2.9216,
      "theoretical_loss": 3.6712181042119933,
      "tokens_seen": 939065344
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036134403209628883,
      "loss": 2.9742,
      "theoretical_loss": 3.6711934385584666,
      "tokens_seen": 939130880
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036133400200601807,
      "loss": 2.6858,
      "theoretical_loss": 3.671168775108053,
      "tokens_seen": 939196416
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036132397191574725,
      "loss": 2.9586,
      "theoretical_loss": 3.6711441138604037,
      "tokens_seen": 939261952
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036131394182547643,
      "loss": 2.7265,
      "theoretical_loss": 3.6711194548151678,
      "tokens_seen": 939327488
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003613039117352056,
      "loss": 2.8468,
      "theoretical_loss": 3.671094797971995,
      "tokens_seen": 939393024
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036129388164493484,
      "loss": 2.9549,
      "theoretical_loss": 3.6710701433305344,
      "tokens_seen": 939458560
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036128385155466397,
      "loss": 2.9086,
      "theoretical_loss": 3.6710454908904366,
      "tokens_seen": 939524096
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003612738214643932,
      "loss": 2.5473,
      "theoretical_loss": 3.671020840651351,
      "tokens_seen": 939589632
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036126379137412233,
      "loss": 3.2493,
      "theoretical_loss": 3.6709961926129284,
      "tokens_seen": 939655168
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036125376128385157,
      "loss": 2.8769,
      "theoretical_loss": 3.670971546774818,
      "tokens_seen": 939720704
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036124373119358075,
      "loss": 2.8278,
      "theoretical_loss": 3.67094690313667,
      "tokens_seen": 939786240
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036123370110330993,
      "loss": 2.9992,
      "theoretical_loss": 3.6709222616981356,
      "tokens_seen": 939851776
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003612236710130391,
      "loss": 3.0023,
      "theoretical_loss": 3.670897622458864,
      "tokens_seen": 939917312
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003612136409227683,
      "loss": 2.8862,
      "theoretical_loss": 3.670872985418506,
      "tokens_seen": 939982848
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003612036108324975,
      "loss": 2.8499,
      "theoretical_loss": 3.670848350576712,
      "tokens_seen": 940048384
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003611935807422267,
      "loss": 2.7971,
      "theoretical_loss": 3.670823717933133,
      "tokens_seen": 940113920
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036118355065195584,
      "loss": 3.0286,
      "theoretical_loss": 3.6707990874874192,
      "tokens_seen": 940179456
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036117352056168507,
      "loss": 3.1045,
      "theoretical_loss": 3.670774459239221,
      "tokens_seen": 940244992
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003611634904714142,
      "loss": 2.9726,
      "theoretical_loss": 3.6707498331881903,
      "tokens_seen": 940310528
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.00036115346038114343,
      "loss": 2.7632,
      "theoretical_loss": 3.670725209333976,
      "tokens_seen": 940376064
    },
    {
      "epoch": 0.28,
      "objective/train/advantage_avg": 0.4957437217235565,
      "objective/train/docs_used": 539932,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.6353485584259033,
      "objective/train/original_loss": 2.6353487968444824,
      "objective/train/theoretical_loss": 3.6707005876762313,
      "objective/train/tokens_used": 960901600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24581576883792877,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0508239269256592,
      "objective/train/weighted_lm_loss": 2.769211530685425,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0253151655197144,
      "theoretical_loss": 3.6707005876762313,
      "tokens_seen": 940441600
    },
    {
      "epoch": 0.28,
      "learning_rate": 0.0003611434302908726,
      "loss": 2.7616,
      "theoretical_loss": 3.6707005876762313,
      "tokens_seen": 940441600
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003611334002006018,
      "loss": 2.6578,
      "theoretical_loss": 3.670675968214606,
      "tokens_seen": 940507136
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.000361123370110331,
      "loss": 2.7026,
      "theoretical_loss": 3.6706513509487513,
      "tokens_seen": 940572672
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003611133400200602,
      "loss": 3.0025,
      "theoretical_loss": 3.6706267358783187,
      "tokens_seen": 940638208
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003611033099297894,
      "loss": 2.8423,
      "theoretical_loss": 3.6706021230029586,
      "tokens_seen": 940703744
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003610932798395186,
      "loss": 2.9028,
      "theoretical_loss": 3.670577512322324,
      "tokens_seen": 940769280
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00036108324974924776,
      "loss": 2.8523,
      "theoretical_loss": 3.6705529038360645,
      "tokens_seen": 940834816
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00036107321965897694,
      "loss": 3.2359,
      "theoretical_loss": 3.670528297543833,
      "tokens_seen": 940900352
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003610631895687062,
      "loss": 3.002,
      "theoretical_loss": 3.6705036934452804,
      "tokens_seen": 940965888
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003610531594784353,
      "loss": 2.953,
      "theoretical_loss": 3.6704790915400585,
      "tokens_seen": 941031424
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00036104312938816454,
      "loss": 2.8934,
      "theoretical_loss": 3.6704544918278192,
      "tokens_seen": 941096960
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00036103309929789366,
      "loss": 2.9312,
      "theoretical_loss": 3.670429894308214,
      "tokens_seen": 941162496
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003610230692076229,
      "loss": 2.9316,
      "theoretical_loss": 3.670405298980895,
      "tokens_seen": 941228032
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003610130391173521,
      "loss": 2.9907,
      "theoretical_loss": 3.6703807058455142,
      "tokens_seen": 941293568
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00036100300902708126,
      "loss": 3.0,
      "theoretical_loss": 3.6703561149017236,
      "tokens_seen": 941359104
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00036099297893681044,
      "loss": 2.7706,
      "theoretical_loss": 3.670331526149176,
      "tokens_seen": 941424640
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003609829488465396,
      "loss": 2.9677,
      "theoretical_loss": 3.670306939587522,
      "tokens_seen": 941490176
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003609729187562688,
      "loss": 2.8227,
      "theoretical_loss": 3.6702823552164157,
      "tokens_seen": 941555712
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00036096288866599804,
      "loss": 2.8056,
      "theoretical_loss": 3.6702577730355084,
      "tokens_seen": 941621248
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00036095285857572717,
      "loss": 2.9214,
      "theoretical_loss": 3.670233193044453,
      "tokens_seen": 941686784
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003609428284854564,
      "loss": 2.7877,
      "theoretical_loss": 3.6702086152429017,
      "tokens_seen": 941752320
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003609327983951856,
      "loss": 2.9107,
      "theoretical_loss": 3.670184039630507,
      "tokens_seen": 941817856
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00036092276830491476,
      "loss": 2.8464,
      "theoretical_loss": 3.670159466206923,
      "tokens_seen": 941883392
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00036091273821464394,
      "loss": 2.9239,
      "theoretical_loss": 3.6701348949718002,
      "tokens_seen": 941948928
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003609027081243731,
      "loss": 2.7735,
      "theoretical_loss": 3.670110325924793,
      "tokens_seen": 942014464
    },
    {
      "epoch": 0.29,
      "objective/train/advantage_avg": 0.494937002658844,
      "objective/train/docs_used": 541311,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.673893928527832,
      "objective/train/original_loss": 2.6738932132720947,
      "objective/train/theoretical_loss": 3.6700857590655547,
      "objective/train/tokens_used": 962540000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24501048028469086,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507392883300781,
      "objective/train/weighted_lm_loss": 2.809519052505493,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0363596677780151,
      "theoretical_loss": 3.6700857590655547,
      "tokens_seen": 942080000
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003608926780341023,
      "loss": 2.7166,
      "theoretical_loss": 3.6700857590655547,
      "tokens_seen": 942080000
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00036088264794383154,
      "loss": 3.0083,
      "theoretical_loss": 3.6700611943937367,
      "tokens_seen": 942145536
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00036087261785356067,
      "loss": 3.1133,
      "theoretical_loss": 3.670036631908993,
      "tokens_seen": 942211072
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003608625877632899,
      "loss": 2.9167,
      "theoretical_loss": 3.6700120716109774,
      "tokens_seen": 942276608
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00036085255767301903,
      "loss": 3.0794,
      "theoretical_loss": 3.6699875134993425,
      "tokens_seen": 942342144
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00036084252758274827,
      "loss": 2.8742,
      "theoretical_loss": 3.6699629575737416,
      "tokens_seen": 942407680
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00036083249749247745,
      "loss": 2.7008,
      "theoretical_loss": 3.669938403833828,
      "tokens_seen": 942473216
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00036082246740220663,
      "loss": 2.9868,
      "theoretical_loss": 3.6699138522792554,
      "tokens_seen": 942538752
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003608124373119358,
      "loss": 2.7533,
      "theoretical_loss": 3.6698893029096773,
      "tokens_seen": 942604288
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00036080240722166505,
      "loss": 2.9279,
      "theoretical_loss": 3.6698647557247472,
      "tokens_seen": 942669824
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00036079237713139417,
      "loss": 2.7526,
      "theoretical_loss": 3.6698402107241193,
      "tokens_seen": 942735360
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003607823470411234,
      "loss": 3.0715,
      "theoretical_loss": 3.6698156679074474,
      "tokens_seen": 942800896
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00036077231695085253,
      "loss": 2.9683,
      "theoretical_loss": 3.6697911272743844,
      "tokens_seen": 942866432
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00036076228686058177,
      "loss": 3.1206,
      "theoretical_loss": 3.6697665888245856,
      "tokens_seen": 942931968
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00036075225677031095,
      "loss": 2.635,
      "theoretical_loss": 3.6697420525577042,
      "tokens_seen": 942997504
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00036074222668004013,
      "loss": 2.7532,
      "theoretical_loss": 3.6697175184733943,
      "tokens_seen": 943063040
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003607321965897693,
      "loss": 2.8491,
      "theoretical_loss": 3.6696929865713104,
      "tokens_seen": 943128576
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003607221664994985,
      "loss": 2.7781,
      "theoretical_loss": 3.669668456851107,
      "tokens_seen": 943194112
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003607121364092277,
      "loss": 2.9209,
      "theoretical_loss": 3.6696439293124383,
      "tokens_seen": 943259648
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003607021063189569,
      "loss": 2.7977,
      "theoretical_loss": 3.6696194039549583,
      "tokens_seen": 943325184
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00036069207622868604,
      "loss": 3.1096,
      "theoretical_loss": 3.669594880778322,
      "tokens_seen": 943390720
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00036068204613841527,
      "loss": 2.8776,
      "theoretical_loss": 3.669570359782184,
      "tokens_seen": 943456256
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003606720160481444,
      "loss": 2.9595,
      "theoretical_loss": 3.6695458409661983,
      "tokens_seen": 943521792
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00036066198595787364,
      "loss": 2.719,
      "theoretical_loss": 3.66952132433002,
      "tokens_seen": 943587328
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003606519558676028,
      "loss": 2.9329,
      "theoretical_loss": 3.669496809873305,
      "tokens_seen": 943652864
    },
    {
      "epoch": 0.29,
      "objective/train/advantage_avg": 0.49430620670318604,
      "objective/train/docs_used": 542104,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.1689090728759766,
      "objective/train/original_loss": 3.1689088344573975,
      "objective/train/theoretical_loss": 3.6694722975957066,
      "objective/train/tokens_used": 964178400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24439860880374908,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506728887557983,
      "objective/train/weighted_lm_loss": 3.329526901245117,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0363596677780151,
      "theoretical_loss": 3.6694722975957066,
      "tokens_seen": 943718400
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.000360641925777332,
      "loss": 3.0907,
      "theoretical_loss": 3.6694722975957066,
      "tokens_seen": 943718400
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003606318956870612,
      "loss": 2.9078,
      "theoretical_loss": 3.6694477874968805,
      "tokens_seen": 943783936
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003606218655967904,
      "loss": 2.9835,
      "theoretical_loss": 3.669423279576482,
      "tokens_seen": 943849472
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00036061183550651954,
      "loss": 2.9547,
      "theoretical_loss": 3.669398773834166,
      "tokens_seen": 943915008
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003606018054162488,
      "loss": 3.1779,
      "theoretical_loss": 3.6693742702695875,
      "tokens_seen": 943980544
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003605917753259779,
      "loss": 3.1172,
      "theoretical_loss": 3.6693497688824026,
      "tokens_seen": 944046080
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00036058174523570714,
      "loss": 2.8392,
      "theoretical_loss": 3.6693252696722656,
      "tokens_seen": 944111616
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003605717151454363,
      "loss": 2.7023,
      "theoretical_loss": 3.6693007726388327,
      "tokens_seen": 944177152
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003605616850551655,
      "loss": 2.6991,
      "theoretical_loss": 3.669276277781759,
      "tokens_seen": 944242688
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003605516549648947,
      "loss": 2.9923,
      "theoretical_loss": 3.6692517851007,
      "tokens_seen": 944308224
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00036054162487462386,
      "loss": 2.7856,
      "theoretical_loss": 3.6692272945953124,
      "tokens_seen": 944373760
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00036053159478435304,
      "loss": 2.8218,
      "theoretical_loss": 3.669202806265251,
      "tokens_seen": 944439296
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003605215646940823,
      "loss": 2.7229,
      "theoretical_loss": 3.6691783201101718,
      "tokens_seen": 944504832
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003605115346038114,
      "loss": 2.853,
      "theoretical_loss": 3.6691538361297313,
      "tokens_seen": 944570368
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00036050150451354064,
      "loss": 2.8437,
      "theoretical_loss": 3.6691293543235846,
      "tokens_seen": 944635904
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00036049147442326977,
      "loss": 2.8237,
      "theoretical_loss": 3.6691048746913886,
      "tokens_seen": 944701440
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.000360481444332999,
      "loss": 3.0866,
      "theoretical_loss": 3.6690803972327988,
      "tokens_seen": 944766976
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003604714142427282,
      "loss": 2.9667,
      "theoretical_loss": 3.669055921947472,
      "tokens_seen": 944832512
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00036046138415245737,
      "loss": 2.9816,
      "theoretical_loss": 3.669031448835064,
      "tokens_seen": 944898048
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00036045135406218655,
      "loss": 2.6633,
      "theoretical_loss": 3.669006977895232,
      "tokens_seen": 944963584
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003604413239719158,
      "loss": 2.742,
      "theoretical_loss": 3.668982509127632,
      "tokens_seen": 945029120
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003604312938816449,
      "loss": 2.887,
      "theoretical_loss": 3.66895804253192,
      "tokens_seen": 945094656
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00036042126379137414,
      "loss": 2.7359,
      "theoretical_loss": 3.6689335781077537,
      "tokens_seen": 945160192
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00036041123370110327,
      "loss": 2.9376,
      "theoretical_loss": 3.6689091158547886,
      "tokens_seen": 945225728
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003604012036108325,
      "loss": 3.047,
      "theoretical_loss": 3.6688846557726826,
      "tokens_seen": 945291264
    },
    {
      "epoch": 0.29,
      "objective/train/advantage_avg": 0.49485278129577637,
      "objective/train/docs_used": 543471,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.79586124420166,
      "objective/train/original_loss": 2.795861005783081,
      "objective/train/theoretical_loss": 3.6688601978610915,
      "objective/train/tokens_used": 965816800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24493557214736938,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507304668426514,
      "objective/train/weighted_lm_loss": 2.937467336654663,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0417574644088745,
      "theoretical_loss": 3.6688601978610915,
      "tokens_seen": 945356800
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003603911735205617,
      "loss": 2.8759,
      "theoretical_loss": 3.6688601978610915,
      "tokens_seen": 945356800
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00036038114343029087,
      "loss": 3.0301,
      "theoretical_loss": 3.668835742119674,
      "tokens_seen": 945422336
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00036037111334002005,
      "loss": 3.0272,
      "theoretical_loss": 3.6688112885480857,
      "tokens_seen": 945487872
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00036036108324974923,
      "loss": 2.8982,
      "theoretical_loss": 3.6687868371459835,
      "tokens_seen": 945553408
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00036035105315947847,
      "loss": 2.8103,
      "theoretical_loss": 3.6687623879130253,
      "tokens_seen": 945618944
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00036034102306920765,
      "loss": 2.8521,
      "theoretical_loss": 3.6687379408488683,
      "tokens_seen": 945684480
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00036033099297893683,
      "loss": 2.8498,
      "theoretical_loss": 3.668713495953169,
      "tokens_seen": 945750016
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.000360320962888666,
      "loss": 2.6674,
      "theoretical_loss": 3.6686890532255862,
      "tokens_seen": 945815552
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00036031093279839525,
      "loss": 2.8064,
      "theoretical_loss": 3.6686646126657765,
      "tokens_seen": 945881088
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00036030090270812437,
      "loss": 2.9598,
      "theoretical_loss": 3.6686401742733974,
      "tokens_seen": 945946624
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003602908726178536,
      "loss": 2.7748,
      "theoretical_loss": 3.668615738048107,
      "tokens_seen": 946012160
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00036028084252758273,
      "loss": 2.8725,
      "theoretical_loss": 3.6685913039895635,
      "tokens_seen": 946077696
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00036027081243731197,
      "loss": 2.8178,
      "theoretical_loss": 3.668566872097423,
      "tokens_seen": 946143232
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00036026078234704115,
      "loss": 2.7844,
      "theoretical_loss": 3.6685424423713444,
      "tokens_seen": 946208768
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00036025075225677033,
      "loss": 3.0786,
      "theoretical_loss": 3.6685180148109864,
      "tokens_seen": 946274304
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003602407221664995,
      "loss": 2.8115,
      "theoretical_loss": 3.6684935894160056,
      "tokens_seen": 946339840
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003602306920762287,
      "loss": 2.8269,
      "theoretical_loss": 3.6684691661860613,
      "tokens_seen": 946405376
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003602206619859579,
      "loss": 3.0386,
      "theoretical_loss": 3.6684447451208113,
      "tokens_seen": 946470912
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003602106318956871,
      "loss": 2.8051,
      "theoretical_loss": 3.6684203262199135,
      "tokens_seen": 946536448
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00036020060180541624,
      "loss": 2.8452,
      "theoretical_loss": 3.668395909483026,
      "tokens_seen": 946601984
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003601905717151455,
      "loss": 2.9254,
      "theoretical_loss": 3.6683714949098087,
      "tokens_seen": 946667520
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003601805416248746,
      "loss": 3.0526,
      "theoretical_loss": 3.6683470824999187,
      "tokens_seen": 946733056
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00036017051153460384,
      "loss": 2.6715,
      "theoretical_loss": 3.668322672253015,
      "tokens_seen": 946798592
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.000360160481444333,
      "loss": 2.8445,
      "theoretical_loss": 3.6682982641687563,
      "tokens_seen": 946864128
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003601504513540622,
      "loss": 2.9374,
      "theoretical_loss": 3.6682738582468013,
      "tokens_seen": 946929664
    },
    {
      "epoch": 0.29,
      "objective/train/advantage_avg": 0.4953087270259857,
      "objective/train/docs_used": 544140,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.904677629470825,
      "objective/train/original_loss": 2.9046778678894043,
      "objective/train/theoretical_loss": 3.668249454486809,
      "objective/train/tokens_used": 967455200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24539422988891602,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507783889770508,
      "objective/train/weighted_lm_loss": 3.052062749862671,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 3.668249454486809,
      "tokens_seen": 946995200
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003601404212637914,
      "loss": 2.9164,
      "theoretical_loss": 3.668249454486809,
      "tokens_seen": 946995200
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003601303911735206,
      "loss": 3.1429,
      "theoretical_loss": 3.6682250528884377,
      "tokens_seen": 947060736
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00036012036108324974,
      "loss": 2.9727,
      "theoretical_loss": 3.668200653451347,
      "tokens_seen": 947126272
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.000360110330992979,
      "loss": 2.932,
      "theoretical_loss": 3.6681762561751956,
      "tokens_seen": 947191808
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003601003009027081,
      "loss": 3.0624,
      "theoretical_loss": 3.668151861059643,
      "tokens_seen": 947257344
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00036009027081243734,
      "loss": 2.9286,
      "theoretical_loss": 3.6681274681043483,
      "tokens_seen": 947322880
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003600802407221665,
      "loss": 3.072,
      "theoretical_loss": 3.6681030773089702,
      "tokens_seen": 947388416
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003600702106318957,
      "loss": 3.0062,
      "theoretical_loss": 3.6680786886731687,
      "tokens_seen": 947453952
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003600601805416249,
      "loss": 2.7653,
      "theoretical_loss": 3.6680543021966026,
      "tokens_seen": 947519488
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00036005015045135406,
      "loss": 2.7027,
      "theoretical_loss": 3.668029917878932,
      "tokens_seen": 947585024
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00036004012036108324,
      "loss": 3.0416,
      "theoretical_loss": 3.6680055357198165,
      "tokens_seen": 947650560
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003600300902708125,
      "loss": 2.6508,
      "theoretical_loss": 3.6679811557189153,
      "tokens_seen": 947716096
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003600200601805416,
      "loss": 2.9634,
      "theoretical_loss": 3.667956777875889,
      "tokens_seen": 947781632
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00036001003009027084,
      "loss": 2.7576,
      "theoretical_loss": 3.667932402190396,
      "tokens_seen": 947847168
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035999999999999997,
      "loss": 2.6475,
      "theoretical_loss": 3.6679080286620973,
      "tokens_seen": 947912704
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003599899699097292,
      "loss": 2.8717,
      "theoretical_loss": 3.667883657290653,
      "tokens_seen": 947978240
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003599799398194584,
      "loss": 2.8419,
      "theoretical_loss": 3.6678592880757224,
      "tokens_seen": 948043776
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035996990972918757,
      "loss": 2.8146,
      "theoretical_loss": 3.667834921016966,
      "tokens_seen": 948109312
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035995987963891675,
      "loss": 2.5884,
      "theoretical_loss": 3.6678105561140435,
      "tokens_seen": 948174848
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.000359949849548646,
      "loss": 2.9922,
      "theoretical_loss": 3.6677861933666156,
      "tokens_seen": 948240384
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003599398194583751,
      "loss": 2.8775,
      "theoretical_loss": 3.6677618327743433,
      "tokens_seen": 948305920
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035992978936810434,
      "loss": 3.0922,
      "theoretical_loss": 3.667737474336886,
      "tokens_seen": 948371456
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035991975927783347,
      "loss": 2.8939,
      "theoretical_loss": 3.667713118053905,
      "tokens_seen": 948436992
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003599097291875627,
      "loss": 2.612,
      "theoretical_loss": 3.66768876392506,
      "tokens_seen": 948502528
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003598996990972919,
      "loss": 2.9247,
      "theoretical_loss": 3.6676644119500126,
      "tokens_seen": 948568064
    },
    {
      "epoch": 0.29,
      "objective/train/advantage_avg": 0.49585816264152527,
      "objective/train/docs_used": 544543,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.619891881942749,
      "objective/train/original_loss": 2.61989164352417,
      "objective/train/theoretical_loss": 3.667640062128423,
      "objective/train/tokens_used": 969093600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24591238796710968,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0508358478546143,
      "objective/train/weighted_lm_loss": 2.752969741821289,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.044721245765686,
      "theoretical_loss": 3.667640062128423,
      "tokens_seen": 948633600
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035988966900702107,
      "loss": 2.9553,
      "theoretical_loss": 3.667640062128423,
      "tokens_seen": 948633600
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035987963891675025,
      "loss": 2.9052,
      "theoretical_loss": 3.667615714459952,
      "tokens_seen": 948699136
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035986960882647943,
      "loss": 2.9139,
      "theoretical_loss": 3.6675913689442607,
      "tokens_seen": 948764672
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003598595787362086,
      "loss": 2.9107,
      "theoretical_loss": 3.66756702558101,
      "tokens_seen": 948830208
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035984954864593785,
      "loss": 2.7639,
      "theoretical_loss": 3.6675426843698613,
      "tokens_seen": 948895744
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.000359839518555667,
      "loss": 2.8825,
      "theoretical_loss": 3.6675183453104747,
      "tokens_seen": 948961280
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003598294884653962,
      "loss": 2.9653,
      "theoretical_loss": 3.667494008402513,
      "tokens_seen": 949026816
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035981945837512534,
      "loss": 2.7965,
      "theoretical_loss": 3.667469673645636,
      "tokens_seen": 949092352
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035980942828485457,
      "loss": 2.789,
      "theoretical_loss": 3.667445341039505,
      "tokens_seen": 949157888
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035979939819458375,
      "loss": 2.87,
      "theoretical_loss": 3.667421010583783,
      "tokens_seen": 949223424
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035978936810431293,
      "loss": 2.938,
      "theoretical_loss": 3.66739668227813,
      "tokens_seen": 949288960
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003597793380140421,
      "loss": 2.7842,
      "theoretical_loss": 3.6673723561222085,
      "tokens_seen": 949354496
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035976930792377135,
      "loss": 3.1367,
      "theoretical_loss": 3.6673480321156795,
      "tokens_seen": 949420032
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003597592778335005,
      "loss": 2.8667,
      "theoretical_loss": 3.667323710258205,
      "tokens_seen": 949485568
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003597492477432297,
      "loss": 2.9643,
      "theoretical_loss": 3.667299390549447,
      "tokens_seen": 949551104
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035973921765295884,
      "loss": 3.0231,
      "theoretical_loss": 3.6672750729890673,
      "tokens_seen": 949616640
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003597291875626881,
      "loss": 2.9421,
      "theoretical_loss": 3.6672507575767277,
      "tokens_seen": 949682176
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035971915747241726,
      "loss": 2.8639,
      "theoretical_loss": 3.6672264443120905,
      "tokens_seen": 949747712
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035970912738214644,
      "loss": 3.0734,
      "theoretical_loss": 3.6672021331948175,
      "tokens_seen": 949813248
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003596990972918756,
      "loss": 3.0936,
      "theoretical_loss": 3.667177824224571,
      "tokens_seen": 949878784
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003596890672016048,
      "loss": 3.0814,
      "theoretical_loss": 3.6671535174010135,
      "tokens_seen": 949944320
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.000359679037111334,
      "loss": 2.952,
      "theoretical_loss": 3.6671292127238067,
      "tokens_seen": 950009856
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003596690070210632,
      "loss": 3.1003,
      "theoretical_loss": 3.667104910192614,
      "tokens_seen": 950075392
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035965897693079234,
      "loss": 2.8838,
      "theoretical_loss": 3.6670806098070976,
      "tokens_seen": 950140928
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003596489468405216,
      "loss": 2.8805,
      "theoretical_loss": 3.6670563115669195,
      "tokens_seen": 950206464
    },
    {
      "epoch": 0.29,
      "objective/train/advantage_avg": 0.49384042620658875,
      "objective/train/docs_used": 545942,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.8101847171783447,
      "objective/train/original_loss": 2.8101844787597656,
      "objective/train/theoretical_loss": 3.667032015471743,
      "objective/train/tokens_used": 970732000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24397726356983185,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506242513656616,
      "objective/train/weighted_lm_loss": 2.9520795345306396,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0338950157165527,
      "theoretical_loss": 3.667032015471743,
      "tokens_seen": 950272000
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035963891675025076,
      "loss": 2.8042,
      "theoretical_loss": 3.667032015471743,
      "tokens_seen": 950272000
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035962888665997994,
      "loss": 2.9961,
      "theoretical_loss": 3.6670077215212302,
      "tokens_seen": 950337536
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003596188565697091,
      "loss": 2.5972,
      "theoretical_loss": 3.666983429715045,
      "tokens_seen": 950403072
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003596088264794383,
      "loss": 2.6146,
      "theoretical_loss": 3.6669591400528487,
      "tokens_seen": 950468608
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035959879638916754,
      "loss": 2.9162,
      "theoretical_loss": 3.6669348525343057,
      "tokens_seen": 950534144
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003595887662988967,
      "loss": 2.6885,
      "theoretical_loss": 3.6669105671590785,
      "tokens_seen": 950599680
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003595787362086259,
      "loss": 2.8269,
      "theoretical_loss": 3.6668862839268304,
      "tokens_seen": 950665216
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003595687061183551,
      "loss": 2.8669,
      "theoretical_loss": 3.666862002837224,
      "tokens_seen": 950730752
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035955867602808426,
      "loss": 3.0139,
      "theoretical_loss": 3.666837723889923,
      "tokens_seen": 950796288
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035954864593781344,
      "loss": 2.8418,
      "theoretical_loss": 3.66681344708459,
      "tokens_seen": 950861824
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003595386158475427,
      "loss": 2.8773,
      "theoretical_loss": 3.66678917242089,
      "tokens_seen": 950927360
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003595285857572718,
      "loss": 2.9095,
      "theoretical_loss": 3.6667648998984856,
      "tokens_seen": 950992896
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035951855566700104,
      "loss": 3.0471,
      "theoretical_loss": 3.66674062951704,
      "tokens_seen": 951058432
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035950852557673017,
      "loss": 2.9345,
      "theoretical_loss": 3.6667163612762175,
      "tokens_seen": 951123968
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003594984954864594,
      "loss": 3.0282,
      "theoretical_loss": 3.666692095175682,
      "tokens_seen": 951189504
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003594884653961886,
      "loss": 2.8462,
      "theoretical_loss": 3.6666678312150958,
      "tokens_seen": 951255040
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035947843530591777,
      "loss": 2.864,
      "theoretical_loss": 3.6666435693941244,
      "tokens_seen": 951320576
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035946840521564695,
      "loss": 2.6474,
      "theoretical_loss": 3.6666193097124307,
      "tokens_seen": 951386112
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003594583751253762,
      "loss": 3.025,
      "theoretical_loss": 3.6665950521696793,
      "tokens_seen": 951451648
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003594483450351053,
      "loss": 2.9659,
      "theoretical_loss": 3.6665707967655345,
      "tokens_seen": 951517184
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035943831494483454,
      "loss": 2.5982,
      "theoretical_loss": 3.66654654349966,
      "tokens_seen": 951582720
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035942828485456367,
      "loss": 3.0173,
      "theoretical_loss": 3.6665222923717202,
      "tokens_seen": 951648256
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003594182547642929,
      "loss": 2.854,
      "theoretical_loss": 3.66649804338138,
      "tokens_seen": 951713792
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003594082246740221,
      "loss": 3.0918,
      "theoretical_loss": 3.6664737965283027,
      "tokens_seen": 951779328
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035939819458375127,
      "loss": 2.9132,
      "theoretical_loss": 3.6664495518121534,
      "tokens_seen": 951844864
    },
    {
      "epoch": 0.29,
      "objective/train/advantage_avg": 0.4949176609516144,
      "objective/train/docs_used": 546639,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.8294246196746826,
      "objective/train/original_loss": 2.8294246196746826,
      "objective/train/theoretical_loss": 3.6664253092325962,
      "objective/train/tokens_used": 972370400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24499675631523132,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507371425628662,
      "objective/train/weighted_lm_loss": 2.9730165004730225,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0442860126495361,
      "theoretical_loss": 3.6664253092325962,
      "tokens_seen": 951910400
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035938816449348045,
      "loss": 2.8081,
      "theoretical_loss": 3.6664253092325962,
      "tokens_seen": 951910400
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035937813440320963,
      "loss": 2.7999,
      "theoretical_loss": 3.6664010687892965,
      "tokens_seen": 951975936
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003593681043129388,
      "loss": 2.9372,
      "theoretical_loss": 3.666376830481919,
      "tokens_seen": 952041472
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035935807422266805,
      "loss": 2.8882,
      "theoretical_loss": 3.666352594310127,
      "tokens_seen": 952107008
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003593480441323972,
      "loss": 2.7655,
      "theoretical_loss": 3.666328360273588,
      "tokens_seen": 952172544
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003593380140421264,
      "loss": 2.829,
      "theoretical_loss": 3.6663041283719644,
      "tokens_seen": 952238080
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035932798395185554,
      "loss": 2.714,
      "theoretical_loss": 3.6662798986049223,
      "tokens_seen": 952303616
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035931795386158477,
      "loss": 2.8823,
      "theoretical_loss": 3.666255670972127,
      "tokens_seen": 952369152
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035930792377131395,
      "loss": 2.7362,
      "theoretical_loss": 3.666231445473243,
      "tokens_seen": 952434688
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035929789368104313,
      "loss": 2.8476,
      "theoretical_loss": 3.666207222107936,
      "tokens_seen": 952500224
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003592878635907723,
      "loss": 2.8995,
      "theoretical_loss": 3.666183000875871,
      "tokens_seen": 952565760
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035927783350050155,
      "loss": 2.6639,
      "theoretical_loss": 3.666158781776714,
      "tokens_seen": 952631296
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003592678034102307,
      "loss": 3.1473,
      "theoretical_loss": 3.66613456481013,
      "tokens_seen": 952696832
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003592577733199599,
      "loss": 2.9312,
      "theoretical_loss": 3.6661103499757846,
      "tokens_seen": 952762368
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035924774322968904,
      "loss": 2.8939,
      "theoretical_loss": 3.6660861372733438,
      "tokens_seen": 952827904
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003592377131394183,
      "loss": 2.8994,
      "theoretical_loss": 3.666061926702472,
      "tokens_seen": 952893440
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035922768304914746,
      "loss": 2.8985,
      "theoretical_loss": 3.6660377182628365,
      "tokens_seen": 952958976
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035921765295887664,
      "loss": 3.0815,
      "theoretical_loss": 3.6660135119541026,
      "tokens_seen": 953024512
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003592076228686058,
      "loss": 2.8175,
      "theoretical_loss": 3.6659893077759356,
      "tokens_seen": 953090048
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.000359197592778335,
      "loss": 2.9135,
      "theoretical_loss": 3.6659651057280023,
      "tokens_seen": 953155584
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003591875626880642,
      "loss": 3.1436,
      "theoretical_loss": 3.665940905809969,
      "tokens_seen": 953221120
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003591775325977934,
      "loss": 2.9653,
      "theoretical_loss": 3.6659167080215003,
      "tokens_seen": 953286656
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035916750250752254,
      "loss": 2.6592,
      "theoretical_loss": 3.665892512362264,
      "tokens_seen": 953352192
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003591574724172518,
      "loss": 2.8859,
      "theoretical_loss": 3.6658683188319263,
      "tokens_seen": 953417728
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035914744232698096,
      "loss": 2.5884,
      "theoretical_loss": 3.6658441274301525,
      "tokens_seen": 953483264
    },
    {
      "epoch": 0.29,
      "objective/train/advantage_avg": 0.49426060914993286,
      "objective/train/docs_used": 547886,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.4510951042175293,
      "objective/train/original_loss": 2.451094627380371,
      "objective/train/theoretical_loss": 3.66581993815661,
      "objective/train/tokens_used": 974008800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24438095092773438,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506683588027954,
      "objective/train/weighted_lm_loss": 2.575307607650757,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 3.66581993815661,
      "tokens_seen": 953548800
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035913741223671014,
      "loss": 2.7725,
      "theoretical_loss": 3.66581993815661,
      "tokens_seen": 953548800
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003591273821464393,
      "loss": 2.9941,
      "theoretical_loss": 3.6657957510109647,
      "tokens_seen": 953614336
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003591173520561685,
      "loss": 2.675,
      "theoretical_loss": 3.6657715659928836,
      "tokens_seen": 953679872
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003591073219658977,
      "loss": 2.6113,
      "theoretical_loss": 3.6657473831020333,
      "tokens_seen": 953745408
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003590972918756269,
      "loss": 2.9773,
      "theoretical_loss": 3.6657232023380804,
      "tokens_seen": 953810944
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035908726178535605,
      "loss": 2.8097,
      "theoretical_loss": 3.665699023700692,
      "tokens_seen": 953876480
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003590772316950853,
      "loss": 3.0145,
      "theoretical_loss": 3.6656748471895346,
      "tokens_seen": 953942016
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003590672016048144,
      "loss": 2.9863,
      "theoretical_loss": 3.6656506728042753,
      "tokens_seen": 954007552
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035905717151454364,
      "loss": 2.8556,
      "theoretical_loss": 3.6656265005445814,
      "tokens_seen": 954073088
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003590471414242728,
      "loss": 2.7848,
      "theoretical_loss": 3.6656023304101195,
      "tokens_seen": 954138624
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.000359037111334002,
      "loss": 2.7681,
      "theoretical_loss": 3.665578162400558,
      "tokens_seen": 954204160
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003590270812437312,
      "loss": 2.8232,
      "theoretical_loss": 3.665553996515562,
      "tokens_seen": 954269696
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035901705115346037,
      "loss": 3.0136,
      "theoretical_loss": 3.6655298327548014,
      "tokens_seen": 954335232
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035900702106318955,
      "loss": 2.9906,
      "theoretical_loss": 3.6655056711179412,
      "tokens_seen": 954400768
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003589969909729188,
      "loss": 2.8973,
      "theoretical_loss": 3.665481511604651,
      "tokens_seen": 954466304
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003589869608826479,
      "loss": 2.9414,
      "theoretical_loss": 3.665457354214597,
      "tokens_seen": 954531840
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035897693079237715,
      "loss": 2.6699,
      "theoretical_loss": 3.665433198947447,
      "tokens_seen": 954597376
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035896690070210633,
      "loss": 3.0302,
      "theoretical_loss": 3.6654090458028694,
      "tokens_seen": 954662912
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003589568706118355,
      "loss": 2.9723,
      "theoretical_loss": 3.665384894780532,
      "tokens_seen": 954728448
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003589468405215647,
      "loss": 3.1838,
      "theoretical_loss": 3.6653607458801014,
      "tokens_seen": 954793984
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035893681043129387,
      "loss": 2.7278,
      "theoretical_loss": 3.6653365991012468,
      "tokens_seen": 954859520
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035892678034102305,
      "loss": 2.857,
      "theoretical_loss": 3.6653124544436357,
      "tokens_seen": 954925056
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003589167502507523,
      "loss": 3.1473,
      "theoretical_loss": 3.665288311906936,
      "tokens_seen": 954990592
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003589067201604814,
      "loss": 2.9196,
      "theoretical_loss": 3.6652641714908167,
      "tokens_seen": 955056128
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035889669007021065,
      "loss": 3.112,
      "theoretical_loss": 3.6652400331949453,
      "tokens_seen": 955121664
    },
    {
      "epoch": 0.29,
      "objective/train/advantage_avg": 0.49475640058517456,
      "objective/train/docs_used": 548540,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.483546018600464,
      "objective/train/original_loss": 2.483546257019043,
      "objective/train/theoretical_loss": 3.66521589701899,
      "objective/train/tokens_used": 975647200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24487629532814026,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507203340530396,
      "objective/train/weighted_lm_loss": 2.609163999557495,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0363596677780151,
      "theoretical_loss": 3.66521589701899,
      "tokens_seen": 955187200
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003588866599799398,
      "loss": 2.859,
      "theoretical_loss": 3.66521589701899,
      "tokens_seen": 955187200
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.000358876629889669,
      "loss": 3.0289,
      "theoretical_loss": 3.66519176296262,
      "tokens_seen": 955252736
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003588665997993982,
      "loss": 3.0635,
      "theoretical_loss": 3.6651676310255032,
      "tokens_seen": 955318272
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003588565697091274,
      "loss": 3.0928,
      "theoretical_loss": 3.665143501207308,
      "tokens_seen": 955383808
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003588465396188566,
      "loss": 2.822,
      "theoretical_loss": 3.6651193735077032,
      "tokens_seen": 955449344
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035883650952858574,
      "loss": 3.0525,
      "theoretical_loss": 3.6650952479263577,
      "tokens_seen": 955514880
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035882647943831497,
      "loss": 3.0163,
      "theoretical_loss": 3.6650711244629406,
      "tokens_seen": 955580416
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035881644934804415,
      "loss": 2.766,
      "theoretical_loss": 3.6650470031171194,
      "tokens_seen": 955645952
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035880641925777333,
      "loss": 2.8767,
      "theoretical_loss": 3.665022883888564,
      "tokens_seen": 955711488
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003587963891675025,
      "loss": 3.045,
      "theoretical_loss": 3.6649987667769435,
      "tokens_seen": 955777024
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035878635907723175,
      "loss": 2.9258,
      "theoretical_loss": 3.6649746517819266,
      "tokens_seen": 955842560
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003587763289869609,
      "loss": 2.8719,
      "theoretical_loss": 3.6649505389031827,
      "tokens_seen": 955908096
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003587662988966901,
      "loss": 2.677,
      "theoretical_loss": 3.6649264281403804,
      "tokens_seen": 955973632
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035875626880641924,
      "loss": 2.7447,
      "theoretical_loss": 3.66490231949319,
      "tokens_seen": 956039168
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003587462387161485,
      "loss": 3.0102,
      "theoretical_loss": 3.6648782129612796,
      "tokens_seen": 956104704
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035873620862587766,
      "loss": 3.0406,
      "theoretical_loss": 3.66485410854432,
      "tokens_seen": 956170240
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035872617853560684,
      "loss": 2.9955,
      "theoretical_loss": 3.6648300062419796,
      "tokens_seen": 956235776
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.000358716148445336,
      "loss": 3.0038,
      "theoretical_loss": 3.664805906053928,
      "tokens_seen": 956301312
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003587061183550652,
      "loss": 3.1232,
      "theoretical_loss": 3.6647818079798364,
      "tokens_seen": 956366848
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003586960882647944,
      "loss": 2.763,
      "theoretical_loss": 3.6647577120193726,
      "tokens_seen": 956432384
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003586860581745236,
      "loss": 2.8673,
      "theoretical_loss": 3.6647336181722077,
      "tokens_seen": 956497920
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035867602808425274,
      "loss": 2.7927,
      "theoretical_loss": 3.6647095264380107,
      "tokens_seen": 956563456
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.000358665997993982,
      "loss": 2.9548,
      "theoretical_loss": 3.664685436816452,
      "tokens_seen": 956628992
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035865596790371116,
      "loss": 2.7299,
      "theoretical_loss": 3.6646613493072016,
      "tokens_seen": 956694528
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035864593781344034,
      "loss": 2.8413,
      "theoretical_loss": 3.6646372639099294,
      "tokens_seen": 956760064
    },
    {
      "epoch": 0.29,
      "objective/train/advantage_avg": 0.4939250349998474,
      "objective/train/docs_used": 549880,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.880634069442749,
      "objective/train/original_loss": 2.880634069442749,
      "objective/train/theoretical_loss": 3.664613180624306,
      "objective/train/tokens_used": 977285600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.244038388133049,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506330728530884,
      "objective/train/weighted_lm_loss": 3.0265934467315674,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 3.664613180624306,
      "tokens_seen": 956825600
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003586359077231695,
      "loss": 2.9144,
      "theoretical_loss": 3.664613180624306,
      "tokens_seen": 956825600
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003586258776328987,
      "loss": 2.7711,
      "theoretical_loss": 3.6645890994500014,
      "tokens_seen": 956891136
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003586158475426279,
      "loss": 2.8703,
      "theoretical_loss": 3.6645650203866857,
      "tokens_seen": 956956672
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003586058174523571,
      "loss": 2.8492,
      "theoretical_loss": 3.6645409434340293,
      "tokens_seen": 957022208
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035859578736208625,
      "loss": 2.57,
      "theoretical_loss": 3.6645168685917033,
      "tokens_seen": 957087744
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003585857572718155,
      "loss": 3.2411,
      "theoretical_loss": 3.6644927958593776,
      "tokens_seen": 957153280
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003585757271815446,
      "loss": 2.5939,
      "theoretical_loss": 3.6644687252367234,
      "tokens_seen": 957218816
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035856569709127384,
      "loss": 2.8945,
      "theoretical_loss": 3.6644446567234112,
      "tokens_seen": 957284352
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.000358555667001003,
      "loss": 2.5971,
      "theoretical_loss": 3.6644205903191107,
      "tokens_seen": 957349888
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003585456369107322,
      "loss": 2.9347,
      "theoretical_loss": 3.6643965260234945,
      "tokens_seen": 957415424
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003585356068204614,
      "loss": 3.0086,
      "theoretical_loss": 3.6643724638362327,
      "tokens_seen": 957480960
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035852557673019057,
      "loss": 3.0413,
      "theoretical_loss": 3.6643484037569958,
      "tokens_seen": 957546496
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035851554663991975,
      "loss": 2.7326,
      "theoretical_loss": 3.6643243457854555,
      "tokens_seen": 957612032
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.000358505516549649,
      "loss": 3.01,
      "theoretical_loss": 3.6643002899212824,
      "tokens_seen": 957677568
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003584954864593781,
      "loss": 2.9477,
      "theoretical_loss": 3.6642762361641488,
      "tokens_seen": 957743104
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035848545636910735,
      "loss": 3.0453,
      "theoretical_loss": 3.6642521845137246,
      "tokens_seen": 957808640
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035847542627883653,
      "loss": 2.8535,
      "theoretical_loss": 3.664228134969682,
      "tokens_seen": 957874176
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003584653961885657,
      "loss": 2.9996,
      "theoretical_loss": 3.664204087531693,
      "tokens_seen": 957939712
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003584553660982949,
      "loss": 2.8036,
      "theoretical_loss": 3.664180042199427,
      "tokens_seen": 958005248
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035844533600802407,
      "loss": 2.9954,
      "theoretical_loss": 3.664155998972558,
      "tokens_seen": 958070784
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035843530591775325,
      "loss": 2.831,
      "theoretical_loss": 3.6641319578507563,
      "tokens_seen": 958136320
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003584252758274825,
      "loss": 2.9285,
      "theoretical_loss": 3.664107918833694,
      "tokens_seen": 958201856
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003584152457372116,
      "loss": 2.9299,
      "theoretical_loss": 3.6640838819210426,
      "tokens_seen": 958267392
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035840521564694085,
      "loss": 2.9546,
      "theoretical_loss": 3.6640598471124743,
      "tokens_seen": 958332928
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035839518555667,
      "loss": 3.0603,
      "theoretical_loss": 3.664035814407661,
      "tokens_seen": 958398464
    },
    {
      "epoch": 0.29,
      "objective/train/advantage_avg": 0.4933430552482605,
      "objective/train/docs_used": 550706,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.875967502593994,
      "objective/train/original_loss": 2.8759679794311523,
      "objective/train/theoretical_loss": 3.6640117838062745,
      "objective/train/tokens_used": 978924000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2434692531824112,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505719184875488,
      "objective/train/weighted_lm_loss": 3.021599054336548,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0253151655197144,
      "theoretical_loss": 3.6640117838062745,
      "tokens_seen": 958464000
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003583851554663992,
      "loss": 2.9467,
      "theoretical_loss": 3.6640117838062745,
      "tokens_seen": 958464000
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003583751253761284,
      "loss": 2.857,
      "theoretical_loss": 3.6639877553079874,
      "tokens_seen": 958529536
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003583650952858576,
      "loss": 2.9074,
      "theoretical_loss": 3.6639637289124716,
      "tokens_seen": 958595072
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035835506519558676,
      "loss": 2.9638,
      "theoretical_loss": 3.6639397046193984,
      "tokens_seen": 958660608
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035834503510531594,
      "loss": 2.9495,
      "theoretical_loss": 3.663915682428442,
      "tokens_seen": 958726144
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003583350050150451,
      "loss": 2.5298,
      "theoretical_loss": 3.663891662339273,
      "tokens_seen": 958791680
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035832497492477435,
      "loss": 2.9032,
      "theoretical_loss": 3.663867644351565,
      "tokens_seen": 958857216
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003583149448345035,
      "loss": 2.9553,
      "theoretical_loss": 3.6638436284649902,
      "tokens_seen": 958922752
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003583049147442327,
      "loss": 2.9367,
      "theoretical_loss": 3.6638196146792215,
      "tokens_seen": 958988288
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003582948846539619,
      "loss": 3.1834,
      "theoretical_loss": 3.6637956029939307,
      "tokens_seen": 959053824
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003582848545636911,
      "loss": 2.9997,
      "theoretical_loss": 3.6637715934087915,
      "tokens_seen": 959119360
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035827482447342026,
      "loss": 2.861,
      "theoretical_loss": 3.6637475859234763,
      "tokens_seen": 959184896
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035826479438314944,
      "loss": 3.1196,
      "theoretical_loss": 3.663723580537658,
      "tokens_seen": 959250432
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003582547642928786,
      "loss": 2.9333,
      "theoretical_loss": 3.66369957725101,
      "tokens_seen": 959315968
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035824473420260786,
      "loss": 2.5524,
      "theoretical_loss": 3.6636755760632043,
      "tokens_seen": 959381504
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.000358234704112337,
      "loss": 2.8811,
      "theoretical_loss": 3.663651576973915,
      "tokens_seen": 959447040
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003582246740220662,
      "loss": 2.9512,
      "theoretical_loss": 3.6636275799828155,
      "tokens_seen": 959512576
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035821464393179535,
      "loss": 3.0708,
      "theoretical_loss": 3.663603585089578,
      "tokens_seen": 959578112
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003582046138415246,
      "loss": 2.9894,
      "theoretical_loss": 3.6635795922938765,
      "tokens_seen": 959643648
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035819458375125376,
      "loss": 2.941,
      "theoretical_loss": 3.663555601595384,
      "tokens_seen": 959709184
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035818455366098294,
      "loss": 2.8642,
      "theoretical_loss": 3.6635316129937747,
      "tokens_seen": 959774720
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003581745235707121,
      "loss": 2.7931,
      "theoretical_loss": 3.6635076264887223,
      "tokens_seen": 959840256
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035816449348044136,
      "loss": 2.9758,
      "theoretical_loss": 3.663483642079899,
      "tokens_seen": 959905792
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003581544633901705,
      "loss": 2.751,
      "theoretical_loss": 3.66345965976698,
      "tokens_seen": 959971328
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003581444332998997,
      "loss": 2.8532,
      "theoretical_loss": 3.6634356795496386,
      "tokens_seen": 960036864
    },
    {
      "epoch": 0.29,
      "objective/train/advantage_avg": 0.49330374598503113,
      "objective/train/docs_used": 551132,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.7352023124694824,
      "objective/train/original_loss": 2.7352025508880615,
      "objective/train/theoretical_loss": 3.663411701427548,
      "objective/train/tokens_used": 980562400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24351052939891815,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505681037902832,
      "objective/train/weighted_lm_loss": 2.873396635055542,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0396549701690674,
      "theoretical_loss": 3.663411701427548,
      "tokens_seen": 960102400
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035813440320962885,
      "loss": 3.146,
      "theoretical_loss": 3.663411701427548,
      "tokens_seen": 960102400
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003581243731193581,
      "loss": 2.8671,
      "theoretical_loss": 3.6633877254003835,
      "tokens_seen": 960167936
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035811434302908727,
      "loss": 3.1139,
      "theoretical_loss": 3.6633637514678177,
      "tokens_seen": 960233472
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035810431293881645,
      "loss": 2.9659,
      "theoretical_loss": 3.663339779629525,
      "tokens_seen": 960299008
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003580942828485457,
      "loss": 2.776,
      "theoretical_loss": 3.6633158098851806,
      "tokens_seen": 960364544
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003580842527582748,
      "loss": 2.897,
      "theoretical_loss": 3.6632918422344574,
      "tokens_seen": 960430080
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035807422266800404,
      "loss": 2.8986,
      "theoretical_loss": 3.66326787667703,
      "tokens_seen": 960495616
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003580641925777332,
      "loss": 2.7852,
      "theoretical_loss": 3.6632439132125736,
      "tokens_seen": 960561152
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003580541624874624,
      "loss": 2.8462,
      "theoretical_loss": 3.6632199518407615,
      "tokens_seen": 960626688
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003580441323971916,
      "loss": 2.8818,
      "theoretical_loss": 3.6631959925612696,
      "tokens_seen": 960692224
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035803410230692077,
      "loss": 3.1131,
      "theoretical_loss": 3.663172035373771,
      "tokens_seen": 960757760
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035802407221664995,
      "loss": 2.78,
      "theoretical_loss": 3.663148080277941,
      "tokens_seen": 960823296
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003580140421263792,
      "loss": 3.0144,
      "theoretical_loss": 3.663124127273454,
      "tokens_seen": 960888832
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003580040120361083,
      "loss": 2.7388,
      "theoretical_loss": 3.6631001763599853,
      "tokens_seen": 960954368
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035799398194583755,
      "loss": 2.8157,
      "theoretical_loss": 3.6630762275372097,
      "tokens_seen": 961019904
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035798395185556673,
      "loss": 2.9503,
      "theoretical_loss": 3.663052280804802,
      "tokens_seen": 961085440
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003579739217652959,
      "loss": 2.6766,
      "theoretical_loss": 3.663028336162437,
      "tokens_seen": 961150976
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003579638916750251,
      "loss": 2.9913,
      "theoretical_loss": 3.663004393609791,
      "tokens_seen": 961216512
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035795386158475427,
      "loss": 2.7444,
      "theoretical_loss": 3.6629804531465373,
      "tokens_seen": 961282048
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035794383149448345,
      "loss": 2.8392,
      "theoretical_loss": 3.662956514772352,
      "tokens_seen": 961347584
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003579338014042127,
      "loss": 2.9873,
      "theoretical_loss": 3.6629325784869113,
      "tokens_seen": 961413120
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003579237713139418,
      "loss": 2.6126,
      "theoretical_loss": 3.662908644289889,
      "tokens_seen": 961478656
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035791374122367105,
      "loss": 2.854,
      "theoretical_loss": 3.6628847121809613,
      "tokens_seen": 961544192
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003579037111334002,
      "loss": 2.8586,
      "theoretical_loss": 3.662860782159804,
      "tokens_seen": 961609728
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003578936810431294,
      "loss": 2.8207,
      "theoretical_loss": 3.662836854226092,
      "tokens_seen": 961675264
    },
    {
      "epoch": 0.29,
      "objective/train/advantage_avg": 0.49155303835868835,
      "objective/train/docs_used": 552317,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.1930103302001953,
      "objective/train/original_loss": 3.193009853363037,
      "objective/train/theoretical_loss": 3.662812928379502,
      "objective/train/tokens_used": 982200800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24172161519527435,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0503840446472168,
      "objective/train/weighted_lm_loss": 3.353759527206421,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0408108234405518,
      "theoretical_loss": 3.662812928379502,
      "tokens_seen": 961740800
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003578836509528586,
      "loss": 3.2146,
      "theoretical_loss": 3.662812928379502,
      "tokens_seen": 961740800
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003578736208625878,
      "loss": 2.6792,
      "theoretical_loss": 3.662789004619709,
      "tokens_seen": 961806336
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035786359077231696,
      "loss": 3.059,
      "theoretical_loss": 3.6627650829463887,
      "tokens_seen": 961871872
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035785356068204614,
      "loss": 2.9314,
      "theoretical_loss": 3.6627411633592173,
      "tokens_seen": 961937408
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003578435305917753,
      "loss": 2.8126,
      "theoretical_loss": 3.662717245857871,
      "tokens_seen": 962002944
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035783350050150455,
      "loss": 3.0876,
      "theoretical_loss": 3.6626933304420257,
      "tokens_seen": 962068480
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003578234704112337,
      "loss": 2.8722,
      "theoretical_loss": 3.6626694171113567,
      "tokens_seen": 962134016
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003578134403209629,
      "loss": 2.9346,
      "theoretical_loss": 3.662645505865541,
      "tokens_seen": 962199552
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003578034102306921,
      "loss": 2.8306,
      "theoretical_loss": 3.662621596704255,
      "tokens_seen": 962265088
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003577933801404213,
      "loss": 2.7103,
      "theoretical_loss": 3.662597689627175,
      "tokens_seen": 962330624
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035778335005015046,
      "loss": 3.0203,
      "theoretical_loss": 3.6625737846339765,
      "tokens_seen": 962396160
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035777331995987964,
      "loss": 2.7944,
      "theoretical_loss": 3.6625498817243365,
      "tokens_seen": 962461696
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003577632898696088,
      "loss": 2.7962,
      "theoretical_loss": 3.6625259808979322,
      "tokens_seen": 962527232
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035775325977933806,
      "loss": 3.0415,
      "theoretical_loss": 3.662502082154439,
      "tokens_seen": 962592768
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003577432296890672,
      "loss": 2.6446,
      "theoretical_loss": 3.6624781854935344,
      "tokens_seen": 962658304
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003577331995987964,
      "loss": 2.8908,
      "theoretical_loss": 3.6624542909148956,
      "tokens_seen": 962723840
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035772316950852555,
      "loss": 2.8383,
      "theoretical_loss": 3.6624303984181985,
      "tokens_seen": 962789376
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003577131394182548,
      "loss": 3.1216,
      "theoretical_loss": 3.6624065080031203,
      "tokens_seen": 962854912
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035770310932798396,
      "loss": 3.0457,
      "theoretical_loss": 3.662382619669338,
      "tokens_seen": 962920448
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035769307923771314,
      "loss": 3.0413,
      "theoretical_loss": 3.6623587334165286,
      "tokens_seen": 962985984
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003576830491474423,
      "loss": 2.8027,
      "theoretical_loss": 3.6623348492443686,
      "tokens_seen": 963051520
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035767301905717156,
      "loss": 3.0234,
      "theoretical_loss": 3.6623109671525365,
      "tokens_seen": 963117056
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003576629889669007,
      "loss": 2.7542,
      "theoretical_loss": 3.662287087140709,
      "tokens_seen": 963182592
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003576529588766299,
      "loss": 2.6845,
      "theoretical_loss": 3.6622632092085636,
      "tokens_seen": 963248128
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035764292878635905,
      "loss": 2.767,
      "theoretical_loss": 3.662239333355777,
      "tokens_seen": 963313664
    },
    {
      "epoch": 0.29,
      "objective/train/advantage_avg": 0.49493563175201416,
      "objective/train/docs_used": 552873,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.9472851753234863,
      "objective/train/original_loss": 2.9472854137420654,
      "objective/train/theoretical_loss": 3.662215459582027,
      "objective/train/tokens_used": 983839200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24499300122261047,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050739049911499,
      "objective/train/weighted_lm_loss": 3.0968527793884277,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0465034246444702,
      "theoretical_loss": 3.662215459582027,
      "tokens_seen": 963379200
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003576328986960883,
      "loss": 3.0909,
      "theoretical_loss": 3.662215459582027,
      "tokens_seen": 963379200
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035762286860581747,
      "loss": 3.1021,
      "theoretical_loss": 3.6621915878869915,
      "tokens_seen": 963444736
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035761283851554665,
      "loss": 2.8798,
      "theoretical_loss": 3.662167718270348,
      "tokens_seen": 963510272
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035760280842527583,
      "loss": 2.6455,
      "theoretical_loss": 3.6621438507317743,
      "tokens_seen": 963575808
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.000357592778335005,
      "loss": 3.0923,
      "theoretical_loss": 3.662119985270947,
      "tokens_seen": 963641344
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003575827482447342,
      "loss": 2.8867,
      "theoretical_loss": 3.6620961218875463,
      "tokens_seen": 963706880
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003575727181544634,
      "loss": 2.9584,
      "theoretical_loss": 3.662072260581248,
      "tokens_seen": 963772416
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035756268806419255,
      "loss": 2.7766,
      "theoretical_loss": 3.6620484013517305,
      "tokens_seen": 963837952
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003575526579739218,
      "loss": 2.7999,
      "theoretical_loss": 3.6620245441986725,
      "tokens_seen": 963903488
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003575426278836509,
      "loss": 2.8326,
      "theoretical_loss": 3.6620006891217516,
      "tokens_seen": 963969024
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035753259779338015,
      "loss": 3.0314,
      "theoretical_loss": 3.6619768361206466,
      "tokens_seen": 964034560
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035752256770310933,
      "loss": 2.6228,
      "theoretical_loss": 3.661952985195035,
      "tokens_seen": 964100096
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003575125376128385,
      "loss": 2.6728,
      "theoretical_loss": 3.661929136344596,
      "tokens_seen": 964165632
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003575025075225677,
      "loss": 3.1415,
      "theoretical_loss": 3.661905289569007,
      "tokens_seen": 964231168
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035749247743229693,
      "loss": 2.8888,
      "theoretical_loss": 3.661881444867947,
      "tokens_seen": 964296704
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035748244734202606,
      "loss": 2.7387,
      "theoretical_loss": 3.6618576022410947,
      "tokens_seen": 964362240
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003574724172517553,
      "loss": 2.957,
      "theoretical_loss": 3.6618337616881282,
      "tokens_seen": 964427776
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003574623871614844,
      "loss": 2.5649,
      "theoretical_loss": 3.661809923208727,
      "tokens_seen": 964493312
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035745235707121365,
      "loss": 2.781,
      "theoretical_loss": 3.66178608680257,
      "tokens_seen": 964558848
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035744232698094283,
      "loss": 2.9146,
      "theoretical_loss": 3.6617622524693347,
      "tokens_seen": 964624384
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.000357432296890672,
      "loss": 2.7463,
      "theoretical_loss": 3.6617384202087004,
      "tokens_seen": 964689920
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003574222668004012,
      "loss": 2.8023,
      "theoretical_loss": 3.661714590020347,
      "tokens_seen": 964755456
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003574122367101304,
      "loss": 3.098,
      "theoretical_loss": 3.6616907619039534,
      "tokens_seen": 964820992
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035740220661985956,
      "loss": 2.9499,
      "theoretical_loss": 3.6616669358591984,
      "tokens_seen": 964886528
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003573921765295888,
      "loss": 2.7347,
      "theoretical_loss": 3.661643111885761,
      "tokens_seen": 964952064
    },
    {
      "epoch": 0.29,
      "objective/train/advantage_avg": 0.49228546023368835,
      "objective/train/docs_used": 554313,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.091707468032837,
      "objective/train/original_loss": 3.091707229614258,
      "objective/train/theoretical_loss": 3.66161928998332,
      "objective/train/tokens_used": 985477600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.242436483502388,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0504608154296875,
      "objective/train/weighted_lm_loss": 3.2479567527770996,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0363596677780151,
      "theoretical_loss": 3.66161928998332,
      "tokens_seen": 965017600
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003573821464393179,
      "loss": 2.8151,
      "theoretical_loss": 3.66161928998332,
      "tokens_seen": 965017600
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035737211634904716,
      "loss": 2.8645,
      "theoretical_loss": 3.661595470151556,
      "tokens_seen": 965083136
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003573620862587763,
      "loss": 3.0017,
      "theoretical_loss": 3.6615716523901476,
      "tokens_seen": 965148672
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003573520561685055,
      "loss": 3.036,
      "theoretical_loss": 3.661547836698775,
      "tokens_seen": 965214208
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035734202607823475,
      "loss": 2.9399,
      "theoretical_loss": 3.6615240230771167,
      "tokens_seen": 965279744
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003573319959879639,
      "loss": 2.8105,
      "theoretical_loss": 3.661500211524853,
      "tokens_seen": 965345280
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003573219658976931,
      "loss": 2.9594,
      "theoretical_loss": 3.6614764020416635,
      "tokens_seen": 965410816
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003573119358074223,
      "loss": 2.5834,
      "theoretical_loss": 3.661452594627228,
      "tokens_seen": 965476352
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003573019057171515,
      "loss": 2.7787,
      "theoretical_loss": 3.6614287892812265,
      "tokens_seen": 965541888
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035729187562688066,
      "loss": 2.713,
      "theoretical_loss": 3.6614049860033386,
      "tokens_seen": 965607424
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035728184553660984,
      "loss": 2.5833,
      "theoretical_loss": 3.6613811847932443,
      "tokens_seen": 965672960
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.000357271815446339,
      "loss": 2.7105,
      "theoretical_loss": 3.6613573856506236,
      "tokens_seen": 965738496
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035726178535606826,
      "loss": 2.9531,
      "theoretical_loss": 3.661333588575157,
      "tokens_seen": 965804032
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003572517552657974,
      "loss": 2.7121,
      "theoretical_loss": 3.6613097935665246,
      "tokens_seen": 965869568
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003572417251755266,
      "loss": 2.7066,
      "theoretical_loss": 3.661286000624406,
      "tokens_seen": 965935104
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035723169508525575,
      "loss": 2.9674,
      "theoretical_loss": 3.6612622097484824,
      "tokens_seen": 966000640
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.000357221664994985,
      "loss": 2.4698,
      "theoretical_loss": 3.6612384209384343,
      "tokens_seen": 966066176
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035721163490471416,
      "loss": 2.8211,
      "theoretical_loss": 3.661214634193941,
      "tokens_seen": 966131712
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035720160481444334,
      "loss": 2.7148,
      "theoretical_loss": 3.6611908495146848,
      "tokens_seen": 966197248
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003571915747241725,
      "loss": 2.7911,
      "theoretical_loss": 3.6611670669003447,
      "tokens_seen": 966262784
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035718154463390176,
      "loss": 2.8821,
      "theoretical_loss": 3.661143286350602,
      "tokens_seen": 966328320
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003571715145436309,
      "loss": 2.7362,
      "theoretical_loss": 3.661119507865138,
      "tokens_seen": 966393856
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003571614844533601,
      "loss": 3.0129,
      "theoretical_loss": 3.661095731443633,
      "tokens_seen": 966459392
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035715145436308925,
      "loss": 2.9294,
      "theoretical_loss": 3.6610719570857677,
      "tokens_seen": 966524928
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003571414242728185,
      "loss": 2.8218,
      "theoretical_loss": 3.6610481847912233,
      "tokens_seen": 966590464
    },
    {
      "epoch": 0.29,
      "objective/train/advantage_avg": 0.4914894700050354,
      "objective/train/docs_used": 554946,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.652301788330078,
      "objective/train/original_loss": 2.652301549911499,
      "objective/train/theoretical_loss": 3.661024414559681,
      "objective/train/tokens_used": 987116000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24168215692043304,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0503774881362915,
      "objective/train/weighted_lm_loss": 2.785728931427002,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0363596677780151,
      "theoretical_loss": 3.661024414559681,
      "tokens_seen": 966656000
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035713139418254767,
      "loss": 2.991,
      "theoretical_loss": 3.661024414559681,
      "tokens_seen": 966656000
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035712136409227685,
      "loss": 3.01,
      "theoretical_loss": 3.6610006463908222,
      "tokens_seen": 966721536
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035711133400200603,
      "loss": 2.6954,
      "theoretical_loss": 3.6609768802843274,
      "tokens_seen": 966787072
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003571013039117352,
      "loss": 3.002,
      "theoretical_loss": 3.6609531162398783,
      "tokens_seen": 966852608
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003570912738214644,
      "loss": 2.9962,
      "theoretical_loss": 3.6609293542571564,
      "tokens_seen": 966918144
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003570812437311936,
      "loss": 3.1054,
      "theoretical_loss": 3.6609055943358424,
      "tokens_seen": 966983680
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035707121364092275,
      "loss": 2.8944,
      "theoretical_loss": 3.6608818364756184,
      "tokens_seen": 967049216
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.000357061183550652,
      "loss": 2.7955,
      "theoretical_loss": 3.6608580806761655,
      "tokens_seen": 967114752
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003570511534603811,
      "loss": 2.6985,
      "theoretical_loss": 3.6608343269371657,
      "tokens_seen": 967180288
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035704112337011035,
      "loss": 2.9037,
      "theoretical_loss": 3.6608105752583007,
      "tokens_seen": 967245824
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035703109327983953,
      "loss": 2.7965,
      "theoretical_loss": 3.660786825639252,
      "tokens_seen": 967311360
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003570210631895687,
      "loss": 2.7794,
      "theoretical_loss": 3.6607630780797016,
      "tokens_seen": 967376896
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003570110330992979,
      "loss": 2.8266,
      "theoretical_loss": 3.6607393325793316,
      "tokens_seen": 967442432
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035700100300902713,
      "loss": 2.8607,
      "theoretical_loss": 3.6607155891378236,
      "tokens_seen": 967507968
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035699097291875626,
      "loss": 2.835,
      "theoretical_loss": 3.6606918477548596,
      "tokens_seen": 967573504
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003569809428284855,
      "loss": 2.8811,
      "theoretical_loss": 3.660668108430122,
      "tokens_seen": 967639040
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003569709127382146,
      "loss": 2.8404,
      "theoretical_loss": 3.6606443711632926,
      "tokens_seen": 967704576
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035696088264794385,
      "loss": 2.7913,
      "theoretical_loss": 3.6606206359540545,
      "tokens_seen": 967770112
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035695085255767303,
      "loss": 2.8027,
      "theoretical_loss": 3.660596902802089,
      "tokens_seen": 967835648
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003569408224674022,
      "loss": 2.7159,
      "theoretical_loss": 3.660573171707079,
      "tokens_seen": 967901184
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003569307923771314,
      "loss": 2.6638,
      "theoretical_loss": 3.6605494426687075,
      "tokens_seen": 967966720
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003569207622868606,
      "loss": 3.0196,
      "theoretical_loss": 3.6605257156866555,
      "tokens_seen": 968032256
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035691073219658976,
      "loss": 2.9495,
      "theoretical_loss": 3.6605019907606073,
      "tokens_seen": 968097792
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.000356900702106319,
      "loss": 2.9338,
      "theoretical_loss": 3.6604782678902446,
      "tokens_seen": 968163328
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003568906720160481,
      "loss": 2.9878,
      "theoretical_loss": 3.6604545470752505,
      "tokens_seen": 968228864
    },
    {
      "epoch": 0.29,
      "objective/train/advantage_avg": 0.49482208490371704,
      "objective/train/docs_used": 556500,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.991096019744873,
      "objective/train/original_loss": 2.9910964965820312,
      "objective/train/theoretical_loss": 3.6604308283153073,
      "objective/train/tokens_used": 988754400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24490055441856384,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507272481918335,
      "objective/train/weighted_lm_loss": 3.1427412033081055,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0437887907028198,
      "theoretical_loss": 3.6604308283153073,
      "tokens_seen": 968294400
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035688064192577736,
      "loss": 2.9275,
      "theoretical_loss": 3.6604308283153073,
      "tokens_seen": 968294400
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003568706118355065,
      "loss": 2.5736,
      "theoretical_loss": 3.6604071116100987,
      "tokens_seen": 968359936
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003568605817452357,
      "loss": 2.8652,
      "theoretical_loss": 3.6603833969593067,
      "tokens_seen": 968425472
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003568505516549649,
      "loss": 3.1164,
      "theoretical_loss": 3.6603596843626156,
      "tokens_seen": 968491008
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003568405215646941,
      "loss": 2.9878,
      "theoretical_loss": 3.6603359738197074,
      "tokens_seen": 968556544
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035683049147442326,
      "loss": 2.9886,
      "theoretical_loss": 3.660312265330266,
      "tokens_seen": 968622080
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003568204613841525,
      "loss": 3.0731,
      "theoretical_loss": 3.660288558893974,
      "tokens_seen": 968687616
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003568104312938816,
      "loss": 2.8069,
      "theoretical_loss": 3.660264854510515,
      "tokens_seen": 968753152
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035680040120361086,
      "loss": 3.0118,
      "theoretical_loss": 3.660241152179572,
      "tokens_seen": 968818688
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035679037111334,
      "loss": 2.6405,
      "theoretical_loss": 3.66021745190083,
      "tokens_seen": 968884224
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003567803410230692,
      "loss": 3.142,
      "theoretical_loss": 3.6601937536739704,
      "tokens_seen": 968949760
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003567703109327984,
      "loss": 2.7867,
      "theoretical_loss": 3.6601700574986777,
      "tokens_seen": 969015296
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003567602808425276,
      "loss": 2.9079,
      "theoretical_loss": 3.6601463633746363,
      "tokens_seen": 969080832
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035675025075225677,
      "loss": 3.1063,
      "theoretical_loss": 3.6601226713015285,
      "tokens_seen": 969146368
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035674022066198595,
      "loss": 2.8994,
      "theoretical_loss": 3.660098981279039,
      "tokens_seen": 969211904
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035673019057171513,
      "loss": 2.7879,
      "theoretical_loss": 3.6600752933068517,
      "tokens_seen": 969277440
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035672016048144436,
      "loss": 3.1502,
      "theoretical_loss": 3.6600516073846503,
      "tokens_seen": 969342976
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003567101303911735,
      "loss": 2.8613,
      "theoretical_loss": 3.6600279235121196,
      "tokens_seen": 969408512
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003567001003009027,
      "loss": 2.949,
      "theoretical_loss": 3.660004241688942,
      "tokens_seen": 969474048
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003567001003009027,
      "loss": 2.771,
      "theoretical_loss": 3.6599805619148027,
      "tokens_seen": 969539584
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035669007021063185,
      "loss": 2.9987,
      "theoretical_loss": 3.6599568841893864,
      "tokens_seen": 969605120
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003566800401203611,
      "loss": 3.0104,
      "theoretical_loss": 3.6599332085123764,
      "tokens_seen": 969670656
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035667001003009027,
      "loss": 3.0228,
      "theoretical_loss": 3.6599095348834574,
      "tokens_seen": 969736192
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035665997993981945,
      "loss": 2.9709,
      "theoretical_loss": 3.659885863302314,
      "tokens_seen": 969801728
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035664994984954863,
      "loss": 2.6705,
      "theoretical_loss": 3.659862193768631,
      "tokens_seen": 969867264
    },
    {
      "epoch": 0.29,
      "objective/train/advantage_avg": 0.49326005578041077,
      "objective/train/docs_used": 557201,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.8820509910583496,
      "objective/train/original_loss": 2.8820509910583496,
      "objective/train/theoretical_loss": 3.6598385262820923,
      "objective/train/tokens_used": 990392800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24338312447071075,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050563097000122,
      "objective/train/weighted_lm_loss": 3.0275652408599854,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0437887907028198,
      "theoretical_loss": 3.6598385262820923,
      "tokens_seen": 969932800
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035663991975927787,
      "loss": 2.9358,
      "theoretical_loss": 3.6598385262820923,
      "tokens_seen": 969932800
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.000356629889669007,
      "loss": 2.9702,
      "theoretical_loss": 3.659814860842383,
      "tokens_seen": 969998336
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035661985957873623,
      "loss": 3.0834,
      "theoretical_loss": 3.659791197449188,
      "tokens_seen": 970063872
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035660982948846536,
      "loss": 2.9523,
      "theoretical_loss": 3.6597675361021906,
      "tokens_seen": 970129408
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003565997993981946,
      "loss": 3.1129,
      "theoretical_loss": 3.6597438768010777,
      "tokens_seen": 970194944
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003565897693079238,
      "loss": 2.9049,
      "theoretical_loss": 3.659720219545533,
      "tokens_seen": 970260480
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035657973921765295,
      "loss": 2.8365,
      "theoretical_loss": 3.659696564335242,
      "tokens_seen": 970326016
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003565697091273822,
      "loss": 2.9521,
      "theoretical_loss": 3.6596729111698894,
      "tokens_seen": 970391552
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003565596790371113,
      "loss": 2.8735,
      "theoretical_loss": 3.659649260049161,
      "tokens_seen": 970457088
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035654964894684055,
      "loss": 3.1717,
      "theoretical_loss": 3.659625610972741,
      "tokens_seen": 970522624
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035653961885656973,
      "loss": 2.9123,
      "theoretical_loss": 3.6596019639403146,
      "tokens_seen": 970588160
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003565295887662989,
      "loss": 2.7447,
      "theoretical_loss": 3.659578318951569,
      "tokens_seen": 970653696
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003565195586760281,
      "loss": 3.1139,
      "theoretical_loss": 3.6595546760061874,
      "tokens_seen": 970719232
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035650952858575733,
      "loss": 2.7534,
      "theoretical_loss": 3.6595310351038566,
      "tokens_seen": 970784768
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035649949849548646,
      "loss": 2.8166,
      "theoretical_loss": 3.6595073962442615,
      "tokens_seen": 970850304
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003564894684052157,
      "loss": 3.0432,
      "theoretical_loss": 3.659483759427088,
      "tokens_seen": 970915840
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003564794383149448,
      "loss": 2.7039,
      "theoretical_loss": 3.659460124652022,
      "tokens_seen": 970981376
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035646940822467405,
      "loss": 3.2684,
      "theoretical_loss": 3.6594364919187488,
      "tokens_seen": 971046912
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035645937813440323,
      "loss": 2.8356,
      "theoretical_loss": 3.6594128612269543,
      "tokens_seen": 971112448
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003564493480441324,
      "loss": 2.8664,
      "theoretical_loss": 3.659389232576325,
      "tokens_seen": 971177984
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003564393179538616,
      "loss": 2.7549,
      "theoretical_loss": 3.6593656059665465,
      "tokens_seen": 971243520
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003564292878635908,
      "loss": 3.0392,
      "theoretical_loss": 3.6593419813973043,
      "tokens_seen": 971309056
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035641925777331996,
      "loss": 3.0131,
      "theoretical_loss": 3.6593183588682847,
      "tokens_seen": 971374592
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003564092276830492,
      "loss": 2.994,
      "theoretical_loss": 3.6592947383791747,
      "tokens_seen": 971440128
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003563991975927783,
      "loss": 2.7219,
      "theoretical_loss": 3.6592711199296595,
      "tokens_seen": 971505664
    },
    {
      "epoch": 0.29,
      "objective/train/advantage_avg": 0.4951452314853668,
      "objective/train/docs_used": 558372,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.605295419692993,
      "objective/train/original_loss": 2.6052956581115723,
      "objective/train/theoretical_loss": 3.6592475035194267,
      "objective/train/tokens_used": 992031200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2452402561903,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507612228393555,
      "objective/train/weighted_lm_loss": 2.7374818325042725,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0304546356201172,
      "theoretical_loss": 3.6592475035194267,
      "tokens_seen": 971571200
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035638916750250756,
      "loss": 3.0577,
      "theoretical_loss": 3.6592475035194267,
      "tokens_seen": 971571200
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003563791374122367,
      "loss": 3.0573,
      "theoretical_loss": 3.659223889148161,
      "tokens_seen": 971636736
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003563691073219659,
      "loss": 2.9173,
      "theoretical_loss": 3.6592002768155503,
      "tokens_seen": 971702272
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003563590772316951,
      "loss": 3.3591,
      "theoretical_loss": 3.6591766665212804,
      "tokens_seen": 971767808
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003563490471414243,
      "loss": 3.0483,
      "theoretical_loss": 3.659153058265038,
      "tokens_seen": 971833344
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035633901705115346,
      "loss": 2.8579,
      "theoretical_loss": 3.6591294520465096,
      "tokens_seen": 971898880
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003563289869608827,
      "loss": 2.7179,
      "theoretical_loss": 3.6591058478653817,
      "tokens_seen": 971964416
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003563189568706118,
      "loss": 2.6519,
      "theoretical_loss": 3.6590822457213426,
      "tokens_seen": 972029952
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035630892678034106,
      "loss": 3.3105,
      "theoretical_loss": 3.6590586456140777,
      "tokens_seen": 972095488
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003562988966900702,
      "loss": 2.9313,
      "theoretical_loss": 3.659035047543274,
      "tokens_seen": 972161024
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003562888665997994,
      "loss": 2.9975,
      "theoretical_loss": 3.659011451508619,
      "tokens_seen": 972226560
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003562788365095286,
      "loss": 2.7743,
      "theoretical_loss": 3.6589878575098,
      "tokens_seen": 972292096
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003562688064192578,
      "loss": 3.1795,
      "theoretical_loss": 3.658964265546503,
      "tokens_seen": 972357632
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035625877632898697,
      "loss": 2.9516,
      "theoretical_loss": 3.6589406756184166,
      "tokens_seen": 972423168
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035624874623871615,
      "loss": 3.0186,
      "theoretical_loss": 3.6589170877252277,
      "tokens_seen": 972488704
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035623871614844533,
      "loss": 2.8315,
      "theoretical_loss": 3.658893501866623,
      "tokens_seen": 972554240
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035622868605817456,
      "loss": 3.0425,
      "theoretical_loss": 3.6588699180422903,
      "tokens_seen": 972619776
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003562186559679037,
      "loss": 2.8307,
      "theoretical_loss": 3.658846336251917,
      "tokens_seen": 972685312
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003562086258776329,
      "loss": 2.7643,
      "theoretical_loss": 3.658822756495191,
      "tokens_seen": 972750848
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035619859578736205,
      "loss": 2.9591,
      "theoretical_loss": 3.6587991787717997,
      "tokens_seen": 972816384
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003561885656970913,
      "loss": 2.6873,
      "theoretical_loss": 3.6587756030814313,
      "tokens_seen": 972881920
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035617853560682047,
      "loss": 2.9696,
      "theoretical_loss": 3.6587520294237725,
      "tokens_seen": 972947456
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035616850551654965,
      "loss": 2.8447,
      "theoretical_loss": 3.658728457798512,
      "tokens_seen": 973012992
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035615847542627883,
      "loss": 2.8854,
      "theoretical_loss": 3.658704888205337,
      "tokens_seen": 973078528
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035614844533600807,
      "loss": 2.9239,
      "theoretical_loss": 3.6586813206439364,
      "tokens_seen": 973144064
    },
    {
      "epoch": 0.29,
      "objective/train/advantage_avg": 0.4946426749229431,
      "objective/train/docs_used": 558840,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.355797529220581,
      "objective/train/original_loss": 3.355797529220581,
      "objective/train/theoretical_loss": 3.6586577551139974,
      "objective/train/tokens_used": 993669600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24475502967834473,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507084131240845,
      "objective/train/weighted_lm_loss": 3.526076555252075,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0338950157165527,
      "theoretical_loss": 3.6586577551139974,
      "tokens_seen": 973209600
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003561384152457372,
      "loss": 2.9783,
      "theoretical_loss": 3.6586577551139974,
      "tokens_seen": 973209600
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035612838515546643,
      "loss": 2.7518,
      "theoretical_loss": 3.6586341916152083,
      "tokens_seen": 973275136
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035611835506519556,
      "loss": 2.8337,
      "theoretical_loss": 3.6586106301472583,
      "tokens_seen": 973340672
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.0003561083249749248,
      "loss": 3.1615,
      "theoretical_loss": 3.6585870707098342,
      "tokens_seen": 973406208
    },
    {
      "epoch": 0.29,
      "learning_rate": 0.00035609829488465397,
      "loss": 2.9522,
      "theoretical_loss": 3.658563513302625,
      "tokens_seen": 973471744
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035608826479438315,
      "loss": 2.9133,
      "theoretical_loss": 3.658539957925319,
      "tokens_seen": 973537280
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035607823470411233,
      "loss": 3.1219,
      "theoretical_loss": 3.6585164045776044,
      "tokens_seen": 973602816
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003560682046138415,
      "loss": 3.1672,
      "theoretical_loss": 3.6584928532591703,
      "tokens_seen": 973668352
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003560581745235707,
      "loss": 2.9307,
      "theoretical_loss": 3.6584693039697047,
      "tokens_seen": 973733888
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035604814443329993,
      "loss": 2.7411,
      "theoretical_loss": 3.658445756708897,
      "tokens_seen": 973799424
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035603811434302906,
      "loss": 2.9884,
      "theoretical_loss": 3.6584222114764353,
      "tokens_seen": 973864960
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003560280842527583,
      "loss": 2.6719,
      "theoretical_loss": 3.6583986682720084,
      "tokens_seen": 973930496
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003560180541624874,
      "loss": 2.7728,
      "theoretical_loss": 3.6583751270953053,
      "tokens_seen": 973996032
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035600802407221666,
      "loss": 2.8124,
      "theoretical_loss": 3.6583515879460147,
      "tokens_seen": 974061568
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035599799398194584,
      "loss": 2.9457,
      "theoretical_loss": 3.658328050823826,
      "tokens_seen": 974127104
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.000355987963891675,
      "loss": 3.2697,
      "theoretical_loss": 3.658304515728429,
      "tokens_seen": 974192640
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003559779338014042,
      "loss": 3.0861,
      "theoretical_loss": 3.658280982659511,
      "tokens_seen": 974258176
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035596790371113344,
      "loss": 2.9481,
      "theoretical_loss": 3.6582574516167625,
      "tokens_seen": 974323712
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035595787362086256,
      "loss": 2.8762,
      "theoretical_loss": 3.658233922599873,
      "tokens_seen": 974389248
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003559478435305918,
      "loss": 3.0098,
      "theoretical_loss": 3.658210395608531,
      "tokens_seen": 974454784
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003559378134403209,
      "loss": 2.7771,
      "theoretical_loss": 3.658186870642426,
      "tokens_seen": 974520320
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035592778335005016,
      "loss": 2.9495,
      "theoretical_loss": 3.6581633477012483,
      "tokens_seen": 974585856
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035591775325977934,
      "loss": 2.9789,
      "theoretical_loss": 3.6581398267846863,
      "tokens_seen": 974651392
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003559077231695085,
      "loss": 2.799,
      "theoretical_loss": 3.6581163078924304,
      "tokens_seen": 974716928
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003558976930792377,
      "loss": 2.9762,
      "theoretical_loss": 3.65809279102417,
      "tokens_seen": 974782464
    },
    {
      "epoch": 0.3,
      "objective/train/advantage_avg": 0.4944697320461273,
      "objective/train/docs_used": 560118,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.480991840362549,
      "objective/train/original_loss": 2.480991840362549,
      "objective/train/theoretical_loss": 3.658069276179595,
      "objective/train/tokens_used": 995308000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24454887211322784,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506900548934937,
      "objective/train/weighted_lm_loss": 2.6067185401916504,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0454468727111816,
      "theoretical_loss": 3.658069276179595,
      "tokens_seen": 974848000
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003558876629889669,
      "loss": 2.6039,
      "theoretical_loss": 3.658069276179595,
      "tokens_seen": 974848000
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035587763289869607,
      "loss": 2.6761,
      "theoretical_loss": 3.6580457633583956,
      "tokens_seen": 974913536
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003558676028084253,
      "loss": 2.8862,
      "theoretical_loss": 3.658022252560261,
      "tokens_seen": 974979072
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035585757271815443,
      "loss": 2.9801,
      "theoretical_loss": 3.6579987437848813,
      "tokens_seen": 975044608
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035584754262788366,
      "loss": 2.9878,
      "theoretical_loss": 3.657975237031947,
      "tokens_seen": 975110144
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003558375125376129,
      "loss": 2.7534,
      "theoretical_loss": 3.657951732301148,
      "tokens_seen": 975175680
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.000355827482447342,
      "loss": 3.0862,
      "theoretical_loss": 3.657928229592174,
      "tokens_seen": 975241216
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035581745235707126,
      "loss": 2.8905,
      "theoretical_loss": 3.6579047289047155,
      "tokens_seen": 975306752
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003558074222668004,
      "loss": 2.9538,
      "theoretical_loss": 3.6578812302384636,
      "tokens_seen": 975372288
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003557973921765296,
      "loss": 3.0093,
      "theoretical_loss": 3.6578577335931075,
      "tokens_seen": 975437824
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003557873620862588,
      "loss": 2.892,
      "theoretical_loss": 3.657834238968338,
      "tokens_seen": 975503360
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.000355777331995988,
      "loss": 2.9776,
      "theoretical_loss": 3.657810746363846,
      "tokens_seen": 975568896
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035576730190571717,
      "loss": 3.1503,
      "theoretical_loss": 3.6577872557793216,
      "tokens_seen": 975634432
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035575727181544635,
      "loss": 2.8398,
      "theoretical_loss": 3.657763767214456,
      "tokens_seen": 975699968
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035574724172517553,
      "loss": 3.1983,
      "theoretical_loss": 3.657740280668939,
      "tokens_seen": 975765504
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035573721163490476,
      "loss": 2.8984,
      "theoretical_loss": 3.657716796142463,
      "tokens_seen": 975831040
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003557271815446339,
      "loss": 2.8288,
      "theoretical_loss": 3.657693313634717,
      "tokens_seen": 975896576
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003557171514543631,
      "loss": 2.9963,
      "theoretical_loss": 3.657669833145393,
      "tokens_seen": 975962112
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035570712136409225,
      "loss": 2.9753,
      "theoretical_loss": 3.6576463546741813,
      "tokens_seen": 976027648
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003556970912738215,
      "loss": 3.1272,
      "theoretical_loss": 3.6576228782207743,
      "tokens_seen": 976093184
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035568706118355067,
      "loss": 2.9427,
      "theoretical_loss": 3.6575994037848614,
      "tokens_seen": 976158720
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035567703109327985,
      "loss": 2.7927,
      "theoretical_loss": 3.657575931366135,
      "tokens_seen": 976224256
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035566700100300903,
      "loss": 3.0791,
      "theoretical_loss": 3.6575524609642858,
      "tokens_seen": 976289792
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035565697091273827,
      "loss": 3.0862,
      "theoretical_loss": 3.6575289925790053,
      "tokens_seen": 976355328
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003556469408224674,
      "loss": 3.1117,
      "theoretical_loss": 3.657505526209985,
      "tokens_seen": 976420864
    },
    {
      "epoch": 0.3,
      "objective/train/advantage_avg": 0.4937366545200348,
      "objective/train/docs_used": 560789,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.8850655555725098,
      "objective/train/original_loss": 2.885065793991089,
      "objective/train/theoretical_loss": 3.657482061856916,
      "objective/train/tokens_used": 996946400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24386052787303925,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050613284111023,
      "objective/train/weighted_lm_loss": 3.0311641693115234,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0417574644088745,
      "theoretical_loss": 3.657482061856916,
      "tokens_seen": 976486400
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035563691073219663,
      "loss": 3.0786,
      "theoretical_loss": 3.657482061856916,
      "tokens_seen": 976486400
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035562688064192576,
      "loss": 3.0497,
      "theoretical_loss": 3.65745859951949,
      "tokens_seen": 976551936
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.000355616850551655,
      "loss": 2.9553,
      "theoretical_loss": 3.657435139197399,
      "tokens_seen": 976617472
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035560682046138417,
      "loss": 2.9181,
      "theoretical_loss": 3.6574116808903345,
      "tokens_seen": 976683008
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035559679037111335,
      "loss": 3.022,
      "theoretical_loss": 3.657388224597988,
      "tokens_seen": 976748544
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035558676028084253,
      "loss": 2.656,
      "theoretical_loss": 3.657364770320051,
      "tokens_seen": 976814080
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003555767301905717,
      "loss": 3.0619,
      "theoretical_loss": 3.6573413180562158,
      "tokens_seen": 976879616
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003555667001003009,
      "loss": 2.8356,
      "theoretical_loss": 3.657317867806175,
      "tokens_seen": 976945152
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035555667001003013,
      "loss": 2.7914,
      "theoretical_loss": 3.6572944195696193,
      "tokens_seen": 977010688
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035554663991975926,
      "loss": 3.0482,
      "theoretical_loss": 3.6572709733462414,
      "tokens_seen": 977076224
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003555366098294885,
      "loss": 3.1004,
      "theoretical_loss": 3.6572475291357334,
      "tokens_seen": 977141760
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003555265797392176,
      "loss": 3.0336,
      "theoretical_loss": 3.657224086937788,
      "tokens_seen": 977207296
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035551654964894686,
      "loss": 2.9636,
      "theoretical_loss": 3.6572006467520968,
      "tokens_seen": 977272832
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035550651955867604,
      "loss": 3.066,
      "theoretical_loss": 3.657177208578352,
      "tokens_seen": 977338368
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003554964894684052,
      "loss": 3.0081,
      "theoretical_loss": 3.6571537724162466,
      "tokens_seen": 977403904
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003554864593781344,
      "loss": 2.5719,
      "theoretical_loss": 3.657130338265473,
      "tokens_seen": 977469440
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035547642928786364,
      "loss": 2.8774,
      "theoretical_loss": 3.6571069061257235,
      "tokens_seen": 977534976
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035546639919759276,
      "loss": 2.9468,
      "theoretical_loss": 3.6570834759966906,
      "tokens_seen": 977600512
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.000355456369107322,
      "loss": 2.8861,
      "theoretical_loss": 3.6570600478780673,
      "tokens_seen": 977666048
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003554463390170511,
      "loss": 2.9689,
      "theoretical_loss": 3.6570366217695467,
      "tokens_seen": 977731584
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035543630892678036,
      "loss": 3.0918,
      "theoretical_loss": 3.6570131976708202,
      "tokens_seen": 977797120
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035542627883650954,
      "loss": 2.7061,
      "theoretical_loss": 3.6569897755815823,
      "tokens_seen": 977862656
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003554162487462387,
      "loss": 2.5445,
      "theoretical_loss": 3.6569663555015257,
      "tokens_seen": 977928192
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003554062186559679,
      "loss": 2.8145,
      "theoretical_loss": 3.656942937430342,
      "tokens_seen": 977993728
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003553961885656971,
      "loss": 2.8467,
      "theoretical_loss": 3.656919521367726,
      "tokens_seen": 978059264
    },
    {
      "epoch": 0.3,
      "objective/train/advantage_avg": 0.49411216378211975,
      "objective/train/docs_used": 561934,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.8479368686676025,
      "objective/train/original_loss": 2.8479371070861816,
      "objective/train/theoretical_loss": 3.6568961073133703,
      "objective/train/tokens_used": 998584800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24420768022537231,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506526231765747,
      "objective/train/weighted_lm_loss": 2.992053985595703,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.038975715637207,
      "theoretical_loss": 3.6568961073133703,
      "tokens_seen": 978124800
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035538615847542627,
      "loss": 2.9781,
      "theoretical_loss": 3.6568961073133703,
      "tokens_seen": 978124800
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003553761283851555,
      "loss": 2.7887,
      "theoretical_loss": 3.656872695266968,
      "tokens_seen": 978190336
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035536609829488463,
      "loss": 3.0882,
      "theoretical_loss": 3.656849285228212,
      "tokens_seen": 978255872
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035535606820461386,
      "loss": 3.0231,
      "theoretical_loss": 3.6568258771967965,
      "tokens_seen": 978321408
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035534603811434304,
      "loss": 2.8634,
      "theoretical_loss": 3.656802471172414,
      "tokens_seen": 978386944
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003553360080240722,
      "loss": 3.177,
      "theoretical_loss": 3.656779067154759,
      "tokens_seen": 978452480
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003553259779338014,
      "loss": 3.0088,
      "theoretical_loss": 3.656755665143524,
      "tokens_seen": 978518016
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003553159478435306,
      "loss": 2.9621,
      "theoretical_loss": 3.656732265138403,
      "tokens_seen": 978583552
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035530591775325977,
      "loss": 2.7821,
      "theoretical_loss": 3.6567088671390904,
      "tokens_seen": 978649088
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.000355295887662989,
      "loss": 3.2106,
      "theoretical_loss": 3.6566854711452796,
      "tokens_seen": 978714624
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035528585757271813,
      "loss": 2.9919,
      "theoretical_loss": 3.656662077156664,
      "tokens_seen": 978780160
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035527582748244737,
      "loss": 2.745,
      "theoretical_loss": 3.656638685172937,
      "tokens_seen": 978845696
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003552657973921765,
      "loss": 2.8758,
      "theoretical_loss": 3.6566152951937942,
      "tokens_seen": 978911232
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035525576730190573,
      "loss": 2.7658,
      "theoretical_loss": 3.656591907218928,
      "tokens_seen": 978976768
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003552457372116349,
      "loss": 2.72,
      "theoretical_loss": 3.656568521248034,
      "tokens_seen": 979042304
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003552357071213641,
      "loss": 2.9501,
      "theoretical_loss": 3.6565451372808053,
      "tokens_seen": 979107840
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035522567703109327,
      "loss": 2.8646,
      "theoretical_loss": 3.6565217553169367,
      "tokens_seen": 979173376
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035521564694082245,
      "loss": 2.7555,
      "theoretical_loss": 3.6564983753561218,
      "tokens_seen": 979238912
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035520561685055163,
      "loss": 2.9355,
      "theoretical_loss": 3.6564749973980555,
      "tokens_seen": 979304448
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035519558676028087,
      "loss": 2.836,
      "theoretical_loss": 3.6564516214424323,
      "tokens_seen": 979369984
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035518555667001,
      "loss": 2.9995,
      "theoretical_loss": 3.6564282474889467,
      "tokens_seen": 979435520
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035517552657973923,
      "loss": 3.0851,
      "theoretical_loss": 3.6564048755372927,
      "tokens_seen": 979501056
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003551654964894684,
      "loss": 3.0175,
      "theoretical_loss": 3.6563815055871656,
      "tokens_seen": 979566592
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003551554663991976,
      "loss": 3.0827,
      "theoretical_loss": 3.65635813763826,
      "tokens_seen": 979632128
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003551454363089268,
      "loss": 2.8613,
      "theoretical_loss": 3.6563347716902697,
      "tokens_seen": 979697664
    },
    {
      "epoch": 0.3,
      "objective/train/advantage_avg": 0.4954160451889038,
      "objective/train/docs_used": 562605,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.7971866130828857,
      "objective/train/original_loss": 2.7971863746643066,
      "objective/train/theoretical_loss": 3.656311407742891,
      "objective/train/tokens_used": 1000223200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24549566209316254,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507895946502686,
      "objective/train/weighted_lm_loss": 2.9391658306121826,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0442860126495361,
      "theoretical_loss": 3.656311407742891,
      "tokens_seen": 979763200
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035513540621865596,
      "loss": 2.8517,
      "theoretical_loss": 3.656311407742891,
      "tokens_seen": 979763200
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035512537612838514,
      "loss": 2.8974,
      "theoretical_loss": 3.6562880457958182,
      "tokens_seen": 979828736
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035511534603811437,
      "loss": 2.8063,
      "theoretical_loss": 3.656264685848746,
      "tokens_seen": 979894272
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003551053159478435,
      "loss": 2.7139,
      "theoretical_loss": 3.6562413279013697,
      "tokens_seen": 979959808
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035509528585757273,
      "loss": 3.0372,
      "theoretical_loss": 3.6562179719533843,
      "tokens_seen": 980025344
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003550852557673019,
      "loss": 2.8854,
      "theoretical_loss": 3.656194618004485,
      "tokens_seen": 980090880
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003550752256770311,
      "loss": 2.989,
      "theoretical_loss": 3.6561712660543675,
      "tokens_seen": 980156416
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035506519558676033,
      "loss": 2.5661,
      "theoretical_loss": 3.656147916102726,
      "tokens_seen": 980221952
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035505516549648946,
      "loss": 2.8044,
      "theoretical_loss": 3.656124568149257,
      "tokens_seen": 980287488
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003550451354062187,
      "loss": 3.0738,
      "theoretical_loss": 3.6561012221936555,
      "tokens_seen": 980353024
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003550351053159478,
      "loss": 2.911,
      "theoretical_loss": 3.656077878235617,
      "tokens_seen": 980418560
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035502507522567706,
      "loss": 2.8706,
      "theoretical_loss": 3.6560545362748367,
      "tokens_seen": 980484096
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035501504513540624,
      "loss": 2.9034,
      "theoretical_loss": 3.6560311963110106,
      "tokens_seen": 980549632
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003550050150451354,
      "loss": 2.9946,
      "theoretical_loss": 3.656007858343835,
      "tokens_seen": 980615168
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003549949849548646,
      "loss": 2.8991,
      "theoretical_loss": 3.6559845223730045,
      "tokens_seen": 980680704
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035498495486459384,
      "loss": 2.9652,
      "theoretical_loss": 3.6559611883982157,
      "tokens_seen": 980746240
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035497492477432296,
      "loss": 2.9634,
      "theoretical_loss": 3.6559378564191647,
      "tokens_seen": 980811776
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003549648946840522,
      "loss": 3.2149,
      "theoretical_loss": 3.655914526435547,
      "tokens_seen": 980877312
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003549548645937813,
      "loss": 2.803,
      "theoretical_loss": 3.6558911984470583,
      "tokens_seen": 980942848
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035494483450351056,
      "loss": 2.9409,
      "theoretical_loss": 3.655867872453395,
      "tokens_seen": 981008384
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035493480441323974,
      "loss": 3.1622,
      "theoretical_loss": 3.6558445484542537,
      "tokens_seen": 981073920
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003549247743229689,
      "loss": 2.9347,
      "theoretical_loss": 3.6558212264493304,
      "tokens_seen": 981139456
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003549147442326981,
      "loss": 2.6974,
      "theoretical_loss": 3.6557979064383215,
      "tokens_seen": 981204992
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003549047141424273,
      "loss": 2.8607,
      "theoretical_loss": 3.6557745884209227,
      "tokens_seen": 981270528
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035489468405215647,
      "loss": 2.5517,
      "theoretical_loss": 3.6557512723968313,
      "tokens_seen": 981336064
    },
    {
      "epoch": 0.3,
      "objective/train/advantage_avg": 0.49321675300598145,
      "objective/train/docs_used": 563033,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.681551218032837,
      "objective/train/original_loss": 2.681551218032837,
      "objective/train/theoretical_loss": 3.6557279583657434,
      "objective/train/tokens_used": 1001861600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24335360527038574,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505585670471191,
      "objective/train/weighted_lm_loss": 2.8171279430389404,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0408108234405518,
      "theoretical_loss": 3.6557279583657434,
      "tokens_seen": 981401600
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003548846539618857,
      "loss": 2.8362,
      "theoretical_loss": 3.6557279583657434,
      "tokens_seen": 981401600
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035487462387161483,
      "loss": 2.8743,
      "theoretical_loss": 3.6557046463273557,
      "tokens_seen": 981467136
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035486459378134406,
      "loss": 2.8229,
      "theoretical_loss": 3.655681336281365,
      "tokens_seen": 981532672
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035485456369107324,
      "loss": 2.978,
      "theoretical_loss": 3.6556580282274673,
      "tokens_seen": 981598208
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003548445336008024,
      "loss": 2.8789,
      "theoretical_loss": 3.6556347221653605,
      "tokens_seen": 981663744
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003548345035105316,
      "loss": 3.0429,
      "theoretical_loss": 3.65561141809474,
      "tokens_seen": 981729280
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003548244734202608,
      "loss": 3.0656,
      "theoretical_loss": 3.655588116015304,
      "tokens_seen": 981794816
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035481444332998997,
      "loss": 2.8457,
      "theoretical_loss": 3.6555648159267484,
      "tokens_seen": 981860352
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003548044132397192,
      "loss": 2.6512,
      "theoretical_loss": 3.6555415178287713,
      "tokens_seen": 981925888
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035479438314944833,
      "loss": 2.8668,
      "theoretical_loss": 3.655518221721069,
      "tokens_seen": 981991424
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035478435305917757,
      "loss": 3.0236,
      "theoretical_loss": 3.655494927603339,
      "tokens_seen": 982056960
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003547743229689067,
      "loss": 2.9456,
      "theoretical_loss": 3.655471635475279,
      "tokens_seen": 982122496
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035476429287863593,
      "loss": 3.0275,
      "theoretical_loss": 3.655448345336585,
      "tokens_seen": 982188032
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003547542627883651,
      "loss": 2.9741,
      "theoretical_loss": 3.6554250571869558,
      "tokens_seen": 982253568
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003547442326980943,
      "loss": 2.9864,
      "theoretical_loss": 3.6554017710260878,
      "tokens_seen": 982319104
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035473420260782347,
      "loss": 2.9288,
      "theoretical_loss": 3.655378486853679,
      "tokens_seen": 982384640
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035472417251755265,
      "loss": 3.2372,
      "theoretical_loss": 3.655355204669427,
      "tokens_seen": 982450176
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035471414242728183,
      "loss": 2.9866,
      "theoretical_loss": 3.655331924473029,
      "tokens_seen": 982515712
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035470411233701107,
      "loss": 2.7407,
      "theoretical_loss": 3.6553086462641833,
      "tokens_seen": 982581248
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003546940822467402,
      "loss": 2.8436,
      "theoretical_loss": 3.6552853700425874,
      "tokens_seen": 982646784
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035468405215646943,
      "loss": 2.9558,
      "theoretical_loss": 3.655262095807939,
      "tokens_seen": 982712320
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003546740220661986,
      "loss": 3.009,
      "theoretical_loss": 3.6552388235599356,
      "tokens_seen": 982777856
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003546639919759278,
      "loss": 2.646,
      "theoretical_loss": 3.655215553298276,
      "tokens_seen": 982843392
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.000354653961885657,
      "loss": 2.9478,
      "theoretical_loss": 3.655192285022658,
      "tokens_seen": 982908928
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035464393179538616,
      "loss": 2.6803,
      "theoretical_loss": 3.655169018732779,
      "tokens_seen": 982974464
    },
    {
      "debugging/Self-BLEU-5": 0.2700858122828388,
      "debugging/distinct-1-grams": 0.785191637630662,
      "debugging/distinct-2-grams": 0.9938271604938271,
      "debugging/entropy-1-grams": 4.521436327200685,
      "debugging/entropy-2-grams": 5.001393331688789,
      "debugging/length": 569.0,
      "debugging/num_segments": 2,
      "debugging/raw_token_scores_avg": 0.006858057342469692,
      "debugging/raw_token_scores_std": 0.010178866796195507,
      "debugging/score": 0.001652892561983471,
      "debugging/score_std": 0.001652892561983471,
      "epoch": 0.3,
      "objective/train/advantage_avg": 0.4931398928165436,
      "objective/train/docs_used": 564216,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.7921459674835205,
      "objective/train/original_loss": 2.7921454906463623,
      "objective/train/theoretical_loss": 3.6551457544283386,
      "objective/train/tokens_used": 1003500000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24329060316085815,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505506992340088,
      "objective/train/weighted_lm_loss": 2.9333269596099854,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0425469875335693,
      "theoretical_loss": 3.6551457544283386,
      "tokens_seen": 983040000
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035463390170511534,
      "loss": 2.936,
      "theoretical_loss": 3.6551457544283386,
      "tokens_seen": 983040000
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035462387161484457,
      "loss": 3.0299,
      "theoretical_loss": 3.655122492109034,
      "tokens_seen": 983105536
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003546138415245737,
      "loss": 2.9496,
      "theoretical_loss": 3.655099231774563,
      "tokens_seen": 983171072
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035460381143430293,
      "loss": 2.815,
      "theoretical_loss": 3.6550759734246245,
      "tokens_seen": 983236608
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035459378134403206,
      "loss": 2.7902,
      "theoretical_loss": 3.6550527170589175,
      "tokens_seen": 983302144
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003545837512537613,
      "loss": 2.8957,
      "theoretical_loss": 3.6550294626771405,
      "tokens_seen": 983367680
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003545737211634905,
      "loss": 2.6732,
      "theoretical_loss": 3.655006210278991,
      "tokens_seen": 983433216
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035456369107321966,
      "loss": 2.9324,
      "theoretical_loss": 3.654982959864168,
      "tokens_seen": 983498752
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035455366098294884,
      "loss": 2.8657,
      "theoretical_loss": 3.6549597114323706,
      "tokens_seen": 983564288
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.000354543630892678,
      "loss": 2.8017,
      "theoretical_loss": 3.6549364649832974,
      "tokens_seen": 983629824
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003545336008024072,
      "loss": 2.8468,
      "theoretical_loss": 3.6549132205166472,
      "tokens_seen": 983695360
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035452357071213644,
      "loss": 2.7388,
      "theoretical_loss": 3.6548899780321187,
      "tokens_seen": 983760896
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035451354062186556,
      "loss": 2.8988,
      "theoretical_loss": 3.654866737529411,
      "tokens_seen": 983826432
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003545035105315948,
      "loss": 2.8097,
      "theoretical_loss": 3.6548434990082233,
      "tokens_seen": 983891968
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.000354493480441324,
      "loss": 3.1389,
      "theoretical_loss": 3.654820262468254,
      "tokens_seen": 983957504
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035448345035105316,
      "loss": 2.9224,
      "theoretical_loss": 3.6547970279092032,
      "tokens_seen": 984023040
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035447342026078234,
      "loss": 3.0548,
      "theoretical_loss": 3.6547737953307697,
      "tokens_seen": 984088576
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003544633901705115,
      "loss": 2.9666,
      "theoretical_loss": 3.6547505647326526,
      "tokens_seen": 984154112
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003544533600802407,
      "loss": 2.8563,
      "theoretical_loss": 3.6547273361145507,
      "tokens_seen": 984219648
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035444332998996994,
      "loss": 2.8438,
      "theoretical_loss": 3.6547041094761648,
      "tokens_seen": 984285184
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035443329989969907,
      "loss": 2.9818,
      "theoretical_loss": 3.6546808848171937,
      "tokens_seen": 984350720
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003544232698094283,
      "loss": 2.976,
      "theoretical_loss": 3.654657662137337,
      "tokens_seen": 984416256
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035441323971915743,
      "loss": 2.9446,
      "theoretical_loss": 3.6546344414362935,
      "tokens_seen": 984481792
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035440320962888667,
      "loss": 2.6522,
      "theoretical_loss": 3.6546112227137644,
      "tokens_seen": 984547328
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035439317953861585,
      "loss": 2.8068,
      "theoretical_loss": 3.6545880059694484,
      "tokens_seen": 984612864
    },
    {
      "epoch": 0.3,
      "objective/train/advantage_avg": 0.49422624707221985,
      "objective/train/docs_used": 564602,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.0432255268096924,
      "objective/train/original_loss": 2.0432260036468506,
      "objective/train/theoretical_loss": 3.654564791203045,
      "objective/train/tokens_used": 1005138400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24431075155735016,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506645441055298,
      "objective/train/weighted_lm_loss": 2.1468849182128906,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0469889640808105,
      "theoretical_loss": 3.654564791203045,
      "tokens_seen": 984678400
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035438314944834503,
      "loss": 2.6506,
      "theoretical_loss": 3.654564791203045,
      "tokens_seen": 984678400
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003543731193580742,
      "loss": 2.7132,
      "theoretical_loss": 3.654541578414255,
      "tokens_seen": 984743936
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035436308926780344,
      "loss": 2.7249,
      "theoretical_loss": 3.654518367602778,
      "tokens_seen": 984809472
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003543530591775326,
      "loss": 2.9414,
      "theoretical_loss": 3.654495158768314,
      "tokens_seen": 984875008
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003543430290872618,
      "loss": 2.7041,
      "theoretical_loss": 3.654471951910563,
      "tokens_seen": 984940544
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.000354332998996991,
      "loss": 2.7464,
      "theoretical_loss": 3.6544487470292246,
      "tokens_seen": 985006080
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035432296890672017,
      "loss": 2.7071,
      "theoretical_loss": 3.6544255441240003,
      "tokens_seen": 985071616
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003543129388164494,
      "loss": 2.7146,
      "theoretical_loss": 3.6544023431945893,
      "tokens_seen": 985137152
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035430290872617853,
      "loss": 2.9128,
      "theoretical_loss": 3.654379144240692,
      "tokens_seen": 985202688
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035429287863590777,
      "loss": 2.8534,
      "theoretical_loss": 3.654355947262009,
      "tokens_seen": 985268224
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003542828485456369,
      "loss": 2.949,
      "theoretical_loss": 3.6543327522582416,
      "tokens_seen": 985333760
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035427281845536613,
      "loss": 3.1328,
      "theoretical_loss": 3.6543095592290884,
      "tokens_seen": 985399296
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003542627883650953,
      "loss": 3.1324,
      "theoretical_loss": 3.654286368174252,
      "tokens_seen": 985464832
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003542527582748245,
      "loss": 2.7894,
      "theoretical_loss": 3.654263179093432,
      "tokens_seen": 985530368
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035424272818455367,
      "loss": 2.8827,
      "theoretical_loss": 3.6542399919863287,
      "tokens_seen": 985595904
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035423269809428285,
      "loss": 2.6608,
      "theoretical_loss": 3.6542168068526433,
      "tokens_seen": 985661440
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035422266800401203,
      "loss": 2.9931,
      "theoretical_loss": 3.654193623692078,
      "tokens_seen": 985726976
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035421263791374127,
      "loss": 3.0622,
      "theoretical_loss": 3.6541704425043315,
      "tokens_seen": 985792512
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003542026078234704,
      "loss": 2.7202,
      "theoretical_loss": 3.654147263289106,
      "tokens_seen": 985858048
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035419257773319963,
      "loss": 2.816,
      "theoretical_loss": 3.654124086046102,
      "tokens_seen": 985923584
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003541825476429288,
      "loss": 2.8632,
      "theoretical_loss": 3.6541009107750217,
      "tokens_seen": 985989120
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.000354172517552658,
      "loss": 3.0435,
      "theoretical_loss": 3.654077737475565,
      "tokens_seen": 986054656
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003541624874623872,
      "loss": 2.8574,
      "theoretical_loss": 3.654054566147434,
      "tokens_seen": 986120192
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035415245737211636,
      "loss": 3.1404,
      "theoretical_loss": 3.65403139679033,
      "tokens_seen": 986185728
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035414242728184554,
      "loss": 2.894,
      "theoretical_loss": 3.6540082294039538,
      "tokens_seen": 986251264
    },
    {
      "epoch": 0.3,
      "objective/train/advantage_avg": 0.4937376081943512,
      "objective/train/docs_used": 565772,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.031085252761841,
      "objective/train/original_loss": 3.031085252761841,
      "objective/train/theoretical_loss": 3.6539850639880065,
      "objective/train/tokens_used": 1006776800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24383626878261566,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050613284111023,
      "objective/train/weighted_lm_loss": 3.1845598220825195,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0442860126495361,
      "theoretical_loss": 3.6539850639880065,
      "tokens_seen": 986316800
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035413239719157477,
      "loss": 3.1466,
      "theoretical_loss": 3.6539850639880065,
      "tokens_seen": 986316800
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003541223671013039,
      "loss": 2.7969,
      "theoretical_loss": 3.653961900542191,
      "tokens_seen": 986382336
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035411233701103314,
      "loss": 2.9641,
      "theoretical_loss": 3.6539387390662075,
      "tokens_seen": 986447872
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035410230692076226,
      "loss": 3.0826,
      "theoretical_loss": 3.6539155795597584,
      "tokens_seen": 986513408
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003540922768304915,
      "loss": 2.8022,
      "theoretical_loss": 3.6538924220225457,
      "tokens_seen": 986578944
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003540822467402207,
      "loss": 3.038,
      "theoretical_loss": 3.6538692664542696,
      "tokens_seen": 986644480
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035407221664994986,
      "loss": 2.9443,
      "theoretical_loss": 3.653846112854634,
      "tokens_seen": 986710016
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035406218655967904,
      "loss": 2.9415,
      "theoretical_loss": 3.653822961223339,
      "tokens_seen": 986775552
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003540521564694082,
      "loss": 2.8073,
      "theoretical_loss": 3.653799811560088,
      "tokens_seen": 986841088
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003540421263791374,
      "loss": 2.6901,
      "theoretical_loss": 3.6537766638645817,
      "tokens_seen": 986906624
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035403209628886664,
      "loss": 2.8943,
      "theoretical_loss": 3.653753518136523,
      "tokens_seen": 986972160
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035402206619859577,
      "loss": 3.0549,
      "theoretical_loss": 3.6537303743756144,
      "tokens_seen": 987037696
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.000354012036108325,
      "loss": 2.9072,
      "theoretical_loss": 3.6537072325815574,
      "tokens_seen": 987103232
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003540020060180542,
      "loss": 2.9935,
      "theoretical_loss": 3.6536840927540544,
      "tokens_seen": 987168768
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035399197592778336,
      "loss": 3.0165,
      "theoretical_loss": 3.653660954892808,
      "tokens_seen": 987234304
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035398194583751254,
      "loss": 2.8337,
      "theoretical_loss": 3.6536378189975203,
      "tokens_seen": 987299840
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003539719157472417,
      "loss": 3.0475,
      "theoretical_loss": 3.653614685067894,
      "tokens_seen": 987365376
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003539618856569709,
      "loss": 3.09,
      "theoretical_loss": 3.6535915531036314,
      "tokens_seen": 987430912
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035395185556670014,
      "loss": 3.121,
      "theoretical_loss": 3.6535684231044354,
      "tokens_seen": 987496448
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035394182547642927,
      "loss": 2.7419,
      "theoretical_loss": 3.6535452950700082,
      "tokens_seen": 987561984
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003539317953861585,
      "loss": 2.8533,
      "theoretical_loss": 3.6535221690000537,
      "tokens_seen": 987627520
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035392176529588763,
      "loss": 2.8141,
      "theoretical_loss": 3.6534990448942732,
      "tokens_seen": 987693056
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035391173520561687,
      "loss": 3.0012,
      "theoretical_loss": 3.6534759227523708,
      "tokens_seen": 987758592
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035390170511534605,
      "loss": 2.9436,
      "theoretical_loss": 3.6534528025740483,
      "tokens_seen": 987824128
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035389167502507523,
      "loss": 3.0629,
      "theoretical_loss": 3.6534296843590095,
      "tokens_seen": 987889664
    },
    {
      "epoch": 0.3,
      "objective/train/advantage_avg": 0.4952625632286072,
      "objective/train/docs_used": 566398,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.9973642826080322,
      "objective/train/original_loss": 2.9973645210266113,
      "objective/train/theoretical_loss": 3.6534065681069574,
      "objective/train/tokens_used": 1008415200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2453421801328659,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507733821868896,
      "objective/train/weighted_lm_loss": 3.14957594871521,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0363596677780151,
      "theoretical_loss": 3.6534065681069574,
      "tokens_seen": 987955200
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003538816449348044,
      "loss": 2.8407,
      "theoretical_loss": 3.6534065681069574,
      "tokens_seen": 987955200
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035387161484453364,
      "loss": 3.1155,
      "theoretical_loss": 3.6533834538175944,
      "tokens_seen": 988020736
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035386158475426277,
      "loss": 2.9089,
      "theoretical_loss": 3.653360341490625,
      "tokens_seen": 988086272
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.000353851554663992,
      "loss": 2.7076,
      "theoretical_loss": 3.653337231125751,
      "tokens_seen": 988151808
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035384152457372113,
      "loss": 2.9161,
      "theoretical_loss": 3.6533141227226764,
      "tokens_seen": 988217344
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035383149448345037,
      "loss": 3.0287,
      "theoretical_loss": 3.653291016281105,
      "tokens_seen": 988282880
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035382146439317955,
      "loss": 2.946,
      "theoretical_loss": 3.6532679118007403,
      "tokens_seen": 988348416
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035381143430290873,
      "loss": 2.9227,
      "theoretical_loss": 3.6532448092812846,
      "tokens_seen": 988413952
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003538014042126379,
      "loss": 2.9307,
      "theoretical_loss": 3.653221708722443,
      "tokens_seen": 988479488
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003537913741223671,
      "loss": 2.8917,
      "theoretical_loss": 3.6531986101239173,
      "tokens_seen": 988545024
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003537813440320963,
      "loss": 2.8149,
      "theoretical_loss": 3.6531755134854125,
      "tokens_seen": 988610560
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003537713139418255,
      "loss": 2.9743,
      "theoretical_loss": 3.6531524188066324,
      "tokens_seen": 988676096
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035376128385155464,
      "loss": 2.8482,
      "theoretical_loss": 3.6531293260872806,
      "tokens_seen": 988741632
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035375125376128387,
      "loss": 3.0189,
      "theoretical_loss": 3.653106235327061,
      "tokens_seen": 988807168
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.000353741223671013,
      "loss": 2.8875,
      "theoretical_loss": 3.6530831465256774,
      "tokens_seen": 988872704
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035373119358074223,
      "loss": 2.6283,
      "theoretical_loss": 3.653060059682834,
      "tokens_seen": 988938240
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003537211634904714,
      "loss": 2.6641,
      "theoretical_loss": 3.6530369747982347,
      "tokens_seen": 989003776
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003537111334002006,
      "loss": 3.0386,
      "theoretical_loss": 3.653013891871584,
      "tokens_seen": 989069312
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003537011033099298,
      "loss": 2.8514,
      "theoretical_loss": 3.652990810902586,
      "tokens_seen": 989134848
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.000353691073219659,
      "loss": 2.8803,
      "theoretical_loss": 3.6529677318909446,
      "tokens_seen": 989200384
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035368104312938814,
      "loss": 2.95,
      "theoretical_loss": 3.6529446548363644,
      "tokens_seen": 989265920
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003536710130391174,
      "loss": 2.9093,
      "theoretical_loss": 3.6529215797385506,
      "tokens_seen": 989331456
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003536609829488465,
      "loss": 2.7897,
      "theoretical_loss": 3.6528985065972064,
      "tokens_seen": 989396992
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035365095285857574,
      "loss": 3.011,
      "theoretical_loss": 3.6528754354120365,
      "tokens_seen": 989462528
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003536409227683049,
      "loss": 3.0244,
      "theoretical_loss": 3.6528523661827466,
      "tokens_seen": 989528064
    },
    {
      "epoch": 0.3,
      "objective/train/advantage_avg": 0.49562814831733704,
      "objective/train/docs_used": 567574,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.0106520652770996,
      "objective/train/original_loss": 3.0106515884399414,
      "objective/train/theoretical_loss": 3.65282929890904,
      "objective/train/tokens_used": 1010053600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2457033097743988,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0508118867874146,
      "objective/train/weighted_lm_loss": 3.163698196411133,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0363596677780151,
      "theoretical_loss": 3.65282929890904,
      "tokens_seen": 989593600
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003536308926780341,
      "loss": 3.0782,
      "theoretical_loss": 3.65282929890904,
      "tokens_seen": 989593600
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003536208625877633,
      "loss": 3.0268,
      "theoretical_loss": 3.6528062335906224,
      "tokens_seen": 989659136
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035361083249749246,
      "loss": 3.0166,
      "theoretical_loss": 3.652783170227198,
      "tokens_seen": 989724672
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003536008024072217,
      "loss": 2.9781,
      "theoretical_loss": 3.6527601088184722,
      "tokens_seen": 989790208
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003535907723169509,
      "loss": 3.0199,
      "theoretical_loss": 3.6527370493641493,
      "tokens_seen": 989855744
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035358074222668006,
      "loss": 2.7737,
      "theoretical_loss": 3.6527139918639353,
      "tokens_seen": 989921280
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035357071213640924,
      "loss": 2.9961,
      "theoretical_loss": 3.6526909363175344,
      "tokens_seen": 989986816
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003535606820461384,
      "loss": 2.9174,
      "theoretical_loss": 3.6526678827246513,
      "tokens_seen": 990052352
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003535506519558676,
      "loss": 3.0244,
      "theoretical_loss": 3.6526448310849924,
      "tokens_seen": 990117888
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035354062186559684,
      "loss": 3.2205,
      "theoretical_loss": 3.6526217813982624,
      "tokens_seen": 990183424
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035353059177532597,
      "loss": 2.8505,
      "theoretical_loss": 3.652598733664166,
      "tokens_seen": 990248960
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003535205616850552,
      "loss": 2.6293,
      "theoretical_loss": 3.6525756878824094,
      "tokens_seen": 990314496
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003535105315947844,
      "loss": 2.8625,
      "theoretical_loss": 3.652552644052698,
      "tokens_seen": 990380032
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035350050150451356,
      "loss": 2.8212,
      "theoretical_loss": 3.6525296021747367,
      "tokens_seen": 990445568
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035349047141424274,
      "loss": 3.1081,
      "theoretical_loss": 3.652506562248232,
      "tokens_seen": 990511104
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003534804413239719,
      "loss": 2.9499,
      "theoretical_loss": 3.6524835242728884,
      "tokens_seen": 990576640
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003534704112337011,
      "loss": 2.8971,
      "theoretical_loss": 3.652460488248412,
      "tokens_seen": 990642176
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035346038114343034,
      "loss": 2.8614,
      "theoretical_loss": 3.652437454174509,
      "tokens_seen": 990707712
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035345035105315947,
      "loss": 2.8945,
      "theoretical_loss": 3.6524144220508847,
      "tokens_seen": 990773248
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003534403209628887,
      "loss": 3.0767,
      "theoretical_loss": 3.6523913918772455,
      "tokens_seen": 990838784
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035343029087261783,
      "loss": 2.9449,
      "theoretical_loss": 3.652368363653297,
      "tokens_seen": 990904320
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035342026078234707,
      "loss": 2.7783,
      "theoretical_loss": 3.652345337378745,
      "tokens_seen": 990969856
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035341023069207625,
      "loss": 3.0659,
      "theoretical_loss": 3.652322313053296,
      "tokens_seen": 991035392
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035340020060180543,
      "loss": 3.1289,
      "theoretical_loss": 3.6522992906766563,
      "tokens_seen": 991100928
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003533901705115346,
      "loss": 2.8368,
      "theoretical_loss": 3.6522762702485316,
      "tokens_seen": 991166464
    },
    {
      "epoch": 0.3,
      "objective/train/advantage_avg": 0.4946490526199341,
      "objective/train/docs_used": 568180,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.4203238487243652,
      "objective/train/original_loss": 2.420323371887207,
      "objective/train/theoretical_loss": 3.652253251768628,
      "objective/train/tokens_used": 1011692000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24474984407424927,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507090091705322,
      "objective/train/weighted_lm_loss": 2.5429279804229736,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0408108234405518,
      "theoretical_loss": 3.652253251768628,
      "tokens_seen": 991232000
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035338014042126384,
      "loss": 2.7412,
      "theoretical_loss": 3.652253251768628,
      "tokens_seen": 991232000
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035337011033099297,
      "loss": 2.9847,
      "theoretical_loss": 3.6522302352366527,
      "tokens_seen": 991297536
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003533600802407222,
      "loss": 2.7827,
      "theoretical_loss": 3.6522072206523113,
      "tokens_seen": 991363072
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035335005015045133,
      "loss": 2.8879,
      "theoretical_loss": 3.6521842080153104,
      "tokens_seen": 991428608
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035334002006018057,
      "loss": 2.9852,
      "theoretical_loss": 3.652161197325357,
      "tokens_seen": 991494144
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035332998996990975,
      "loss": 3.0202,
      "theoretical_loss": 3.652138188582157,
      "tokens_seen": 991559680
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035331995987963893,
      "loss": 2.9099,
      "theoretical_loss": 3.652115181785417,
      "tokens_seen": 991625216
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003533099297893681,
      "loss": 2.8415,
      "theoretical_loss": 3.652092176934845,
      "tokens_seen": 991690752
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003532998996990973,
      "loss": 3.0539,
      "theoretical_loss": 3.6520691740301463,
      "tokens_seen": 991756288
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003532898696088265,
      "loss": 3.0421,
      "theoretical_loss": 3.6520461730710285,
      "tokens_seen": 991821824
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003532798395185557,
      "loss": 2.681,
      "theoretical_loss": 3.6520231740571982,
      "tokens_seen": 991887360
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035326980942828484,
      "loss": 3.0535,
      "theoretical_loss": 3.6520001769883628,
      "tokens_seen": 991952896
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035325977933801407,
      "loss": 2.6798,
      "theoretical_loss": 3.651977181864229,
      "tokens_seen": 992018432
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003532497492477432,
      "loss": 2.4702,
      "theoretical_loss": 3.6519541886845035,
      "tokens_seen": 992083968
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035323971915747243,
      "loss": 2.9112,
      "theoretical_loss": 3.651931197448894,
      "tokens_seen": 992149504
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003532296890672016,
      "loss": 2.6895,
      "theoretical_loss": 3.6519082081571073,
      "tokens_seen": 992215040
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003532196589769308,
      "loss": 2.789,
      "theoretical_loss": 3.6518852208088513,
      "tokens_seen": 992280576
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035320962888666,
      "loss": 2.8679,
      "theoretical_loss": 3.651862235403833,
      "tokens_seen": 992346112
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003531995987963892,
      "loss": 2.7321,
      "theoretical_loss": 3.6518392519417597,
      "tokens_seen": 992411648
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035318956870611834,
      "loss": 2.7513,
      "theoretical_loss": 3.6518162704223394,
      "tokens_seen": 992477184
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003531795386158476,
      "loss": 2.9177,
      "theoretical_loss": 3.6517932908452786,
      "tokens_seen": 992542720
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003531695085255767,
      "loss": 2.736,
      "theoretical_loss": 3.651770313210286,
      "tokens_seen": 992608256
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035315947843530594,
      "loss": 3.0626,
      "theoretical_loss": 3.651747337517069,
      "tokens_seen": 992673792
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003531494483450351,
      "loss": 2.7839,
      "theoretical_loss": 3.651724363765334,
      "tokens_seen": 992739328
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003531394182547643,
      "loss": 2.7021,
      "theoretical_loss": 3.6517013919547905,
      "tokens_seen": 992804864
    },
    {
      "epoch": 0.3,
      "objective/train/advantage_avg": 0.49571818113327026,
      "objective/train/docs_used": 569645,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.7368669509887695,
      "objective/train/original_loss": 2.7368662357330322,
      "objective/train/theoretical_loss": 3.651678422085146,
      "objective/train/tokens_used": 1013330400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2458011955022812,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050821304321289,
      "objective/train/weighted_lm_loss": 2.875945806503296,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0425469875335693,
      "theoretical_loss": 3.651678422085146,
      "tokens_seen": 992870400
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003531293881644935,
      "loss": 2.9283,
      "theoretical_loss": 3.651678422085146,
      "tokens_seen": 992870400
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035311935807422266,
      "loss": 3.116,
      "theoretical_loss": 3.651655454156108,
      "tokens_seen": 992935936
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035310932798395184,
      "loss": 2.9772,
      "theoretical_loss": 3.651632488167385,
      "tokens_seen": 993001472
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003530992978936811,
      "loss": 2.8877,
      "theoretical_loss": 3.6516095241186846,
      "tokens_seen": 993067008
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003530892678034102,
      "loss": 2.8961,
      "theoretical_loss": 3.6515865620097148,
      "tokens_seen": 993132544
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035307923771313944,
      "loss": 2.8715,
      "theoretical_loss": 3.6515636018401842,
      "tokens_seen": 993198080
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035306920762286857,
      "loss": 2.7533,
      "theoretical_loss": 3.6515406436098004,
      "tokens_seen": 993263616
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003530591775325978,
      "loss": 3.0676,
      "theoretical_loss": 3.6515176873182726,
      "tokens_seen": 993329152
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.000353049147442327,
      "loss": 2.8447,
      "theoretical_loss": 3.651494732965308,
      "tokens_seen": 993394688
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035303911735205617,
      "loss": 2.8981,
      "theoretical_loss": 3.651471780550616,
      "tokens_seen": 993460224
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035302908726178535,
      "loss": 3.03,
      "theoretical_loss": 3.651448830073905,
      "tokens_seen": 993525760
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003530190571715146,
      "loss": 2.7871,
      "theoretical_loss": 3.651425881534883,
      "tokens_seen": 993591296
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003530090270812437,
      "loss": 2.8398,
      "theoretical_loss": 3.6514029349332597,
      "tokens_seen": 993656832
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035299899699097294,
      "loss": 2.9846,
      "theoretical_loss": 3.651379990268742,
      "tokens_seen": 993722368
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035298896690070207,
      "loss": 2.8216,
      "theoretical_loss": 3.65135704754104,
      "tokens_seen": 993787904
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003529789368104313,
      "loss": 2.9335,
      "theoretical_loss": 3.651334106749862,
      "tokens_seen": 993853440
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003529689067201605,
      "loss": 2.9329,
      "theoretical_loss": 3.6513111678949173,
      "tokens_seen": 993918976
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035295887662988967,
      "loss": 2.9702,
      "theoretical_loss": 3.651288230975915,
      "tokens_seen": 993984512
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035294884653961885,
      "loss": 3.1801,
      "theoretical_loss": 3.651265295992563,
      "tokens_seen": 994050048
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035293881644934803,
      "loss": 2.8917,
      "theoretical_loss": 3.6512423629445707,
      "tokens_seen": 994115584
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003529287863590772,
      "loss": 3.0817,
      "theoretical_loss": 3.651219431831648,
      "tokens_seen": 994181120
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035291875626880645,
      "loss": 2.7691,
      "theoretical_loss": 3.6511965026535034,
      "tokens_seen": 994246656
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003529087261785356,
      "loss": 2.9561,
      "theoretical_loss": 3.651173575409846,
      "tokens_seen": 994312192
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003528986960882648,
      "loss": 3.1591,
      "theoretical_loss": 3.651150650100386,
      "tokens_seen": 994377728
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035288866599799394,
      "loss": 3.1899,
      "theoretical_loss": 3.6511277267248317,
      "tokens_seen": 994443264
    },
    {
      "epoch": 0.3,
      "objective/train/advantage_avg": 0.49454787373542786,
      "objective/train/docs_used": 570064,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.1332168579101562,
      "objective/train/original_loss": 3.1332168579101562,
      "objective/train/theoretical_loss": 3.651104805282893,
      "objective/train/tokens_used": 1014968800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24463964998722076,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506982803344727,
      "objective/train/weighted_lm_loss": 3.292024612426758,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0253151655197144,
      "theoretical_loss": 3.651104805282893,
      "tokens_seen": 994508800
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035287863590772317,
      "loss": 3.0068,
      "theoretical_loss": 3.651104805282893,
      "tokens_seen": 994508800
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035286860581745235,
      "loss": 2.9814,
      "theoretical_loss": 3.6510818857742797,
      "tokens_seen": 994574336
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035285857572718153,
      "loss": 2.8779,
      "theoretical_loss": 3.651058968198701,
      "tokens_seen": 994639872
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035284854563691077,
      "loss": 3.2789,
      "theoretical_loss": 3.6510360525558667,
      "tokens_seen": 994705408
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035283851554663995,
      "loss": 3.1115,
      "theoretical_loss": 3.651013138845486,
      "tokens_seen": 994770944
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035282848545636913,
      "loss": 3.0039,
      "theoretical_loss": 3.650990227067269,
      "tokens_seen": 994836480
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003528184553660983,
      "loss": 2.6752,
      "theoretical_loss": 3.6509673172209256,
      "tokens_seen": 994902016
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003528084252758275,
      "loss": 3.0672,
      "theoretical_loss": 3.650944409306166,
      "tokens_seen": 994967552
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003527983951855567,
      "loss": 2.8678,
      "theoretical_loss": 3.6509215033226994,
      "tokens_seen": 995033088
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003527883650952859,
      "loss": 2.9307,
      "theoretical_loss": 3.650898599270236,
      "tokens_seen": 995098624
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035277833500501504,
      "loss": 3.0576,
      "theoretical_loss": 3.6508756971484857,
      "tokens_seen": 995164160
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035276830491474427,
      "loss": 2.8461,
      "theoretical_loss": 3.6508527969571594,
      "tokens_seen": 995229696
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003527582748244734,
      "loss": 2.8817,
      "theoretical_loss": 3.6508298986959664,
      "tokens_seen": 995295232
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035274824473420263,
      "loss": 2.8727,
      "theoretical_loss": 3.6508070023646173,
      "tokens_seen": 995360768
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003527382146439318,
      "loss": 3.087,
      "theoretical_loss": 3.6507841079628225,
      "tokens_seen": 995426304
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.000352728184553661,
      "loss": 2.9498,
      "theoretical_loss": 3.6507612154902924,
      "tokens_seen": 995491840
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003527181544633902,
      "loss": 3.1291,
      "theoretical_loss": 3.650738324946737,
      "tokens_seen": 995557376
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003527081243731194,
      "loss": 2.7566,
      "theoretical_loss": 3.6507154363318675,
      "tokens_seen": 995622912
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035269809428284854,
      "loss": 2.8927,
      "theoretical_loss": 3.6506925496453935,
      "tokens_seen": 995688448
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003526880641925778,
      "loss": 2.7073,
      "theoretical_loss": 3.650669664887027,
      "tokens_seen": 995753984
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003526780341023069,
      "loss": 2.9022,
      "theoretical_loss": 3.650646782056477,
      "tokens_seen": 995819520
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035266800401203614,
      "loss": 2.7468,
      "theoretical_loss": 3.6506239011534554,
      "tokens_seen": 995885056
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003526579739217653,
      "loss": 2.9692,
      "theoretical_loss": 3.650601022177673,
      "tokens_seen": 995950592
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003526479438314945,
      "loss": 3.0134,
      "theoretical_loss": 3.65057814512884,
      "tokens_seen": 996016128
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003526379137412237,
      "loss": 3.0147,
      "theoretical_loss": 3.650555270006668,
      "tokens_seen": 996081664
    },
    {
      "epoch": 0.3,
      "objective/train/advantage_avg": 0.4940336048603058,
      "objective/train/docs_used": 571551,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.9804065227508545,
      "objective/train/original_loss": 2.9804062843322754,
      "objective/train/theoretical_loss": 3.6505323968108674,
      "objective/train/tokens_used": 1016607200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.244149312376976,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506445169448853,
      "objective/train/weighted_lm_loss": 3.1312379837036133,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0363596677780151,
      "theoretical_loss": 3.6505323968108674,
      "tokens_seen": 996147200
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035262788365095286,
      "loss": 3.0563,
      "theoretical_loss": 3.6505323968108674,
      "tokens_seen": 996147200
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035261785356068204,
      "loss": 3.1067,
      "theoretical_loss": 3.6505095255411497,
      "tokens_seen": 996212736
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003526078234704113,
      "loss": 2.9649,
      "theoretical_loss": 3.650486656197226,
      "tokens_seen": 996278272
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003525977933801404,
      "loss": 2.8872,
      "theoretical_loss": 3.650463788778807,
      "tokens_seen": 996343808
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035258776328986964,
      "loss": 3.0984,
      "theoretical_loss": 3.650440923285605,
      "tokens_seen": 996409344
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035257773319959877,
      "loss": 3.0094,
      "theoretical_loss": 3.65041805971733,
      "tokens_seen": 996474880
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.000352567703109328,
      "loss": 2.9494,
      "theoretical_loss": 3.650395198073694,
      "tokens_seen": 996540416
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003525576730190572,
      "loss": 2.8021,
      "theoretical_loss": 3.6503723383544084,
      "tokens_seen": 996605952
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035254764292878637,
      "loss": 2.8411,
      "theoretical_loss": 3.6503494805591847,
      "tokens_seen": 996671488
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035253761283851555,
      "loss": 2.7962,
      "theoretical_loss": 3.650326624687735,
      "tokens_seen": 996737024
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003525275827482448,
      "loss": 3.0115,
      "theoretical_loss": 3.65030377073977,
      "tokens_seen": 996802560
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003525175526579739,
      "loss": 3.1383,
      "theoretical_loss": 3.6502809187150014,
      "tokens_seen": 996868096
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035250752256770314,
      "loss": 2.8605,
      "theoretical_loss": 3.650258068613142,
      "tokens_seen": 996933632
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035249749247743227,
      "loss": 2.6864,
      "theoretical_loss": 3.650235220433902,
      "tokens_seen": 996999168
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003524874623871615,
      "loss": 2.953,
      "theoretical_loss": 3.650212374176995,
      "tokens_seen": 997064704
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003524774322968907,
      "loss": 3.0599,
      "theoretical_loss": 3.6501895298421316,
      "tokens_seen": 997130240
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035246740220661987,
      "loss": 3.0537,
      "theoretical_loss": 3.6501666874290244,
      "tokens_seen": 997195776
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035245737211634905,
      "loss": 2.7092,
      "theoretical_loss": 3.6501438469373855,
      "tokens_seen": 997261312
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035244734202607823,
      "loss": 2.6638,
      "theoretical_loss": 3.6501210083669267,
      "tokens_seen": 997326848
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003524373119358074,
      "loss": 3.1162,
      "theoretical_loss": 3.65009817171736,
      "tokens_seen": 997392384
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035242728184553665,
      "loss": 3.0266,
      "theoretical_loss": 3.6500753369883983,
      "tokens_seen": 997457920
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003524172517552658,
      "loss": 2.8024,
      "theoretical_loss": 3.6500525041797536,
      "tokens_seen": 997523456
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.000352407221664995,
      "loss": 3.111,
      "theoretical_loss": 3.6500296732911384,
      "tokens_seen": 997588992
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035239719157472414,
      "loss": 2.969,
      "theoretical_loss": 3.650006844322265,
      "tokens_seen": 997654528
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035238716148445337,
      "loss": 2.8896,
      "theoretical_loss": 3.6499840172728453,
      "tokens_seen": 997720064
    },
    {
      "epoch": 0.3,
      "objective/train/advantage_avg": 0.49412211775779724,
      "objective/train/docs_used": 572250,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.6048202514648438,
      "objective/train/original_loss": 2.6048202514648438,
      "objective/train/theoretical_loss": 3.6499611921425927,
      "objective/train/tokens_used": 1018245600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24421553313732147,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506535768508911,
      "objective/train/weighted_lm_loss": 2.736677408218384,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0437887907028198,
      "theoretical_loss": 3.6499611921425927,
      "tokens_seen": 997785600
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035237713139418255,
      "loss": 2.8516,
      "theoretical_loss": 3.6499611921425927,
      "tokens_seen": 997785600
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035236710130391173,
      "loss": 2.8402,
      "theoretical_loss": 3.6499383689312195,
      "tokens_seen": 997851136
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003523570712136409,
      "loss": 2.8818,
      "theoretical_loss": 3.6499155476384377,
      "tokens_seen": 997916672
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035234704112337015,
      "loss": 2.9739,
      "theoretical_loss": 3.6498927282639615,
      "tokens_seen": 997982208
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003523370110330993,
      "loss": 2.911,
      "theoretical_loss": 3.649869910807503,
      "tokens_seen": 998047744
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003523269809428285,
      "loss": 2.9643,
      "theoretical_loss": 3.6498470952687745,
      "tokens_seen": 998113280
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035231695085255764,
      "loss": 2.8987,
      "theoretical_loss": 3.649824281647489,
      "tokens_seen": 998178816
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003523069207622869,
      "loss": 2.697,
      "theoretical_loss": 3.6498014699433603,
      "tokens_seen": 998244352
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035229689067201606,
      "loss": 2.7107,
      "theoretical_loss": 3.649778660156101,
      "tokens_seen": 998309888
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035228686058174524,
      "loss": 3.0224,
      "theoretical_loss": 3.6497558522854243,
      "tokens_seen": 998375424
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003522768304914744,
      "loss": 2.8887,
      "theoretical_loss": 3.6497330463310433,
      "tokens_seen": 998440960
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003522668004012036,
      "loss": 2.8539,
      "theoretical_loss": 3.649710242292671,
      "tokens_seen": 998506496
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003522567703109328,
      "loss": 2.8649,
      "theoretical_loss": 3.6496874401700206,
      "tokens_seen": 998572032
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.000352246740220662,
      "loss": 2.9592,
      "theoretical_loss": 3.649664639962806,
      "tokens_seen": 998637568
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035223671013039114,
      "loss": 2.7288,
      "theoretical_loss": 3.64964184167074,
      "tokens_seen": 998703104
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003522266800401204,
      "loss": 3.0923,
      "theoretical_loss": 3.6496190452935373,
      "tokens_seen": 998768640
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035221664994984956,
      "loss": 3.1707,
      "theoretical_loss": 3.6495962508309097,
      "tokens_seen": 998834176
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035220661985957874,
      "loss": 2.9169,
      "theoretical_loss": 3.649573458282572,
      "tokens_seen": 998899712
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003521965897693079,
      "loss": 2.7944,
      "theoretical_loss": 3.6495506676482377,
      "tokens_seen": 998965248
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003521865596790371,
      "loss": 3.015,
      "theoretical_loss": 3.64952787892762,
      "tokens_seen": 999030784
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003521765295887663,
      "loss": 2.8621,
      "theoretical_loss": 3.649505092120433,
      "tokens_seen": 999096320
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003521664994984955,
      "loss": 2.9754,
      "theoretical_loss": 3.649482307226391,
      "tokens_seen": 999161856
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035215646940822465,
      "loss": 3.0127,
      "theoretical_loss": 3.6494595242452066,
      "tokens_seen": 999227392
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003521464393179539,
      "loss": 2.8216,
      "theoretical_loss": 3.6494367431765955,
      "tokens_seen": 999292928
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.000352136409227683,
      "loss": 2.8859,
      "theoretical_loss": 3.6494139640202707,
      "tokens_seen": 999358464
    },
    {
      "epoch": 0.3,
      "objective/train/advantage_avg": 0.4948074519634247,
      "objective/train/docs_used": 573370,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.5566787719726562,
      "objective/train/original_loss": 2.5566790103912354,
      "objective/train/theoretical_loss": 3.6493911867759463,
      "objective/train/tokens_used": 1019884000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24488966166973114,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507256984710693,
      "objective/train/weighted_lm_loss": 2.686249017715454,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0425469875335693,
      "theoretical_loss": 3.6493911867759463,
      "tokens_seen": 999424000
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035212637913741224,
      "loss": 2.9852,
      "theoretical_loss": 3.6493911867759463,
      "tokens_seen": 999424000
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003521163490471414,
      "loss": 2.9912,
      "theoretical_loss": 3.6493684114433362,
      "tokens_seen": 999489536
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003521063189568706,
      "loss": 2.8481,
      "theoretical_loss": 3.6493456380221554,
      "tokens_seen": 999555072
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035209628886659984,
      "loss": 2.94,
      "theoretical_loss": 3.649322866512118,
      "tokens_seen": 999620608
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035208625877632897,
      "loss": 2.7017,
      "theoretical_loss": 3.649300096912938,
      "tokens_seen": 999686144
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003520762286860582,
      "loss": 2.7462,
      "theoretical_loss": 3.64927732922433,
      "tokens_seen": 999751680
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003520661985957874,
      "loss": 2.7426,
      "theoretical_loss": 3.649254563446008,
      "tokens_seen": 999817216
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035205616850551657,
      "loss": 3.0646,
      "theoretical_loss": 3.649231799577687,
      "tokens_seen": 999882752
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035204613841524575,
      "loss": 2.7938,
      "theoretical_loss": 3.649209037619082,
      "tokens_seen": 999948288
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.000352036108324975,
      "loss": 2.9681,
      "theoretical_loss": 3.649186277569907,
      "tokens_seen": 1000013824
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003520260782347041,
      "loss": 3.0288,
      "theoretical_loss": 3.6491635194298766,
      "tokens_seen": 1000079360
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035201604814443334,
      "loss": 3.0738,
      "theoretical_loss": 3.6491407631987056,
      "tokens_seen": 1000144896
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035200601805416247,
      "loss": 2.9303,
      "theoretical_loss": 3.6491180088761093,
      "tokens_seen": 1000210432
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003519959879638917,
      "loss": 2.8708,
      "theoretical_loss": 3.649095256461802,
      "tokens_seen": 1000275968
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003519859578736209,
      "loss": 2.92,
      "theoretical_loss": 3.6490725059554996,
      "tokens_seen": 1000341504
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035197592778335007,
      "loss": 2.8737,
      "theoretical_loss": 3.6490497573569165,
      "tokens_seen": 1000407040
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035196589769307925,
      "loss": 2.8778,
      "theoretical_loss": 3.649027010665767,
      "tokens_seen": 1000472576
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035195586760280843,
      "loss": 2.9002,
      "theoretical_loss": 3.6490042658817674,
      "tokens_seen": 1000538112
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003519458375125376,
      "loss": 2.8562,
      "theoretical_loss": 3.6489815230046325,
      "tokens_seen": 1000603648
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035193580742226685,
      "loss": 2.899,
      "theoretical_loss": 3.6489587820340774,
      "tokens_seen": 1000669184
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.000351925777331996,
      "loss": 2.9546,
      "theoretical_loss": 3.648936042969818,
      "tokens_seen": 1000734720
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003519157472417252,
      "loss": 2.9279,
      "theoretical_loss": 3.6489133058115684,
      "tokens_seen": 1000800256
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035190571715145434,
      "loss": 2.9812,
      "theoretical_loss": 3.6488905705590455,
      "tokens_seen": 1000865792
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035189568706118357,
      "loss": 2.815,
      "theoretical_loss": 3.6488678372119643,
      "tokens_seen": 1000931328
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035188565697091275,
      "loss": 2.819,
      "theoretical_loss": 3.64884510577004,
      "tokens_seen": 1000996864
    },
    {
      "epoch": 0.3,
      "objective/train/advantage_avg": 0.4947241246700287,
      "objective/train/docs_used": 574030,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.8419854640960693,
      "objective/train/original_loss": 2.8419854640960693,
      "objective/train/theoretical_loss": 3.6488223762329888,
      "objective/train/tokens_used": 1021522400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24481908977031708,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507168769836426,
      "objective/train/weighted_lm_loss": 2.986077070236206,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0445106029510498,
      "theoretical_loss": 3.6488223762329888,
      "tokens_seen": 1001062400
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035187562688064193,
      "loss": 3.2145,
      "theoretical_loss": 3.6488223762329888,
      "tokens_seen": 1001062400
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003518655967903711,
      "loss": 2.9222,
      "theoretical_loss": 3.6487996486005256,
      "tokens_seen": 1001127936
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035185556670010035,
      "loss": 2.9165,
      "theoretical_loss": 3.6487769228723668,
      "tokens_seen": 1001193472
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003518455366098295,
      "loss": 3.01,
      "theoretical_loss": 3.648754199048228,
      "tokens_seen": 1001259008
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003518355065195587,
      "loss": 2.9801,
      "theoretical_loss": 3.6487314771278245,
      "tokens_seen": 1001324544
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035182547642928784,
      "loss": 3.1513,
      "theoretical_loss": 3.648708757110873,
      "tokens_seen": 1001390080
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003518154463390171,
      "loss": 2.5501,
      "theoretical_loss": 3.6486860389970897,
      "tokens_seen": 1001455616
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035180541624874626,
      "loss": 2.7606,
      "theoretical_loss": 3.6486633227861907,
      "tokens_seen": 1001521152
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035179538615847544,
      "loss": 2.941,
      "theoretical_loss": 3.6486406084778906,
      "tokens_seen": 1001586688
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003517853560682046,
      "loss": 2.8134,
      "theoretical_loss": 3.648617896071907,
      "tokens_seen": 1001652224
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003517753259779338,
      "loss": 2.7603,
      "theoretical_loss": 3.648595185567956,
      "tokens_seen": 1001717760
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.000351765295887663,
      "loss": 2.9316,
      "theoretical_loss": 3.648572476965753,
      "tokens_seen": 1001783296
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003517552657973922,
      "loss": 2.8577,
      "theoretical_loss": 3.6485497702650154,
      "tokens_seen": 1001848832
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035174523570712134,
      "loss": 2.7478,
      "theoretical_loss": 3.6485270654654594,
      "tokens_seen": 1001914368
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003517352056168506,
      "loss": 2.8039,
      "theoretical_loss": 3.6485043625668014,
      "tokens_seen": 1001979904
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035172517552657976,
      "loss": 2.9703,
      "theoretical_loss": 3.6484816615687574,
      "tokens_seen": 1002045440
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035171514543630894,
      "loss": 2.8385,
      "theoretical_loss": 3.6484589624710444,
      "tokens_seen": 1002110976
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003517051153460381,
      "loss": 3.0387,
      "theoretical_loss": 3.648436265273379,
      "tokens_seen": 1002176512
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003516950852557673,
      "loss": 3.1697,
      "theoretical_loss": 3.6484135699754785,
      "tokens_seen": 1002242048
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003516850551654965,
      "loss": 2.8859,
      "theoretical_loss": 3.648390876577059,
      "tokens_seen": 1002307584
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003516750250752257,
      "loss": 2.9701,
      "theoretical_loss": 3.648368185077837,
      "tokens_seen": 1002373120
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035166499498495485,
      "loss": 2.6748,
      "theoretical_loss": 3.6483454954775305,
      "tokens_seen": 1002438656
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003516549648946841,
      "loss": 2.7498,
      "theoretical_loss": 3.648322807775856,
      "tokens_seen": 1002504192
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003516449348044132,
      "loss": 3.0127,
      "theoretical_loss": 3.6483001219725297,
      "tokens_seen": 1002569728
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035163490471414244,
      "loss": 2.9435,
      "theoretical_loss": 3.64827743806727,
      "tokens_seen": 1002635264
    },
    {
      "epoch": 0.3,
      "objective/train/advantage_avg": 0.4947204291820526,
      "objective/train/docs_used": 574442,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.858386993408203,
      "objective/train/original_loss": 2.8583877086639404,
      "objective/train/theoretical_loss": 3.648254756059793,
      "objective/train/tokens_used": 1023160800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24480661749839783,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507164001464844,
      "objective/train/weighted_lm_loss": 3.003246545791626,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0417574644088745,
      "theoretical_loss": 3.648254756059793,
      "tokens_seen": 1002700800
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003516248746238716,
      "loss": 2.9667,
      "theoretical_loss": 3.648254756059793,
      "tokens_seen": 1002700800
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003516148445336008,
      "loss": 3.0566,
      "theoretical_loss": 3.648232075949817,
      "tokens_seen": 1002766336
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035160481444333,
      "loss": 2.9745,
      "theoretical_loss": 3.648209397737058,
      "tokens_seen": 1002831872
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035159478435305917,
      "loss": 2.8736,
      "theoretical_loss": 3.6481867214212342,
      "tokens_seen": 1002897408
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035158475426278835,
      "loss": 2.5101,
      "theoretical_loss": 3.648164047002062,
      "tokens_seen": 1002962944
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003515747241725176,
      "loss": 2.9448,
      "theoretical_loss": 3.648141374479261,
      "tokens_seen": 1003028480
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003515646940822467,
      "loss": 2.9549,
      "theoretical_loss": 3.648118703852546,
      "tokens_seen": 1003094016
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035155466399197595,
      "loss": 2.9621,
      "theoretical_loss": 3.648096035121637,
      "tokens_seen": 1003159552
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035154463390170513,
      "loss": 2.9177,
      "theoretical_loss": 3.6480733682862496,
      "tokens_seen": 1003225088
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003515346038114343,
      "loss": 2.8159,
      "theoretical_loss": 3.6480507033461023,
      "tokens_seen": 1003290624
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003515245737211635,
      "loss": 2.9466,
      "theoretical_loss": 3.648028040300913,
      "tokens_seen": 1003356160
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035151454363089267,
      "loss": 2.9997,
      "theoretical_loss": 3.6480053791504,
      "tokens_seen": 1003421696
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035150451354062185,
      "loss": 3.0205,
      "theoretical_loss": 3.64798271989428,
      "tokens_seen": 1003487232
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003514944834503511,
      "loss": 2.6089,
      "theoretical_loss": 3.6479600625322717,
      "tokens_seen": 1003552768
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003514844533600802,
      "loss": 2.8483,
      "theoretical_loss": 3.647937407064093,
      "tokens_seen": 1003618304
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035147442326980945,
      "loss": 2.823,
      "theoretical_loss": 3.6479147534894616,
      "tokens_seen": 1003683840
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003514643931795386,
      "loss": 3.0513,
      "theoretical_loss": 3.647892101808096,
      "tokens_seen": 1003749376
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003514543630892678,
      "loss": 2.6145,
      "theoretical_loss": 3.647869452019714,
      "tokens_seen": 1003814912
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.000351444332998997,
      "loss": 3.0609,
      "theoretical_loss": 3.647846804124035,
      "tokens_seen": 1003880448
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003514343029087262,
      "loss": 2.9603,
      "theoretical_loss": 3.6478241581207755,
      "tokens_seen": 1003945984
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035142427281845536,
      "loss": 2.9995,
      "theoretical_loss": 3.647801514009655,
      "tokens_seen": 1004011520
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035141424272818454,
      "loss": 2.9553,
      "theoretical_loss": 3.6477788717903916,
      "tokens_seen": 1004077056
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003514042126379137,
      "loss": 2.948,
      "theoretical_loss": 3.647756231462704,
      "tokens_seen": 1004142592
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035139418254764295,
      "loss": 3.0435,
      "theoretical_loss": 3.6477335930263104,
      "tokens_seen": 1004208128
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003513841524573721,
      "loss": 2.7478,
      "theoretical_loss": 3.6477109564809296,
      "tokens_seen": 1004273664
    },
    {
      "epoch": 0.3,
      "objective/train/advantage_avg": 0.49513304233551025,
      "objective/train/docs_used": 575683,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.896829843521118,
      "objective/train/original_loss": 2.896829605102539,
      "objective/train/theoretical_loss": 3.64768832182628,
      "objective/train/tokens_used": 1024799200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24522534012794495,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507599115371704,
      "objective/train/weighted_lm_loss": 3.0438146591186523,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0442860126495361,
      "theoretical_loss": 3.64768832182628,
      "tokens_seen": 1004339200
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003513741223671013,
      "loss": 2.8618,
      "theoretical_loss": 3.64768832182628,
      "tokens_seen": 1004339200
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003513640922768305,
      "loss": 2.7947,
      "theoretical_loss": 3.6476656890620807,
      "tokens_seen": 1004404736
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003513540621865597,
      "loss": 2.8218,
      "theoretical_loss": 3.6476430581880503,
      "tokens_seen": 1004470272
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003513440320962889,
      "loss": 2.9565,
      "theoretical_loss": 3.647620429203908,
      "tokens_seen": 1004535808
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035133400200601804,
      "loss": 3.106,
      "theoretical_loss": 3.6475978021093716,
      "tokens_seen": 1004601344
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003513239719157473,
      "loss": 2.8028,
      "theoretical_loss": 3.6475751769041613,
      "tokens_seen": 1004666880
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035131394182547646,
      "loss": 3.0699,
      "theoretical_loss": 3.647552553587995,
      "tokens_seen": 1004732416
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035130391173520564,
      "loss": 2.9816,
      "theoretical_loss": 3.6475299321605927,
      "tokens_seen": 1004797952
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003512938816449348,
      "loss": 3.2023,
      "theoretical_loss": 3.6475073126216735,
      "tokens_seen": 1004863488
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.000351283851554664,
      "loss": 2.9229,
      "theoretical_loss": 3.6474846949709563,
      "tokens_seen": 1004929024
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003512738214643932,
      "loss": 2.96,
      "theoretical_loss": 3.64746207920816,
      "tokens_seen": 1004994560
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003512637913741224,
      "loss": 2.8483,
      "theoretical_loss": 3.6474394653330044,
      "tokens_seen": 1005060096
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035125376128385154,
      "loss": 3.0592,
      "theoretical_loss": 3.647416853345209,
      "tokens_seen": 1005125632
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003512437311935808,
      "loss": 2.9479,
      "theoretical_loss": 3.647394243244493,
      "tokens_seen": 1005191168
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035123370110330996,
      "loss": 2.802,
      "theoretical_loss": 3.6473716350305754,
      "tokens_seen": 1005256704
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035122367101303914,
      "loss": 2.6616,
      "theoretical_loss": 3.6473490287031765,
      "tokens_seen": 1005322240
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003512136409227683,
      "loss": 3.0031,
      "theoretical_loss": 3.6473264242620154,
      "tokens_seen": 1005387776
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003512036108324975,
      "loss": 2.9384,
      "theoretical_loss": 3.6473038217068128,
      "tokens_seen": 1005453312
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003511935807422267,
      "loss": 2.9056,
      "theoretical_loss": 3.647281221037287,
      "tokens_seen": 1005518848
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003511835506519559,
      "loss": 2.9652,
      "theoretical_loss": 3.6472586222531587,
      "tokens_seen": 1005584384
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035117352056168505,
      "loss": 2.883,
      "theoretical_loss": 3.6472360253541476,
      "tokens_seen": 1005649920
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003511634904714143,
      "loss": 2.8292,
      "theoretical_loss": 3.6472134303399733,
      "tokens_seen": 1005715456
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003511534603811434,
      "loss": 3.0293,
      "theoretical_loss": 3.6471908372103563,
      "tokens_seen": 1005780992
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035114343029087264,
      "loss": 2.826,
      "theoretical_loss": 3.647168245965016,
      "tokens_seen": 1005846528
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003511334002006018,
      "loss": 2.9976,
      "theoretical_loss": 3.6471456566036737,
      "tokens_seen": 1005912064
    },
    {
      "epoch": 0.3,
      "objective/train/advantage_avg": 0.4933144152164459,
      "objective/train/docs_used": 576356,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.7411937713623047,
      "objective/train/original_loss": 2.741194009780884,
      "objective/train/theoretical_loss": 3.6471230691260477,
      "objective/train/tokens_used": 1026437600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2434435784816742,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505688190460205,
      "objective/train/weighted_lm_loss": 2.879876136779785,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.038975715637207,
      "theoretical_loss": 3.6471230691260477,
      "tokens_seen": 1005977600
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.000351123370110331,
      "loss": 2.7051,
      "theoretical_loss": 3.6471230691260477,
      "tokens_seen": 1005977600
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003511133400200602,
      "loss": 2.9,
      "theoretical_loss": 3.64710048353186,
      "tokens_seen": 1006043136
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035110330992978937,
      "loss": 2.7488,
      "theoretical_loss": 3.6470778998208297,
      "tokens_seen": 1006108672
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035109327983951855,
      "loss": 2.8838,
      "theoretical_loss": 3.6470553179926775,
      "tokens_seen": 1006174208
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003510832497492478,
      "loss": 2.8364,
      "theoretical_loss": 3.647032738047124,
      "tokens_seen": 1006239744
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.0003510732196589769,
      "loss": 2.9771,
      "theoretical_loss": 3.647010159983889,
      "tokens_seen": 1006305280
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035106318956870615,
      "loss": 2.7895,
      "theoretical_loss": 3.646987583802694,
      "tokens_seen": 1006370816
    },
    {
      "epoch": 0.3,
      "learning_rate": 0.00035105315947843533,
      "loss": 2.9222,
      "theoretical_loss": 3.646965009503259,
      "tokens_seen": 1006436352
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003510431293881645,
      "loss": 2.9572,
      "theoretical_loss": 3.6469424370853045,
      "tokens_seen": 1006501888
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003510330992978937,
      "loss": 2.9449,
      "theoretical_loss": 3.6469198665485516,
      "tokens_seen": 1006567424
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00035102306920762287,
      "loss": 2.9806,
      "theoretical_loss": 3.6468972978927208,
      "tokens_seen": 1006632960
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00035101303911735205,
      "loss": 2.7118,
      "theoretical_loss": 3.646874731117533,
      "tokens_seen": 1006698496
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003510030090270813,
      "loss": 3.0396,
      "theoretical_loss": 3.646852166222709,
      "tokens_seen": 1006764032
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003509929789368104,
      "loss": 3.0186,
      "theoretical_loss": 3.6468296032079697,
      "tokens_seen": 1006829568
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00035098294884653965,
      "loss": 2.9845,
      "theoretical_loss": 3.646807042073036,
      "tokens_seen": 1006895104
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003509729187562688,
      "loss": 2.7847,
      "theoretical_loss": 3.6467844828176297,
      "tokens_seen": 1006960640
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.000350962888665998,
      "loss": 2.7373,
      "theoretical_loss": 3.6467619254414707,
      "tokens_seen": 1007026176
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003509528585757272,
      "loss": 2.9238,
      "theoretical_loss": 3.6467393699442807,
      "tokens_seen": 1007091712
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003509428284854564,
      "loss": 2.9943,
      "theoretical_loss": 3.6467168163257817,
      "tokens_seen": 1007157248
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00035093279839518556,
      "loss": 3.0035,
      "theoretical_loss": 3.646694264585694,
      "tokens_seen": 1007222784
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00035092276830491474,
      "loss": 2.8542,
      "theoretical_loss": 3.646671714723739,
      "tokens_seen": 1007288320
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003509127382146439,
      "loss": 2.941,
      "theoretical_loss": 3.6466491667396386,
      "tokens_seen": 1007353856
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00035090270812437315,
      "loss": 2.5903,
      "theoretical_loss": 3.6466266206331137,
      "tokens_seen": 1007419392
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003508926780341023,
      "loss": 2.8738,
      "theoretical_loss": 3.6466040764038867,
      "tokens_seen": 1007484928
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003508826479438315,
      "loss": 3.0024,
      "theoretical_loss": 3.6465815340516783,
      "tokens_seen": 1007550464
    },
    {
      "epoch": 0.31,
      "objective/train/advantage_avg": 0.49136701226234436,
      "objective/train/docs_used": 577861,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.7426509857177734,
      "objective/train/original_loss": 2.7426507472991943,
      "objective/train/theoretical_loss": 3.6465589935762104,
      "objective/train/tokens_used": 1028076000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24156326055526733,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.05036461353302,
      "objective/train/weighted_lm_loss": 2.880661964416504,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0396549701690674,
      "theoretical_loss": 3.6465589935762104,
      "tokens_seen": 1007616000
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003508726178535607,
      "loss": 2.5968,
      "theoretical_loss": 3.6465589935762104,
      "tokens_seen": 1007616000
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003508625877632899,
      "loss": 3.0356,
      "theoretical_loss": 3.646536454977205,
      "tokens_seen": 1007681536
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00035085255767301906,
      "loss": 2.7868,
      "theoretical_loss": 3.6465139182543833,
      "tokens_seen": 1007747072
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00035084252758274824,
      "loss": 2.8358,
      "theoretical_loss": 3.6464913834074677,
      "tokens_seen": 1007812608
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003508324974924774,
      "loss": 3.0328,
      "theoretical_loss": 3.64646885043618,
      "tokens_seen": 1007878144
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00035082246740220666,
      "loss": 2.7774,
      "theoretical_loss": 3.646446319340242,
      "tokens_seen": 1007943680
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003508124373119358,
      "loss": 2.93,
      "theoretical_loss": 3.6464237901193757,
      "tokens_seen": 1008009216
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.000350802407221665,
      "loss": 3.2336,
      "theoretical_loss": 3.6464012627733027,
      "tokens_seen": 1008074752
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00035079237713139415,
      "loss": 3.0556,
      "theoretical_loss": 3.646378737301746,
      "tokens_seen": 1008140288
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003507823470411234,
      "loss": 3.0124,
      "theoretical_loss": 3.6463562137044274,
      "tokens_seen": 1008205824
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00035077231695085256,
      "loss": 3.0077,
      "theoretical_loss": 3.646333691981069,
      "tokens_seen": 1008271360
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00035076228686058174,
      "loss": 2.7028,
      "theoretical_loss": 3.6463111721313934,
      "tokens_seen": 1008336896
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003507522567703109,
      "loss": 3.0128,
      "theoretical_loss": 3.6462886541551223,
      "tokens_seen": 1008402432
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00035074222668004016,
      "loss": 2.9736,
      "theoretical_loss": 3.646266138051979,
      "tokens_seen": 1008467968
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003507321965897693,
      "loss": 2.9394,
      "theoretical_loss": 3.646243623821685,
      "tokens_seen": 1008533504
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003507221664994985,
      "loss": 2.9586,
      "theoretical_loss": 3.646221111463964,
      "tokens_seen": 1008599040
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00035071213640922765,
      "loss": 2.9207,
      "theoretical_loss": 3.646198600978538,
      "tokens_seen": 1008664576
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003507021063189569,
      "loss": 2.9647,
      "theoretical_loss": 3.6461760923651294,
      "tokens_seen": 1008730112
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00035069207622868607,
      "loss": 2.7372,
      "theoretical_loss": 3.646153585623461,
      "tokens_seen": 1008795648
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00035068204613841525,
      "loss": 2.9358,
      "theoretical_loss": 3.646131080753255,
      "tokens_seen": 1008861184
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00035067201604814443,
      "loss": 2.8308,
      "theoretical_loss": 3.6461085777542364,
      "tokens_seen": 1008926720
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003506619859578736,
      "loss": 2.7206,
      "theoretical_loss": 3.646086076626126,
      "tokens_seen": 1008992256
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003506519558676028,
      "loss": 2.8429,
      "theoretical_loss": 3.6460635773686474,
      "tokens_seen": 1009057792
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.000350641925777332,
      "loss": 2.9795,
      "theoretical_loss": 3.6460410799815235,
      "tokens_seen": 1009123328
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00035063189568706115,
      "loss": 3.1074,
      "theoretical_loss": 3.6460185844644775,
      "tokens_seen": 1009188864
    },
    {
      "epoch": 0.31,
      "objective/train/advantage_avg": 0.4940352737903595,
      "objective/train/docs_used": 578450,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.7926337718963623,
      "objective/train/original_loss": 2.7926337718963623,
      "objective/train/theoretical_loss": 3.645996090817232,
      "objective/train/tokens_used": 1029714400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24414482712745667,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506445169448853,
      "objective/train/weighted_lm_loss": 2.9339656829833984,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0363596677780151,
      "theoretical_loss": 3.645996090817232,
      "tokens_seen": 1009254400
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003506218655967904,
      "loss": 2.8908,
      "theoretical_loss": 3.645996090817232,
      "tokens_seen": 1009254400
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003506118355065195,
      "loss": 2.9511,
      "theoretical_loss": 3.645973599039511,
      "tokens_seen": 1009319936
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00035060180541624875,
      "loss": 2.8669,
      "theoretical_loss": 3.6459511091310377,
      "tokens_seen": 1009385472
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.000350591775325978,
      "loss": 2.9472,
      "theoretical_loss": 3.645928621091535,
      "tokens_seen": 1009451008
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003505817452357071,
      "loss": 2.851,
      "theoretical_loss": 3.6459061349207262,
      "tokens_seen": 1009516544
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00035057171514543635,
      "loss": 2.8106,
      "theoretical_loss": 3.6458836506183347,
      "tokens_seen": 1009582080
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00035056168505516553,
      "loss": 2.9522,
      "theoretical_loss": 3.6458611681840845,
      "tokens_seen": 1009647616
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003505516549648947,
      "loss": 2.7606,
      "theoretical_loss": 3.6458386876176987,
      "tokens_seen": 1009713152
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003505416248746239,
      "loss": 2.9062,
      "theoretical_loss": 3.645816208918901,
      "tokens_seen": 1009778688
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00035053159478435307,
      "loss": 2.9461,
      "theoretical_loss": 3.645793732087415,
      "tokens_seen": 1009844224
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00035052156469408225,
      "loss": 2.9416,
      "theoretical_loss": 3.6457712571229646,
      "tokens_seen": 1009909760
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003505115346038115,
      "loss": 2.6607,
      "theoretical_loss": 3.6457487840252734,
      "tokens_seen": 1009975296
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003505015045135406,
      "loss": 2.8096,
      "theoretical_loss": 3.645726312794065,
      "tokens_seen": 1010040832
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00035049147442326985,
      "loss": 2.9163,
      "theoretical_loss": 3.6457038434290636,
      "tokens_seen": 1010106368
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.000350481444332999,
      "loss": 2.7849,
      "theoretical_loss": 3.6456813759299935,
      "tokens_seen": 1010171904
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003504714142427282,
      "loss": 2.7809,
      "theoretical_loss": 3.6456589102965777,
      "tokens_seen": 1010237440
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003504613841524574,
      "loss": 2.6476,
      "theoretical_loss": 3.645636446528542,
      "tokens_seen": 1010302976
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003504513540621866,
      "loss": 2.9793,
      "theoretical_loss": 3.6456139846256077,
      "tokens_seen": 1010368512
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00035044132397191576,
      "loss": 2.9543,
      "theoretical_loss": 3.645591524587502,
      "tokens_seen": 1010434048
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00035043129388164494,
      "loss": 2.8995,
      "theoretical_loss": 3.6455690664139473,
      "tokens_seen": 1010499584
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003504212637913741,
      "loss": 2.743,
      "theoretical_loss": 3.6455466101046685,
      "tokens_seen": 1010565120
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00035041123370110335,
      "loss": 3.1383,
      "theoretical_loss": 3.6455241556593903,
      "tokens_seen": 1010630656
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003504012036108325,
      "loss": 2.8704,
      "theoretical_loss": 3.6455017030778363,
      "tokens_seen": 1010696192
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003503911735205617,
      "loss": 2.8576,
      "theoretical_loss": 3.6454792523597312,
      "tokens_seen": 1010761728
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003503811434302909,
      "loss": 2.6627,
      "theoretical_loss": 3.6454568035048003,
      "tokens_seen": 1010827264
    },
    {
      "epoch": 0.31,
      "objective/train/advantage_avg": 0.4939795732498169,
      "objective/train/docs_used": 579792,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.765200138092041,
      "objective/train/original_loss": 2.765200138092041,
      "objective/train/theoretical_loss": 3.645434356512767,
      "objective/train/tokens_used": 1031352800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2441144585609436,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506387948989868,
      "objective/train/weighted_lm_loss": 2.9050846099853516,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0396549701690674,
      "theoretical_loss": 3.645434356512767,
      "tokens_seen": 1010892800
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003503711133400201,
      "loss": 2.9229,
      "theoretical_loss": 3.645434356512767,
      "tokens_seen": 1010892800
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00035036108324974926,
      "loss": 2.9159,
      "theoretical_loss": 3.645411911383357,
      "tokens_seen": 1010958336
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00035035105315947844,
      "loss": 2.8521,
      "theoretical_loss": 3.6453894681162944,
      "tokens_seen": 1011023872
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003503410230692076,
      "loss": 2.8314,
      "theoretical_loss": 3.6453670267113045,
      "tokens_seen": 1011089408
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00035033099297893686,
      "loss": 2.9901,
      "theoretical_loss": 3.645344587168111,
      "tokens_seen": 1011154944
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.000350320962888666,
      "loss": 3.1512,
      "theoretical_loss": 3.6453221494864403,
      "tokens_seen": 1011220480
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003503109327983952,
      "loss": 2.9464,
      "theoretical_loss": 3.6452997136660166,
      "tokens_seen": 1011286016
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00035030090270812435,
      "loss": 2.6916,
      "theoretical_loss": 3.6452772797065642,
      "tokens_seen": 1011351552
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003502908726178536,
      "loss": 3.0025,
      "theoretical_loss": 3.6452548476078093,
      "tokens_seen": 1011417088
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00035028084252758276,
      "loss": 2.6235,
      "theoretical_loss": 3.645232417369477,
      "tokens_seen": 1011482624
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00035027081243731194,
      "loss": 2.7812,
      "theoretical_loss": 3.6452099889912914,
      "tokens_seen": 1011548160
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003502607823470411,
      "loss": 2.8214,
      "theoretical_loss": 3.645187562472979,
      "tokens_seen": 1011613696
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00035025075225677036,
      "loss": 2.9127,
      "theoretical_loss": 3.6451651378142644,
      "tokens_seen": 1011679232
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003502407221664995,
      "loss": 3.1076,
      "theoretical_loss": 3.645142715014873,
      "tokens_seen": 1011744768
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003502306920762287,
      "loss": 3.0762,
      "theoretical_loss": 3.6451202940745295,
      "tokens_seen": 1011810304
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00035022066198595785,
      "loss": 2.769,
      "theoretical_loss": 3.645097874992961,
      "tokens_seen": 1011875840
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003502106318956871,
      "loss": 2.9144,
      "theoretical_loss": 3.645075457769892,
      "tokens_seen": 1011941376
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00035020060180541627,
      "loss": 2.8049,
      "theoretical_loss": 3.6450530424050482,
      "tokens_seen": 1012006912
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00035019057171514545,
      "loss": 2.8963,
      "theoretical_loss": 3.645030628898155,
      "tokens_seen": 1012072448
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00035018054162487463,
      "loss": 2.8745,
      "theoretical_loss": 3.6450082172489395,
      "tokens_seen": 1012137984
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003501705115346038,
      "loss": 2.9205,
      "theoretical_loss": 3.6449858074571253,
      "tokens_seen": 1012203520
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.000350160481444333,
      "loss": 2.8692,
      "theoretical_loss": 3.6449633995224398,
      "tokens_seen": 1012269056
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003501504513540622,
      "loss": 2.7331,
      "theoretical_loss": 3.644940993444608,
      "tokens_seen": 1012334592
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00035014042126379135,
      "loss": 2.7714,
      "theoretical_loss": 3.6449185892233564,
      "tokens_seen": 1012400128
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003501303911735206,
      "loss": 2.9177,
      "theoretical_loss": 3.6448961868584107,
      "tokens_seen": 1012465664
    },
    {
      "epoch": 0.31,
      "objective/train/advantage_avg": 0.4932183623313904,
      "objective/train/docs_used": 580179,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.2976231575012207,
      "objective/train/original_loss": 2.2976229190826416,
      "objective/train/theoretical_loss": 3.644873786349497,
      "objective/train/tokens_used": 1032991200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24334488809108734,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505588054656982,
      "objective/train/weighted_lm_loss": 2.413477659225464,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0454468727111816,
      "theoretical_loss": 3.644873786349497,
      "tokens_seen": 1012531200
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003501203610832497,
      "loss": 2.9087,
      "theoretical_loss": 3.644873786349497,
      "tokens_seen": 1012531200
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00035011033099297895,
      "loss": 2.7186,
      "theoretical_loss": 3.6448513876963418,
      "tokens_seen": 1012596736
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00035010030090270813,
      "loss": 3.0746,
      "theoretical_loss": 3.644828990898671,
      "tokens_seen": 1012662272
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003500902708124373,
      "loss": 2.7719,
      "theoretical_loss": 3.6448065959562106,
      "tokens_seen": 1012727808
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003500802407221665,
      "loss": 2.6163,
      "theoretical_loss": 3.644784202868687,
      "tokens_seen": 1012793344
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00035007021063189573,
      "loss": 2.7899,
      "theoretical_loss": 3.644761811635827,
      "tokens_seen": 1012858880
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00035006018054162486,
      "loss": 2.7872,
      "theoretical_loss": 3.6447394222573557,
      "tokens_seen": 1012924416
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003500501504513541,
      "loss": 2.6812,
      "theoretical_loss": 3.6447170347330013,
      "tokens_seen": 1012989952
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003500401203610832,
      "loss": 2.9928,
      "theoretical_loss": 3.644694649062489,
      "tokens_seen": 1013055488
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00035003009027081245,
      "loss": 2.8272,
      "theoretical_loss": 3.6446722652455463,
      "tokens_seen": 1013121024
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00035002006018054163,
      "loss": 3.0945,
      "theoretical_loss": 3.644649883281899,
      "tokens_seen": 1013186560
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003500100300902708,
      "loss": 2.9858,
      "theoretical_loss": 3.6446275031712743,
      "tokens_seen": 1013252096
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00035,
      "loss": 2.8964,
      "theoretical_loss": 3.6446051249133986,
      "tokens_seen": 1013317632
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003499899699097292,
      "loss": 3.0696,
      "theoretical_loss": 3.6445827485079993,
      "tokens_seen": 1013383168
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034997993981945836,
      "loss": 3.0979,
      "theoretical_loss": 3.6445603739548025,
      "tokens_seen": 1013448704
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003499699097291876,
      "loss": 2.9796,
      "theoretical_loss": 3.6445380012535358,
      "tokens_seen": 1013514240
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003499598796389167,
      "loss": 2.9315,
      "theoretical_loss": 3.644515630403926,
      "tokens_seen": 1013579776
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034994984954864596,
      "loss": 2.7572,
      "theoretical_loss": 3.6444932614057,
      "tokens_seen": 1013645312
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003499398194583751,
      "loss": 2.7549,
      "theoretical_loss": 3.644470894258585,
      "tokens_seen": 1013710848
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003499297893681043,
      "loss": 2.849,
      "theoretical_loss": 3.644448528962308,
      "tokens_seen": 1013776384
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003499197592778335,
      "loss": 2.9144,
      "theoretical_loss": 3.644426165516596,
      "tokens_seen": 1013841920
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003499097291875627,
      "loss": 3.0029,
      "theoretical_loss": 3.6444038039211772,
      "tokens_seen": 1013907456
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034989969909729186,
      "loss": 2.7535,
      "theoretical_loss": 3.644381444175778,
      "tokens_seen": 1013972992
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003498896690070211,
      "loss": 2.9225,
      "theoretical_loss": 3.644359086280126,
      "tokens_seen": 1014038528
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003498796389167502,
      "loss": 2.7728,
      "theoretical_loss": 3.644336730233949,
      "tokens_seen": 1014104064
    },
    {
      "epoch": 0.31,
      "objective/train/advantage_avg": 0.4952164888381958,
      "objective/train/docs_used": 581599,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.7141573429107666,
      "objective/train/original_loss": 2.7141573429107666,
      "objective/train/theoretical_loss": 3.6443143760369736,
      "objective/train/tokens_used": 1034629600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24529395997524261,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507686138153076,
      "objective/train/weighted_lm_loss": 2.851802110671997,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0432155132293701,
      "theoretical_loss": 3.6443143760369736,
      "tokens_seen": 1014169600
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034986960882647946,
      "loss": 2.815,
      "theoretical_loss": 3.6443143760369736,
      "tokens_seen": 1014169600
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003498595787362086,
      "loss": 2.8296,
      "theoretical_loss": 3.644292023688929,
      "tokens_seen": 1014235136
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003498495486459378,
      "loss": 2.7819,
      "theoretical_loss": 3.644269673189541,
      "tokens_seen": 1014300672
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034983951855566706,
      "loss": 2.9663,
      "theoretical_loss": 3.6442473245385387,
      "tokens_seen": 1014366208
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003498294884653962,
      "loss": 2.8967,
      "theoretical_loss": 3.6442249777356492,
      "tokens_seen": 1014431744
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003498194583751254,
      "loss": 2.8593,
      "theoretical_loss": 3.6442026327806003,
      "tokens_seen": 1014497280
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034980942828485455,
      "loss": 2.7828,
      "theoretical_loss": 3.6441802896731197,
      "tokens_seen": 1014562816
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003497993981945838,
      "loss": 2.7622,
      "theoretical_loss": 3.644157948412935,
      "tokens_seen": 1014628352
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034978936810431296,
      "loss": 2.7058,
      "theoretical_loss": 3.6441356089997754,
      "tokens_seen": 1014693888
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034977933801404214,
      "loss": 2.8511,
      "theoretical_loss": 3.644113271433368,
      "tokens_seen": 1014759424
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003497693079237713,
      "loss": 2.6331,
      "theoretical_loss": 3.6440909357134412,
      "tokens_seen": 1014824960
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034975927783350056,
      "loss": 2.9465,
      "theoretical_loss": 3.6440686018397224,
      "tokens_seen": 1014890496
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003497492477432297,
      "loss": 2.8194,
      "theoretical_loss": 3.644046269811941,
      "tokens_seen": 1014956032
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003497392176529589,
      "loss": 3.0007,
      "theoretical_loss": 3.6440239396298244,
      "tokens_seen": 1015021568
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034972918756268805,
      "loss": 2.6793,
      "theoretical_loss": 3.644001611293101,
      "tokens_seen": 1015087104
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003497191574724173,
      "loss": 2.8492,
      "theoretical_loss": 3.6439792848014996,
      "tokens_seen": 1015152640
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034970912738214647,
      "loss": 2.8995,
      "theoretical_loss": 3.643956960154748,
      "tokens_seen": 1015218176
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034969909729187565,
      "loss": 2.9514,
      "theoretical_loss": 3.643934637352575,
      "tokens_seen": 1015283712
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034968906720160483,
      "loss": 3.181,
      "theoretical_loss": 3.643912316394709,
      "tokens_seen": 1015349248
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.000349679037111334,
      "loss": 3.0514,
      "theoretical_loss": 3.6438899972808794,
      "tokens_seen": 1015414784
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003496690070210632,
      "loss": 2.9003,
      "theoretical_loss": 3.6438676800108136,
      "tokens_seen": 1015480320
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003496589769307924,
      "loss": 2.8133,
      "theoretical_loss": 3.6438453645842412,
      "tokens_seen": 1015545856
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034964894684052155,
      "loss": 3.0791,
      "theoretical_loss": 3.643823051000891,
      "tokens_seen": 1015611392
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003496389167502508,
      "loss": 2.8412,
      "theoretical_loss": 3.6438007392604908,
      "tokens_seen": 1015676928
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003496288866599799,
      "loss": 2.8427,
      "theoretical_loss": 3.643778429362771,
      "tokens_seen": 1015742464
    },
    {
      "epoch": 0.31,
      "objective/train/advantage_avg": 0.4943927228450775,
      "objective/train/docs_used": 582330,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.7385354042053223,
      "objective/train/original_loss": 2.7385354042053223,
      "objective/train/theoretical_loss": 3.643756121307459,
      "objective/train/tokens_used": 1036268000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24449387192726135,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506821870803833,
      "objective/train/weighted_lm_loss": 2.8772785663604736,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.042169213294983,
      "theoretical_loss": 3.643756121307459,
      "tokens_seen": 1015808000
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034961885656970915,
      "loss": 2.6997,
      "theoretical_loss": 3.643756121307459,
      "tokens_seen": 1015808000
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034960882647943833,
      "loss": 2.6143,
      "theoretical_loss": 3.643733815094285,
      "tokens_seen": 1015873536
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003495987963891675,
      "loss": 2.7878,
      "theoretical_loss": 3.6437115107229774,
      "tokens_seen": 1015939072
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003495887662988967,
      "loss": 2.948,
      "theoretical_loss": 3.6436892081932655,
      "tokens_seen": 1016004608
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034957873620862593,
      "loss": 2.8181,
      "theoretical_loss": 3.643666907504879,
      "tokens_seen": 1016070144
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034956870611835506,
      "loss": 2.8831,
      "theoretical_loss": 3.643644608657546,
      "tokens_seen": 1016135680
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003495586760280843,
      "loss": 2.7751,
      "theoretical_loss": 3.6436223116509963,
      "tokens_seen": 1016201216
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003495486459378134,
      "loss": 2.9333,
      "theoretical_loss": 3.6436000164849593,
      "tokens_seen": 1016266752
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034953861584754265,
      "loss": 2.9935,
      "theoretical_loss": 3.643577723159165,
      "tokens_seen": 1016332288
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034952858575727183,
      "loss": 2.8472,
      "theoretical_loss": 3.643555431673342,
      "tokens_seen": 1016397824
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.000349518555667001,
      "loss": 2.7778,
      "theoretical_loss": 3.64353314202722,
      "tokens_seen": 1016463360
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003495085255767302,
      "loss": 2.9027,
      "theoretical_loss": 3.643510854220529,
      "tokens_seen": 1016528896
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003494984954864594,
      "loss": 2.8093,
      "theoretical_loss": 3.643488568252998,
      "tokens_seen": 1016594432
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034948846539618856,
      "loss": 2.882,
      "theoretical_loss": 3.643466284124357,
      "tokens_seen": 1016659968
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003494784353059178,
      "loss": 2.8366,
      "theoretical_loss": 3.643444001834336,
      "tokens_seen": 1016725504
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003494684052156469,
      "loss": 2.687,
      "theoretical_loss": 3.6434217213826647,
      "tokens_seen": 1016791040
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034945837512537616,
      "loss": 2.9574,
      "theoretical_loss": 3.643399442769072,
      "tokens_seen": 1016856576
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003494483450351053,
      "loss": 2.816,
      "theoretical_loss": 3.6433771659932894,
      "tokens_seen": 1016922112
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003494383149448345,
      "loss": 2.9738,
      "theoretical_loss": 3.6433548910550457,
      "tokens_seen": 1016987648
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003494282848545637,
      "loss": 2.8352,
      "theoretical_loss": 3.643332617954071,
      "tokens_seen": 1017053184
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003494182547642929,
      "loss": 2.8422,
      "theoretical_loss": 3.6433103466900962,
      "tokens_seen": 1017118720
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034940822467402206,
      "loss": 2.8391,
      "theoretical_loss": 3.643288077262851,
      "tokens_seen": 1017184256
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003493981945837513,
      "loss": 2.9936,
      "theoretical_loss": 3.6432658096720654,
      "tokens_seen": 1017249792
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003493881644934804,
      "loss": 3.0038,
      "theoretical_loss": 3.6432435439174693,
      "tokens_seen": 1017315328
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034937813440320966,
      "loss": 2.7994,
      "theoretical_loss": 3.6432212799987944,
      "tokens_seen": 1017380864
    },
    {
      "epoch": 0.31,
      "objective/train/advantage_avg": 0.49529311060905457,
      "objective/train/docs_used": 582938,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.2492406368255615,
      "objective/train/original_loss": 2.2492408752441406,
      "objective/train/theoretical_loss": 3.6431990179157694,
      "objective/train/tokens_used": 1037906400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24536600708961487,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507766008377075,
      "objective/train/weighted_lm_loss": 2.363377094268799,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0363596677780151,
      "theoretical_loss": 3.6431990179157694,
      "tokens_seen": 1017446400
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003493681043129388,
      "loss": 2.762,
      "theoretical_loss": 3.6431990179157694,
      "tokens_seen": 1017446400
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.000349358074222668,
      "loss": 2.7661,
      "theoretical_loss": 3.643176757668126,
      "tokens_seen": 1017511936
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003493480441323972,
      "loss": 2.8529,
      "theoretical_loss": 3.643154499255594,
      "tokens_seen": 1017577472
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003493380140421264,
      "loss": 2.898,
      "theoretical_loss": 3.643132242677904,
      "tokens_seen": 1017643008
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034932798395185557,
      "loss": 2.9358,
      "theoretical_loss": 3.643109987934787,
      "tokens_seen": 1017708544
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034931795386158475,
      "loss": 2.9043,
      "theoretical_loss": 3.643087735025973,
      "tokens_seen": 1017774080
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034930792377131393,
      "loss": 3.0782,
      "theoretical_loss": 3.6430654839511933,
      "tokens_seen": 1017839616
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034929789368104316,
      "loss": 2.9547,
      "theoretical_loss": 3.6430432347101784,
      "tokens_seen": 1017905152
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003492878635907723,
      "loss": 2.6912,
      "theoretical_loss": 3.6430209873026596,
      "tokens_seen": 1017970688
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003492778335005015,
      "loss": 2.9015,
      "theoretical_loss": 3.642998741728367,
      "tokens_seen": 1018036224
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034926780341023065,
      "loss": 3.0598,
      "theoretical_loss": 3.642976497987032,
      "tokens_seen": 1018101760
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003492577733199599,
      "loss": 2.7843,
      "theoretical_loss": 3.6429542560783856,
      "tokens_seen": 1018167296
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034924774322968907,
      "loss": 2.8543,
      "theoretical_loss": 3.6429320160021588,
      "tokens_seen": 1018232832
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034923771313941825,
      "loss": 2.5563,
      "theoretical_loss": 3.6429097777580823,
      "tokens_seen": 1018298368
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034922768304914743,
      "loss": 2.9295,
      "theoretical_loss": 3.6428875413458877,
      "tokens_seen": 1018363904
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034921765295887667,
      "loss": 2.7864,
      "theoretical_loss": 3.642865306765306,
      "tokens_seen": 1018429440
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003492076228686058,
      "loss": 2.5886,
      "theoretical_loss": 3.6428430740160693,
      "tokens_seen": 1018494976
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034919759277833503,
      "loss": 2.6936,
      "theoretical_loss": 3.642820843097908,
      "tokens_seen": 1018560512
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034918756268806416,
      "loss": 2.924,
      "theoretical_loss": 3.6427986140105535,
      "tokens_seen": 1018626048
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003491775325977934,
      "loss": 3.0681,
      "theoretical_loss": 3.642776386753738,
      "tokens_seen": 1018691584
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034916750250752257,
      "loss": 2.7134,
      "theoretical_loss": 3.642754161327192,
      "tokens_seen": 1018757120
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034915747241725175,
      "loss": 2.7911,
      "theoretical_loss": 3.642731937730648,
      "tokens_seen": 1018822656
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034914744232698093,
      "loss": 2.988,
      "theoretical_loss": 3.642709715963837,
      "tokens_seen": 1018888192
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003491374122367101,
      "loss": 2.5143,
      "theoretical_loss": 3.642687496026491,
      "tokens_seen": 1018953728
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003491273821464393,
      "loss": 2.9234,
      "theoretical_loss": 3.6426652779183417,
      "tokens_seen": 1019019264
    },
    {
      "epoch": 0.31,
      "objective/train/advantage_avg": 0.4934338629245758,
      "objective/train/docs_used": 584109,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.748129367828369,
      "objective/train/original_loss": 2.7481296062469482,
      "objective/train/theoretical_loss": 3.642643061639121,
      "objective/train/tokens_used": 1039544800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2435649037361145,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050581455230713,
      "objective/train/weighted_lm_loss": 2.886993646621704,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0396549701690674,
      "theoretical_loss": 3.642643061639121,
      "tokens_seen": 1019084800
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034911735205616853,
      "loss": 2.893,
      "theoretical_loss": 3.642643061639121,
      "tokens_seen": 1019084800
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034910732196589766,
      "loss": 2.8816,
      "theoretical_loss": 3.64262084718856,
      "tokens_seen": 1019150336
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003490972918756269,
      "loss": 2.7959,
      "theoretical_loss": 3.6425986345663914,
      "tokens_seen": 1019215872
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034908726178535613,
      "loss": 2.8978,
      "theoretical_loss": 3.642576423772347,
      "tokens_seen": 1019281408
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034907723169508526,
      "loss": 3.0549,
      "theoretical_loss": 3.642554214806159,
      "tokens_seen": 1019346944
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003490672016048145,
      "loss": 2.7617,
      "theoretical_loss": 3.6425320076675587,
      "tokens_seen": 1019412480
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003490571715145436,
      "loss": 2.8903,
      "theoretical_loss": 3.6425098023562787,
      "tokens_seen": 1019478016
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034904714142427285,
      "loss": 2.9162,
      "theoretical_loss": 3.6424875988720515,
      "tokens_seen": 1019543552
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034903711133400203,
      "loss": 2.8964,
      "theoretical_loss": 3.6424653972146093,
      "tokens_seen": 1019609088
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003490270812437312,
      "loss": 2.8754,
      "theoretical_loss": 3.6424431973836837,
      "tokens_seen": 1019674624
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003490170511534604,
      "loss": 2.7885,
      "theoretical_loss": 3.642420999379008,
      "tokens_seen": 1019740160
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003490070210631896,
      "loss": 2.8972,
      "theoretical_loss": 3.6423988032003143,
      "tokens_seen": 1019805696
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034899699097291876,
      "loss": 2.8756,
      "theoretical_loss": 3.642376608847334,
      "tokens_seen": 1019871232
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.000348986960882648,
      "loss": 2.733,
      "theoretical_loss": 3.642354416319801,
      "tokens_seen": 1019936768
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003489769307923771,
      "loss": 2.768,
      "theoretical_loss": 3.6423322256174475,
      "tokens_seen": 1020002304
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034896690070210636,
      "loss": 2.8918,
      "theoretical_loss": 3.642310036740006,
      "tokens_seen": 1020067840
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003489568706118355,
      "loss": 2.9423,
      "theoretical_loss": 3.642287849687209,
      "tokens_seen": 1020133376
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003489468405215647,
      "loss": 2.8858,
      "theoretical_loss": 3.6422656644587903,
      "tokens_seen": 1020198912
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003489368104312939,
      "loss": 2.8012,
      "theoretical_loss": 3.6422434810544813,
      "tokens_seen": 1020264448
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003489267803410231,
      "loss": 3.0301,
      "theoretical_loss": 3.6422212994740155,
      "tokens_seen": 1020329984
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034891675025075226,
      "loss": 2.9555,
      "theoretical_loss": 3.6421991197171257,
      "tokens_seen": 1020395520
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003489067201604815,
      "loss": 2.7199,
      "theoretical_loss": 3.6421769417835446,
      "tokens_seen": 1020461056
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003488966900702106,
      "loss": 2.4241,
      "theoretical_loss": 3.642154765673006,
      "tokens_seen": 1020526592
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034888665997993986,
      "loss": 2.8851,
      "theoretical_loss": 3.6421325913852427,
      "tokens_seen": 1020592128
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.000348876629889669,
      "loss": 2.8853,
      "theoretical_loss": 3.6421104189199873,
      "tokens_seen": 1020657664
    },
    {
      "epoch": 0.31,
      "objective/train/advantage_avg": 0.49584394693374634,
      "objective/train/docs_used": 584789,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.661966562271118,
      "objective/train/original_loss": 2.6619668006896973,
      "objective/train/theoretical_loss": 3.642088248276974,
      "objective/train/tokens_used": 1041183200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24590672552585602,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0508344173431396,
      "objective/train/weighted_lm_loss": 2.7970948219299316,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0442860126495361,
      "theoretical_loss": 3.642088248276974,
      "tokens_seen": 1020723200
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003488665997993982,
      "loss": 3.0205,
      "theoretical_loss": 3.642088248276974,
      "tokens_seen": 1020723200
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003488565697091274,
      "loss": 2.7617,
      "theoretical_loss": 3.642066079455935,
      "tokens_seen": 1020788736
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003488465396188566,
      "loss": 2.8041,
      "theoretical_loss": 3.642043912456604,
      "tokens_seen": 1020854272
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034883650952858577,
      "loss": 2.8988,
      "theoretical_loss": 3.642021747278715,
      "tokens_seen": 1020919808
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034882647943831495,
      "loss": 2.692,
      "theoretical_loss": 3.6419995839220007,
      "tokens_seen": 1020985344
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034881644934804413,
      "loss": 2.7142,
      "theoretical_loss": 3.6419774223861943,
      "tokens_seen": 1021050880
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034880641925777336,
      "loss": 2.7836,
      "theoretical_loss": 3.64195526267103,
      "tokens_seen": 1021116416
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003487963891675025,
      "loss": 3.2064,
      "theoretical_loss": 3.6419331047762418,
      "tokens_seen": 1021181952
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003487863590772317,
      "loss": 2.9943,
      "theoretical_loss": 3.641910948701562,
      "tokens_seen": 1021247488
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034877632898696085,
      "loss": 2.6565,
      "theoretical_loss": 3.641888794446725,
      "tokens_seen": 1021313024
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003487662988966901,
      "loss": 2.9537,
      "theoretical_loss": 3.6418666420114647,
      "tokens_seen": 1021378560
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034875626880641927,
      "loss": 2.6334,
      "theoretical_loss": 3.6418444913955152,
      "tokens_seen": 1021444096
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034874623871614845,
      "loss": 2.4848,
      "theoretical_loss": 3.6418223425986094,
      "tokens_seen": 1021509632
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034873620862587763,
      "loss": 2.8983,
      "theoretical_loss": 3.6418001956204824,
      "tokens_seen": 1021575168
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034872617853560687,
      "loss": 2.563,
      "theoretical_loss": 3.6417780504608674,
      "tokens_seen": 1021640704
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.000348716148445336,
      "loss": 2.9514,
      "theoretical_loss": 3.6417559071194985,
      "tokens_seen": 1021706240
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034870611835506523,
      "loss": 3.0421,
      "theoretical_loss": 3.64173376559611,
      "tokens_seen": 1021771776
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034869608826479436,
      "loss": 2.8665,
      "theoretical_loss": 3.6417116258904363,
      "tokens_seen": 1021837312
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003486860581745236,
      "loss": 2.6218,
      "theoretical_loss": 3.6416894880022106,
      "tokens_seen": 1021902848
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034867602808425277,
      "loss": 2.6798,
      "theoretical_loss": 3.6416673519311686,
      "tokens_seen": 1021968384
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034866599799398195,
      "loss": 2.8432,
      "theoretical_loss": 3.641645217677043,
      "tokens_seen": 1022033920
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034865596790371113,
      "loss": 2.8966,
      "theoretical_loss": 3.64162308523957,
      "tokens_seen": 1022099456
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003486459378134403,
      "loss": 2.796,
      "theoretical_loss": 3.6416009546184824,
      "tokens_seen": 1022164992
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003486359077231695,
      "loss": 2.8847,
      "theoretical_loss": 3.6415788258135158,
      "tokens_seen": 1022230528
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034862587763289873,
      "loss": 2.9818,
      "theoretical_loss": 3.641556698824404,
      "tokens_seen": 1022296064
    },
    {
      "epoch": 0.31,
      "objective/train/advantage_avg": 0.4917621612548828,
      "objective/train/docs_used": 586155,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.2256531715393066,
      "objective/train/original_loss": 2.2256531715393066,
      "objective/train/theoretical_loss": 3.6415345736508824,
      "objective/train/tokens_used": 1042821600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2419607788324356,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0504060983657837,
      "objective/train/weighted_lm_loss": 2.3384642601013184,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0338950157165527,
      "theoretical_loss": 3.6415345736508824,
      "tokens_seen": 1022361600
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034861584754262786,
      "loss": 2.81,
      "theoretical_loss": 3.6415345736508824,
      "tokens_seen": 1022361600
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003486058174523571,
      "loss": 2.9552,
      "theoretical_loss": 3.641512450292685,
      "tokens_seen": 1022427136
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003485957873620862,
      "loss": 2.9716,
      "theoretical_loss": 3.641490328749547,
      "tokens_seen": 1022492672
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034858575727181546,
      "loss": 2.6089,
      "theoretical_loss": 3.641468209021202,
      "tokens_seen": 1022558208
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034857572718154464,
      "loss": 2.8975,
      "theoretical_loss": 3.6414460911073863,
      "tokens_seen": 1022623744
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003485656970912738,
      "loss": 2.8531,
      "theoretical_loss": 3.641423975007834,
      "tokens_seen": 1022689280
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.000348555667001003,
      "loss": 2.8112,
      "theoretical_loss": 3.6414018607222802,
      "tokens_seen": 1022754816
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034854563691073223,
      "loss": 2.966,
      "theoretical_loss": 3.64137974825046,
      "tokens_seen": 1022820352
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034853560682046136,
      "loss": 2.7017,
      "theoretical_loss": 3.641357637592108,
      "tokens_seen": 1022885888
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003485255767301906,
      "loss": 2.8003,
      "theoretical_loss": 3.6413355287469598,
      "tokens_seen": 1022951424
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003485155466399197,
      "loss": 2.983,
      "theoretical_loss": 3.6413134217147505,
      "tokens_seen": 1023016960
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034850551654964896,
      "loss": 2.8967,
      "theoretical_loss": 3.6412913164952148,
      "tokens_seen": 1023082496
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034849548645937814,
      "loss": 3.0017,
      "theoretical_loss": 3.641269213088089,
      "tokens_seen": 1023148032
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003484854563691073,
      "loss": 2.9199,
      "theoretical_loss": 3.641247111493107,
      "tokens_seen": 1023213568
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003484754262788365,
      "loss": 3.0036,
      "theoretical_loss": 3.641225011710006,
      "tokens_seen": 1023279104
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003484653961885657,
      "loss": 2.7428,
      "theoretical_loss": 3.64120291373852,
      "tokens_seen": 1023344640
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034845536609829486,
      "loss": 2.6983,
      "theoretical_loss": 3.6411808175783844,
      "tokens_seen": 1023410176
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003484453360080241,
      "loss": 2.9622,
      "theoretical_loss": 3.641158723229336,
      "tokens_seen": 1023475712
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003484353059177532,
      "loss": 2.6968,
      "theoretical_loss": 3.641136630691109,
      "tokens_seen": 1023541248
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034842527582748246,
      "loss": 2.7483,
      "theoretical_loss": 3.64111453996344,
      "tokens_seen": 1023606784
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034841524573721164,
      "loss": 2.9651,
      "theoretical_loss": 3.641092451046064,
      "tokens_seen": 1023672320
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003484052156469408,
      "loss": 2.9147,
      "theoretical_loss": 3.6410703639387174,
      "tokens_seen": 1023737856
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034839518555667,
      "loss": 2.9415,
      "theoretical_loss": 3.6410482786411364,
      "tokens_seen": 1023803392
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003483851554663992,
      "loss": 2.9359,
      "theoretical_loss": 3.6410261951530556,
      "tokens_seen": 1023868928
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034837512537612837,
      "loss": 2.8167,
      "theoretical_loss": 3.641004113474212,
      "tokens_seen": 1023934464
    },
    {
      "epoch": 0.31,
      "objective/train/advantage_avg": 0.4943208396434784,
      "objective/train/docs_used": 586672,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.083000659942627,
      "objective/train/original_loss": 3.083000898361206,
      "objective/train/theoretical_loss": 3.6409820336043413,
      "objective/train/tokens_used": 1044460000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2444504201412201,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506747961044312,
      "objective/train/weighted_lm_loss": 3.239086389541626,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0338950157165527,
      "theoretical_loss": 3.6409820336043413,
      "tokens_seen": 1024000000
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003483650952858576,
      "loss": 2.7583,
      "theoretical_loss": 3.6409820336043413,
      "tokens_seen": 1024000000
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034835506519558673,
      "loss": 2.8665,
      "theoretical_loss": 3.6409599555431793,
      "tokens_seen": 1024065536
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034834503510531597,
      "loss": 2.7641,
      "theoretical_loss": 3.640937879290462,
      "tokens_seen": 1024131072
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034833500501504515,
      "loss": 3.016,
      "theoretical_loss": 3.640915804845926,
      "tokens_seen": 1024196608
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034832497492477433,
      "loss": 2.8534,
      "theoretical_loss": 3.640893732209307,
      "tokens_seen": 1024262144
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034831494483450356,
      "loss": 2.9531,
      "theoretical_loss": 3.6408716613803422,
      "tokens_seen": 1024327680
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003483049147442327,
      "loss": 2.7703,
      "theoretical_loss": 3.6408495923587667,
      "tokens_seen": 1024393216
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003482948846539619,
      "loss": 2.7566,
      "theoretical_loss": 3.640827525144318,
      "tokens_seen": 1024458752
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034828485456369105,
      "loss": 2.7445,
      "theoretical_loss": 3.640805459736732,
      "tokens_seen": 1024524288
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003482748244734203,
      "loss": 2.8313,
      "theoretical_loss": 3.640783396135745,
      "tokens_seen": 1024589824
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034826479438314947,
      "loss": 2.8448,
      "theoretical_loss": 3.640761334341094,
      "tokens_seen": 1024655360
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034825476429287865,
      "loss": 2.9793,
      "theoretical_loss": 3.640739274352515,
      "tokens_seen": 1024720896
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034824473420260783,
      "loss": 3.0653,
      "theoretical_loss": 3.640717216169745,
      "tokens_seen": 1024786432
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034823470411233707,
      "loss": 2.9325,
      "theoretical_loss": 3.640695159792521,
      "tokens_seen": 1024851968
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003482246740220662,
      "loss": 3.1179,
      "theoretical_loss": 3.6406731052205794,
      "tokens_seen": 1024917504
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034821464393179543,
      "loss": 2.6595,
      "theoretical_loss": 3.640651052453657,
      "tokens_seen": 1024983040
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034820461384152456,
      "loss": 2.7686,
      "theoretical_loss": 3.640629001491491,
      "tokens_seen": 1025048576
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003481945837512538,
      "loss": 2.9461,
      "theoretical_loss": 3.640606952333818,
      "tokens_seen": 1025114112
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034818455366098297,
      "loss": 2.8926,
      "theoretical_loss": 3.6405849049803747,
      "tokens_seen": 1025179648
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034817452357071215,
      "loss": 2.6287,
      "theoretical_loss": 3.640562859430899,
      "tokens_seen": 1025245184
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034816449348044133,
      "loss": 2.7095,
      "theoretical_loss": 3.6405408156851276,
      "tokens_seen": 1025310720
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003481544633901705,
      "loss": 2.7112,
      "theoretical_loss": 3.6405187737427975,
      "tokens_seen": 1025376256
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003481444332998997,
      "loss": 2.9555,
      "theoretical_loss": 3.640496733603646,
      "tokens_seen": 1025441792
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034813440320962893,
      "loss": 2.7558,
      "theoretical_loss": 3.64047469526741,
      "tokens_seen": 1025507328
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034812437311935806,
      "loss": 3.0004,
      "theoretical_loss": 3.6404526587338273,
      "tokens_seen": 1025572864
    },
    {
      "epoch": 0.31,
      "objective/train/advantage_avg": 0.49567121267318726,
      "objective/train/docs_used": 587795,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.8272171020507812,
      "objective/train/original_loss": 2.827216863632202,
      "objective/train/theoretical_loss": 3.6404306240026356,
      "objective/train/tokens_used": 1046098400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2457389384508133,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050816297531128,
      "objective/train/weighted_lm_loss": 2.970926523208618,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0437887907028198,
      "theoretical_loss": 3.6404306240026356,
      "tokens_seen": 1025638400
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003481143430290873,
      "loss": 2.7444,
      "theoretical_loss": 3.6404306240026356,
      "tokens_seen": 1025638400
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003481043129388164,
      "loss": 2.6638,
      "theoretical_loss": 3.6404085910735713,
      "tokens_seen": 1025703936
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034809428284854566,
      "loss": 2.9439,
      "theoretical_loss": 3.6403865599463727,
      "tokens_seen": 1025769472
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034808425275827484,
      "loss": 2.8474,
      "theoretical_loss": 3.640364530620777,
      "tokens_seen": 1025835008
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.000348074222668004,
      "loss": 2.9116,
      "theoretical_loss": 3.640342503096522,
      "tokens_seen": 1025900544
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003480641925777332,
      "loss": 2.7507,
      "theoretical_loss": 3.6403204773733453,
      "tokens_seen": 1025966080
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034805416248746243,
      "loss": 2.8277,
      "theoretical_loss": 3.640298453450984,
      "tokens_seen": 1026031616
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034804413239719156,
      "loss": 2.9407,
      "theoretical_loss": 3.640276431329177,
      "tokens_seen": 1026097152
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003480341023069208,
      "loss": 2.8075,
      "theoretical_loss": 3.640254411007661,
      "tokens_seen": 1026162688
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003480240722166499,
      "loss": 2.832,
      "theoretical_loss": 3.6402323924861744,
      "tokens_seen": 1026228224
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034801404212637916,
      "loss": 2.7685,
      "theoretical_loss": 3.6402103757644553,
      "tokens_seen": 1026293760
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034800401203610834,
      "loss": 2.7861,
      "theoretical_loss": 3.6401883608422416,
      "tokens_seen": 1026359296
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003479939819458375,
      "loss": 2.9684,
      "theoretical_loss": 3.6401663477192705,
      "tokens_seen": 1026424832
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003479839518555667,
      "loss": 2.6977,
      "theoretical_loss": 3.6401443363952812,
      "tokens_seen": 1026490368
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003479739217652959,
      "loss": 2.7925,
      "theoretical_loss": 3.640122326870012,
      "tokens_seen": 1026555904
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034796389167502506,
      "loss": 2.5882,
      "theoretical_loss": 3.6401003191431993,
      "tokens_seen": 1026621440
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003479538615847543,
      "loss": 2.5874,
      "theoretical_loss": 3.640078313214583,
      "tokens_seen": 1026686976
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034794383149448343,
      "loss": 2.7911,
      "theoretical_loss": 3.640056309083901,
      "tokens_seen": 1026752512
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034793380140421266,
      "loss": 2.7366,
      "theoretical_loss": 3.640034306750892,
      "tokens_seen": 1026818048
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034792377131394184,
      "loss": 3.0852,
      "theoretical_loss": 3.6400123062152936,
      "tokens_seen": 1026883584
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.000347913741223671,
      "loss": 2.7425,
      "theoretical_loss": 3.639990307476845,
      "tokens_seen": 1026949120
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003479037111334002,
      "loss": 2.9338,
      "theoretical_loss": 3.6399683105352842,
      "tokens_seen": 1027014656
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003478936810431294,
      "loss": 2.9147,
      "theoretical_loss": 3.6399463153903495,
      "tokens_seen": 1027080192
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034788365095285857,
      "loss": 2.812,
      "theoretical_loss": 3.6399243220417805,
      "tokens_seen": 1027145728
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003478736208625878,
      "loss": 2.9904,
      "theoretical_loss": 3.6399023304893157,
      "tokens_seen": 1027211264
    },
    {
      "epoch": 0.31,
      "objective/train/advantage_avg": 0.49628520011901855,
      "objective/train/docs_used": 588595,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.1171886920928955,
      "objective/train/original_loss": 3.1171882152557373,
      "objective/train/theoretical_loss": 3.6398803407326934,
      "objective/train/tokens_used": 1047736800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24633857607841492,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050880789756775,
      "objective/train/weighted_lm_loss": 3.2757461071014404,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0304546356201172,
      "theoretical_loss": 3.6398803407326934,
      "tokens_seen": 1027276800
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034786359077231693,
      "loss": 3.078,
      "theoretical_loss": 3.6398803407326934,
      "tokens_seen": 1027276800
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034785356068204617,
      "loss": 2.9352,
      "theoretical_loss": 3.639858352771652,
      "tokens_seen": 1027342336
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003478435305917753,
      "loss": 2.7486,
      "theoretical_loss": 3.639836366605932,
      "tokens_seen": 1027407872
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034783350050150453,
      "loss": 2.8643,
      "theoretical_loss": 3.6398143822352704,
      "tokens_seen": 1027473408
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003478234704112337,
      "loss": 2.9386,
      "theoretical_loss": 3.639792399659407,
      "tokens_seen": 1027538944
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003478134403209629,
      "loss": 2.9707,
      "theoretical_loss": 3.639770418878081,
      "tokens_seen": 1027604480
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034780341023069207,
      "loss": 2.8965,
      "theoretical_loss": 3.639748439891031,
      "tokens_seen": 1027670016
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034779338014042125,
      "loss": 2.7733,
      "theoretical_loss": 3.639726462697997,
      "tokens_seen": 1027735552
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034778335005015043,
      "loss": 2.7514,
      "theoretical_loss": 3.639704487298717,
      "tokens_seen": 1027801088
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034777331995987967,
      "loss": 2.7082,
      "theoretical_loss": 3.6396825136929314,
      "tokens_seen": 1027866624
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003477632898696088,
      "loss": 2.8142,
      "theoretical_loss": 3.639660541880378,
      "tokens_seen": 1027932160
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034775325977933803,
      "loss": 2.7806,
      "theoretical_loss": 3.6396385718607975,
      "tokens_seen": 1027997696
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003477432296890672,
      "loss": 2.4744,
      "theoretical_loss": 3.639616603633929,
      "tokens_seen": 1028063232
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003477331995987964,
      "loss": 2.7907,
      "theoretical_loss": 3.6395946371995116,
      "tokens_seen": 1028128768
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003477231695085256,
      "loss": 2.8112,
      "theoretical_loss": 3.639572672557285,
      "tokens_seen": 1028194304
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034771313941825476,
      "loss": 3.0008,
      "theoretical_loss": 3.6395507097069886,
      "tokens_seen": 1028259840
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034770310932798394,
      "loss": 2.9044,
      "theoretical_loss": 3.6395287486483623,
      "tokens_seen": 1028325376
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034769307923771317,
      "loss": 2.9089,
      "theoretical_loss": 3.6395067893811457,
      "tokens_seen": 1028390912
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003476830491474423,
      "loss": 2.8997,
      "theoretical_loss": 3.6394848319050785,
      "tokens_seen": 1028456448
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034767301905717153,
      "loss": 2.827,
      "theoretical_loss": 3.6394628762199,
      "tokens_seen": 1028521984
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034766298896690066,
      "loss": 2.9954,
      "theoretical_loss": 3.6394409223253508,
      "tokens_seen": 1028587520
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003476529588766299,
      "loss": 2.8044,
      "theoretical_loss": 3.6394189702211706,
      "tokens_seen": 1028653056
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003476429287863591,
      "loss": 2.8373,
      "theoretical_loss": 3.639397019907099,
      "tokens_seen": 1028718592
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034763289869608826,
      "loss": 2.6738,
      "theoretical_loss": 3.639375071382876,
      "tokens_seen": 1028784128
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034762286860581744,
      "loss": 3.0357,
      "theoretical_loss": 3.6393531246482422,
      "tokens_seen": 1028849664
    },
    {
      "epoch": 0.31,
      "objective/train/advantage_avg": 0.4939960539340973,
      "objective/train/docs_used": 589927,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.6870462894439697,
      "objective/train/original_loss": 2.6870460510253906,
      "objective/train/theoretical_loss": 3.6393311797029373,
      "objective/train/tokens_used": 1049375200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24409128725528717,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506404638290405,
      "objective/train/weighted_lm_loss": 2.8232169151306152,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0428944826126099,
      "theoretical_loss": 3.6393311797029373,
      "tokens_seen": 1028915200
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003476128385155466,
      "loss": 2.9397,
      "theoretical_loss": 3.6393311797029373,
      "tokens_seen": 1028915200
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003476028084252758,
      "loss": 2.6337,
      "theoretical_loss": 3.6393092365467012,
      "tokens_seen": 1028980736
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034759277833500504,
      "loss": 3.1184,
      "theoretical_loss": 3.639287295179275,
      "tokens_seen": 1029046272
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003475827482447342,
      "loss": 3.0469,
      "theoretical_loss": 3.639265355600398,
      "tokens_seen": 1029111808
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003475727181544634,
      "loss": 2.9164,
      "theoretical_loss": 3.639243417809811,
      "tokens_seen": 1029177344
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034756268806419263,
      "loss": 2.7489,
      "theoretical_loss": 3.639221481807254,
      "tokens_seen": 1029242880
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034755265797392176,
      "loss": 2.9474,
      "theoretical_loss": 3.639199547592468,
      "tokens_seen": 1029308416
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.000347542627883651,
      "loss": 2.8766,
      "theoretical_loss": 3.6391776151651936,
      "tokens_seen": 1029373952
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003475325977933801,
      "loss": 2.7277,
      "theoretical_loss": 3.6391556845251705,
      "tokens_seen": 1029439488
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034752256770310936,
      "loss": 2.8465,
      "theoretical_loss": 3.6391337556721393,
      "tokens_seen": 1029505024
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034751253761283854,
      "loss": 2.8531,
      "theoretical_loss": 3.6391118286058415,
      "tokens_seen": 1029570560
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003475025075225677,
      "loss": 2.9134,
      "theoretical_loss": 3.639089903326018,
      "tokens_seen": 1029636096
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003474924774322969,
      "loss": 2.9487,
      "theoretical_loss": 3.639067979832408,
      "tokens_seen": 1029701632
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003474824473420261,
      "loss": 2.7596,
      "theoretical_loss": 3.6390460581247535,
      "tokens_seen": 1029767168
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034747241725175527,
      "loss": 2.6055,
      "theoretical_loss": 3.639024138202795,
      "tokens_seen": 1029832704
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003474623871614845,
      "loss": 2.7231,
      "theoretical_loss": 3.639002220066274,
      "tokens_seen": 1029898240
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034745235707121363,
      "loss": 2.8379,
      "theoretical_loss": 3.6389803037149306,
      "tokens_seen": 1029963776
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034744232698094286,
      "loss": 2.6673,
      "theoretical_loss": 3.638958389148506,
      "tokens_seen": 1030029312
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034743229689067204,
      "loss": 2.8255,
      "theoretical_loss": 3.6389364763667422,
      "tokens_seen": 1030094848
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003474222668004012,
      "loss": 2.8261,
      "theoretical_loss": 3.6389145653693795,
      "tokens_seen": 1030160384
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003474122367101304,
      "loss": 2.8194,
      "theoretical_loss": 3.638892656156159,
      "tokens_seen": 1030225920
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003474022066198596,
      "loss": 2.9986,
      "theoretical_loss": 3.638870748726822,
      "tokens_seen": 1030291456
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034739217652958877,
      "loss": 2.8751,
      "theoretical_loss": 3.63884884308111,
      "tokens_seen": 1030356992
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.000347382146439318,
      "loss": 2.8131,
      "theoretical_loss": 3.6388269392187644,
      "tokens_seen": 1030422528
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034737211634904713,
      "loss": 2.7213,
      "theoretical_loss": 3.6388050371395266,
      "tokens_seen": 1030488064
    },
    {
      "epoch": 0.31,
      "objective/train/advantage_avg": 0.4945930540561676,
      "objective/train/docs_used": 590720,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.405606508255005,
      "objective/train/original_loss": 2.405606269836426,
      "objective/train/theoretical_loss": 3.638783136843138,
      "objective/train/tokens_used": 1051013600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24467112123966217,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507030487060547,
      "objective/train/weighted_lm_loss": 2.5276553630828857,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0396549701690674,
      "theoretical_loss": 3.638783136843138,
      "tokens_seen": 1030553600
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034736208625877637,
      "loss": 2.8548,
      "theoretical_loss": 3.638783136843138,
      "tokens_seen": 1030553600
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003473520561685055,
      "loss": 2.885,
      "theoretical_loss": 3.63876123832934,
      "tokens_seen": 1030619136
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034734202607823473,
      "loss": 2.7348,
      "theoretical_loss": 3.6387393415978746,
      "tokens_seen": 1030684672
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003473319959879639,
      "loss": 2.8825,
      "theoretical_loss": 3.6387174466484824,
      "tokens_seen": 1030750208
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003473219658976931,
      "loss": 2.7175,
      "theoretical_loss": 3.6386955534809062,
      "tokens_seen": 1030815744
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034731193580742227,
      "loss": 2.7038,
      "theoretical_loss": 3.638673662094887,
      "tokens_seen": 1030881280
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034730190571715145,
      "loss": 2.8923,
      "theoretical_loss": 3.6386517724901672,
      "tokens_seen": 1030946816
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034729187562688063,
      "loss": 2.66,
      "theoretical_loss": 3.6386298846664884,
      "tokens_seen": 1031012352
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034728184553660987,
      "loss": 2.7642,
      "theoretical_loss": 3.638607998623592,
      "tokens_seen": 1031077888
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.000347271815446339,
      "loss": 2.9181,
      "theoretical_loss": 3.63858611436122,
      "tokens_seen": 1031143424
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034726178535606823,
      "loss": 2.8691,
      "theoretical_loss": 3.6385642318791156,
      "tokens_seen": 1031208960
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003472517552657974,
      "loss": 2.8156,
      "theoretical_loss": 3.638542351177019,
      "tokens_seen": 1031274496
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003472417251755266,
      "loss": 2.7416,
      "theoretical_loss": 3.6385204722546742,
      "tokens_seen": 1031340032
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003472316950852558,
      "loss": 2.9559,
      "theoretical_loss": 3.638498595111822,
      "tokens_seen": 1031405568
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034722166499498496,
      "loss": 2.8646,
      "theoretical_loss": 3.638476719748205,
      "tokens_seen": 1031471104
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034721163490471414,
      "loss": 2.9578,
      "theoretical_loss": 3.6384548461635653,
      "tokens_seen": 1031536640
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034720160481444337,
      "loss": 2.8692,
      "theoretical_loss": 3.6384329743576456,
      "tokens_seen": 1031602176
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003471915747241725,
      "loss": 2.5042,
      "theoretical_loss": 3.638411104330188,
      "tokens_seen": 1031667712
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034718154463390173,
      "loss": 2.9381,
      "theoretical_loss": 3.6383892360809353,
      "tokens_seen": 1031733248
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034717151454363086,
      "loss": 3.0739,
      "theoretical_loss": 3.6383673696096297,
      "tokens_seen": 1031798784
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003471614844533601,
      "loss": 2.868,
      "theoretical_loss": 3.638345504916013,
      "tokens_seen": 1031864320
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003471514543630893,
      "loss": 2.9349,
      "theoretical_loss": 3.6383236419998295,
      "tokens_seen": 1031929856
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034714142427281846,
      "loss": 2.8738,
      "theoretical_loss": 3.6383017808608207,
      "tokens_seen": 1031995392
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034713139418254764,
      "loss": 3.0128,
      "theoretical_loss": 3.6382799214987287,
      "tokens_seen": 1032060928
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003471213640922768,
      "loss": 3.0387,
      "theoretical_loss": 3.6382580639132978,
      "tokens_seen": 1032126464
    },
    {
      "epoch": 0.31,
      "objective/train/advantage_avg": 0.4933317303657532,
      "objective/train/docs_used": 591991,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.5515425205230713,
      "objective/train/original_loss": 2.5515429973602295,
      "objective/train/theoretical_loss": 3.63823620810427,
      "objective/train/tokens_used": 1052652000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24346166849136353,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505707263946533,
      "objective/train/weighted_lm_loss": 2.680623769760132,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0338950157165527,
      "theoretical_loss": 3.63823620810427,
      "tokens_seen": 1032192000
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.000347111334002006,
      "loss": 2.6366,
      "theoretical_loss": 3.63823620810427,
      "tokens_seen": 1032192000
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034710130391173524,
      "loss": 2.6126,
      "theoretical_loss": 3.6382143540713874,
      "tokens_seen": 1032257536
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034709127382146436,
      "loss": 2.7625,
      "theoretical_loss": 3.6381925018143946,
      "tokens_seen": 1032323072
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003470812437311936,
      "loss": 2.8786,
      "theoretical_loss": 3.638170651333033,
      "tokens_seen": 1032388608
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003470712136409228,
      "loss": 3.0553,
      "theoretical_loss": 3.638148802627046,
      "tokens_seen": 1032454144
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034706118355065196,
      "loss": 2.8655,
      "theoretical_loss": 3.6381269556961775,
      "tokens_seen": 1032519680
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034705115346038114,
      "loss": 2.5937,
      "theoretical_loss": 3.63810511054017,
      "tokens_seen": 1032585216
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003470411233701103,
      "loss": 2.7202,
      "theoretical_loss": 3.638083267158766,
      "tokens_seen": 1032650752
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003470310932798395,
      "loss": 3.0018,
      "theoretical_loss": 3.6380614255517107,
      "tokens_seen": 1032716288
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034702106318956874,
      "loss": 2.892,
      "theoretical_loss": 3.638039585718745,
      "tokens_seen": 1032781824
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034701103309929787,
      "loss": 2.9785,
      "theoretical_loss": 3.638017747659614,
      "tokens_seen": 1032847360
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003470010030090271,
      "loss": 2.8469,
      "theoretical_loss": 3.6379959113740603,
      "tokens_seen": 1032912896
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034699097291875623,
      "loss": 2.5769,
      "theoretical_loss": 3.6379740768618274,
      "tokens_seen": 1032978432
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034698094282848547,
      "loss": 2.7803,
      "theoretical_loss": 3.637952244122659,
      "tokens_seen": 1033043968
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034697091273821465,
      "loss": 2.8939,
      "theoretical_loss": 3.6379304131562984,
      "tokens_seen": 1033109504
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034696088264794383,
      "loss": 2.9618,
      "theoretical_loss": 3.6379085839624894,
      "tokens_seen": 1033175040
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.000346950852557673,
      "loss": 2.8528,
      "theoretical_loss": 3.6378867565409756,
      "tokens_seen": 1033240576
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034694082246740224,
      "loss": 3.0076,
      "theoretical_loss": 3.6378649308915008,
      "tokens_seen": 1033306112
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034693079237713137,
      "loss": 2.8723,
      "theoretical_loss": 3.6378431070138086,
      "tokens_seen": 1033371648
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003469207622868606,
      "loss": 2.7559,
      "theoretical_loss": 3.637821284907643,
      "tokens_seen": 1033437184
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034691073219658973,
      "loss": 2.8722,
      "theoretical_loss": 3.6377994645727476,
      "tokens_seen": 1033502720
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034690070210631897,
      "loss": 2.9158,
      "theoretical_loss": 3.6377776460088667,
      "tokens_seen": 1033568256
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034689067201604815,
      "loss": 3.1778,
      "theoretical_loss": 3.6377558292157435,
      "tokens_seen": 1033633792
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034688064192577733,
      "loss": 2.8091,
      "theoretical_loss": 3.637734014193123,
      "tokens_seen": 1033699328
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003468706118355065,
      "loss": 3.0963,
      "theoretical_loss": 3.637712200940749,
      "tokens_seen": 1033764864
    },
    {
      "epoch": 0.31,
      "objective/train/advantage_avg": 0.49583420157432556,
      "objective/train/docs_used": 592462,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.792963743209839,
      "objective/train/original_loss": 2.7929635047912598,
      "objective/train/theoretical_loss": 3.637690389458365,
      "objective/train/tokens_used": 1054290400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24593158066272736,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0508335828781128,
      "objective/train/weighted_lm_loss": 2.9348974227905273,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0253151655197144,
      "theoretical_loss": 3.637690389458365,
      "tokens_seen": 1033830400
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003468605817452357,
      "loss": 2.622,
      "theoretical_loss": 3.637690389458365,
      "tokens_seen": 1033830400
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003468505516549649,
      "loss": 2.7465,
      "theoretical_loss": 3.637668579745716,
      "tokens_seen": 1033895936
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003468405215646941,
      "loss": 3.1677,
      "theoretical_loss": 3.637646771802546,
      "tokens_seen": 1033961472
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003468304914744233,
      "loss": 2.9405,
      "theoretical_loss": 3.6376249656285995,
      "tokens_seen": 1034027008
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034682046138415247,
      "loss": 2.7318,
      "theoretical_loss": 3.63760316122362,
      "tokens_seen": 1034092544
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034681043129388165,
      "loss": 3.0284,
      "theoretical_loss": 3.637581358587353,
      "tokens_seen": 1034158080
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034680040120361083,
      "loss": 2.8993,
      "theoretical_loss": 3.637559557719542,
      "tokens_seen": 1034223616
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034679037111334007,
      "loss": 3.0291,
      "theoretical_loss": 3.6375377586199322,
      "tokens_seen": 1034289152
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003467803410230692,
      "loss": 2.963,
      "theoretical_loss": 3.637515961288268,
      "tokens_seen": 1034354688
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034677031093279843,
      "loss": 2.7408,
      "theoretical_loss": 3.637494165724294,
      "tokens_seen": 1034420224
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003467602808425276,
      "loss": 2.5449,
      "theoretical_loss": 3.637472371927754,
      "tokens_seen": 1034485760
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003467502507522568,
      "loss": 2.9252,
      "theoretical_loss": 3.6374505798983945,
      "tokens_seen": 1034551296
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.000346740220661986,
      "loss": 2.8316,
      "theoretical_loss": 3.637428789635959,
      "tokens_seen": 1034616832
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034673019057171516,
      "loss": 2.7265,
      "theoretical_loss": 3.6374070011401924,
      "tokens_seen": 1034682368
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034672016048144434,
      "loss": 2.5914,
      "theoretical_loss": 3.6373852144108403,
      "tokens_seen": 1034747904
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034671013039117357,
      "loss": 2.9154,
      "theoretical_loss": 3.6373634294476465,
      "tokens_seen": 1034813440
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003467001003009027,
      "loss": 2.6473,
      "theoretical_loss": 3.6373416462503565,
      "tokens_seen": 1034878976
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034669007021063193,
      "loss": 3.1287,
      "theoretical_loss": 3.637319864818716,
      "tokens_seen": 1034944512
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034668004012036106,
      "loss": 2.8293,
      "theoretical_loss": 3.6372980851524694,
      "tokens_seen": 1035010048
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003466700100300903,
      "loss": 2.8988,
      "theoretical_loss": 3.637276307251362,
      "tokens_seen": 1035075584
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003466599799398195,
      "loss": 3.0941,
      "theoretical_loss": 3.637254531115139,
      "tokens_seen": 1035141120
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034664994984954866,
      "loss": 2.9222,
      "theoretical_loss": 3.637232756743545,
      "tokens_seen": 1035206656
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034663991975927784,
      "loss": 3.1393,
      "theoretical_loss": 3.6372109841363267,
      "tokens_seen": 1035272192
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.000346629889669007,
      "loss": 2.9846,
      "theoretical_loss": 3.6371892132932286,
      "tokens_seen": 1035337728
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003466198595787362,
      "loss": 2.8644,
      "theoretical_loss": 3.637167444213996,
      "tokens_seen": 1035403264
    },
    {
      "epoch": 0.31,
      "objective/train/advantage_avg": 0.4937974810600281,
      "objective/train/docs_used": 593828,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.0478973388671875,
      "objective/train/original_loss": 3.0478971004486084,
      "objective/train/theoretical_loss": 3.637145676898374,
      "objective/train/tokens_used": 1055928800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24389833211898804,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506196022033691,
      "objective/train/weighted_lm_loss": 3.202120780944824,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0417574644088745,
      "theoretical_loss": 3.637145676898374,
      "tokens_seen": 1035468800
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034660982948846544,
      "loss": 2.915,
      "theoretical_loss": 3.637145676898374,
      "tokens_seen": 1035468800
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034659979939819456,
      "loss": 3.0805,
      "theoretical_loss": 3.6371239113461096,
      "tokens_seen": 1035534336
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003465897693079238,
      "loss": 3.0933,
      "theoretical_loss": 3.637102147556947,
      "tokens_seen": 1035599872
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.000346579739217653,
      "loss": 2.977,
      "theoretical_loss": 3.6370803855306324,
      "tokens_seen": 1035665408
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034656970912738216,
      "loss": 3.0133,
      "theoretical_loss": 3.6370586252669117,
      "tokens_seen": 1035730944
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034655967903711134,
      "loss": 2.9144,
      "theoretical_loss": 3.6370368667655297,
      "tokens_seen": 1035796480
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003465496489468405,
      "loss": 2.6995,
      "theoretical_loss": 3.637015110026233,
      "tokens_seen": 1035862016
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003465396188565697,
      "loss": 2.9039,
      "theoretical_loss": 3.6369933550487668,
      "tokens_seen": 1035927552
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034652958876629894,
      "loss": 2.6757,
      "theoretical_loss": 3.6369716018328777,
      "tokens_seen": 1035993088
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034651955867602807,
      "loss": 2.7618,
      "theoretical_loss": 3.6369498503783113,
      "tokens_seen": 1036058624
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003465095285857573,
      "loss": 3.0009,
      "theoretical_loss": 3.6369281006848135,
      "tokens_seen": 1036124160
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034649949849548643,
      "loss": 2.9387,
      "theoretical_loss": 3.6369063527521304,
      "tokens_seen": 1036189696
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034648946840521567,
      "loss": 3.0204,
      "theoretical_loss": 3.6368846065800082,
      "tokens_seen": 1036255232
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034647943831494485,
      "loss": 2.8966,
      "theoretical_loss": 3.6368628621681927,
      "tokens_seen": 1036320768
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034646940822467403,
      "loss": 2.8882,
      "theoretical_loss": 3.636841119516431,
      "tokens_seen": 1036386304
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003464593781344032,
      "loss": 2.8554,
      "theoretical_loss": 3.636819378624468,
      "tokens_seen": 1036451840
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034644934804413244,
      "loss": 2.9782,
      "theoretical_loss": 3.636797639492051,
      "tokens_seen": 1036517376
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034643931795386157,
      "loss": 2.6525,
      "theoretical_loss": 3.6367759021189263,
      "tokens_seen": 1036582912
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003464292878635908,
      "loss": 2.6791,
      "theoretical_loss": 3.63675416650484,
      "tokens_seen": 1036648448
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034641925777331993,
      "loss": 2.9316,
      "theoretical_loss": 3.6367324326495387,
      "tokens_seen": 1036713984
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034640922768304917,
      "loss": 2.6395,
      "theoretical_loss": 3.6367107005527686,
      "tokens_seen": 1036779520
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034639919759277835,
      "loss": 2.9279,
      "theoretical_loss": 3.6366889702142764,
      "tokens_seen": 1036845056
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034638916750250753,
      "loss": 2.7249,
      "theoretical_loss": 3.636667241633809,
      "tokens_seen": 1036910592
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003463791374122367,
      "loss": 2.7696,
      "theoretical_loss": 3.6366455148111134,
      "tokens_seen": 1036976128
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003463691073219659,
      "loss": 2.7503,
      "theoretical_loss": 3.6366237897459355,
      "tokens_seen": 1037041664
    },
    {
      "epoch": 0.31,
      "objective/train/advantage_avg": 0.49424177408218384,
      "objective/train/docs_used": 594513,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.825852155685425,
      "objective/train/original_loss": 2.8258516788482666,
      "objective/train/theoretical_loss": 3.636602066438022,
      "objective/train/tokens_used": 1057567200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24432311952114105,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506662130355835,
      "objective/train/weighted_lm_loss": 2.9690215587615967,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0456793308258057,
      "theoretical_loss": 3.636602066438022,
      "tokens_seen": 1037107200
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003463590772316951,
      "loss": 2.9097,
      "theoretical_loss": 3.636602066438022,
      "tokens_seen": 1037107200
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003463490471414243,
      "loss": 2.8686,
      "theoretical_loss": 3.636580344887121,
      "tokens_seen": 1037172736
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034633901705115344,
      "loss": 2.8067,
      "theoretical_loss": 3.6365586250929782,
      "tokens_seen": 1037238272
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034632898696088267,
      "loss": 2.5973,
      "theoretical_loss": 3.6365369070553406,
      "tokens_seen": 1037303808
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003463189568706118,
      "loss": 2.9614,
      "theoretical_loss": 3.636515190773956,
      "tokens_seen": 1037369344
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034630892678034103,
      "loss": 2.9112,
      "theoretical_loss": 3.6364934762485706,
      "tokens_seen": 1037434880
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003462988966900702,
      "loss": 2.9192,
      "theoretical_loss": 3.636471763478932,
      "tokens_seen": 1037500416
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003462888665997994,
      "loss": 2.7554,
      "theoretical_loss": 3.6364500524647867,
      "tokens_seen": 1037565952
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003462788365095286,
      "loss": 3.0566,
      "theoretical_loss": 3.636428343205883,
      "tokens_seen": 1037631488
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003462688064192578,
      "loss": 2.9137,
      "theoretical_loss": 3.636406635701967,
      "tokens_seen": 1037697024
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034625877632898694,
      "loss": 2.7963,
      "theoretical_loss": 3.6363849299527864,
      "tokens_seen": 1037762560
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003462487462387162,
      "loss": 2.5264,
      "theoretical_loss": 3.636363225958089,
      "tokens_seen": 1037828096
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003462387161484453,
      "loss": 2.9888,
      "theoretical_loss": 3.636341523717622,
      "tokens_seen": 1037893632
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034622868605817454,
      "loss": 2.9254,
      "theoretical_loss": 3.6363198232311325,
      "tokens_seen": 1037959168
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003462186559679037,
      "loss": 2.916,
      "theoretical_loss": 3.636298124498368,
      "tokens_seen": 1038024704
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003462086258776329,
      "loss": 2.9589,
      "theoretical_loss": 3.6362764275190766,
      "tokens_seen": 1038090240
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003461985957873621,
      "loss": 2.7862,
      "theoretical_loss": 3.636254732293006,
      "tokens_seen": 1038155776
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034618856569709126,
      "loss": 3.0236,
      "theoretical_loss": 3.6362330388199027,
      "tokens_seen": 1038221312
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034617853560682044,
      "loss": 2.6608,
      "theoretical_loss": 3.6362113470995157,
      "tokens_seen": 1038286848
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003461685055165497,
      "loss": 2.9059,
      "theoretical_loss": 3.6361896571315926,
      "tokens_seen": 1038352384
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003461584754262788,
      "loss": 2.9895,
      "theoretical_loss": 3.6361679689158803,
      "tokens_seen": 1038417920
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034614844533600804,
      "loss": 2.7835,
      "theoretical_loss": 3.6361462824521276,
      "tokens_seen": 1038483456
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034613841524573717,
      "loss": 2.5993,
      "theoretical_loss": 3.636124597740082,
      "tokens_seen": 1038548992
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003461283851554664,
      "loss": 3.0121,
      "theoretical_loss": 3.6361029147794914,
      "tokens_seen": 1038614528
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003461183550651956,
      "loss": 2.8918,
      "theoretical_loss": 3.6360812335701045,
      "tokens_seen": 1038680064
    },
    {
      "epoch": 0.31,
      "objective/train/advantage_avg": 0.4948311448097229,
      "objective/train/docs_used": 595196,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.797086477279663,
      "objective/train/original_loss": 2.797086238861084,
      "objective/train/theoretical_loss": 3.636059554111668,
      "objective/train/tokens_used": 1059205600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24494938552379608,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507283210754395,
      "objective/train/weighted_lm_loss": 2.9390549659729004,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0338950157165527,
      "theoretical_loss": 3.636059554111668,
      "tokens_seen": 1038745600
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034610832497492476,
      "loss": 2.9227,
      "theoretical_loss": 3.636059554111668,
      "tokens_seen": 1038745600
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034609829488465395,
      "loss": 2.7421,
      "theoretical_loss": 3.6360378764039316,
      "tokens_seen": 1038811136
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003460882647943832,
      "loss": 2.7675,
      "theoretical_loss": 3.636016200446643,
      "tokens_seen": 1038876672
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034607823470411236,
      "loss": 2.5848,
      "theoretical_loss": 3.63599452623955,
      "tokens_seen": 1038942208
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034606820461384154,
      "loss": 2.7058,
      "theoretical_loss": 3.6359728537824014,
      "tokens_seen": 1039007744
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003460581745235707,
      "loss": 3.0162,
      "theoretical_loss": 3.6359511830749454,
      "tokens_seen": 1039073280
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003460481444332999,
      "loss": 2.6292,
      "theoretical_loss": 3.6359295141169303,
      "tokens_seen": 1039138816
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034603811434302914,
      "loss": 2.8296,
      "theoretical_loss": 3.635907846908104,
      "tokens_seen": 1039204352
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034602808425275827,
      "loss": 2.8376,
      "theoretical_loss": 3.635886181448216,
      "tokens_seen": 1039269888
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.0003460180541624875,
      "loss": 2.8056,
      "theoretical_loss": 3.635864517737015,
      "tokens_seen": 1039335424
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034600802407221663,
      "loss": 2.7996,
      "theoretical_loss": 3.635842855774248,
      "tokens_seen": 1039400960
    },
    {
      "epoch": 0.31,
      "learning_rate": 0.00034599799398194587,
      "loss": 2.8196,
      "theoretical_loss": 3.6358211955596653,
      "tokens_seen": 1039466496
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034598796389167505,
      "loss": 2.6337,
      "theoretical_loss": 3.635799537093015,
      "tokens_seen": 1039532032
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034597793380140423,
      "loss": 2.9727,
      "theoretical_loss": 3.6357778803740457,
      "tokens_seen": 1039597568
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003459679037111334,
      "loss": 2.9985,
      "theoretical_loss": 3.635756225402506,
      "tokens_seen": 1039663104
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034595787362086264,
      "loss": 2.7662,
      "theoretical_loss": 3.635734572178146,
      "tokens_seen": 1039728640
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034594784353059177,
      "loss": 2.7949,
      "theoretical_loss": 3.6357129207007137,
      "tokens_seen": 1039794176
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.000345937813440321,
      "loss": 2.6737,
      "theoretical_loss": 3.635691270969958,
      "tokens_seen": 1039859712
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034592778335005013,
      "loss": 3.0372,
      "theoretical_loss": 3.6356696229856276,
      "tokens_seen": 1039925248
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034591775325977937,
      "loss": 2.7248,
      "theoretical_loss": 3.6356479767474723,
      "tokens_seen": 1039990784
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034590772316950855,
      "loss": 2.818,
      "theoretical_loss": 3.6356263322552413,
      "tokens_seen": 1040056320
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034589769307923773,
      "loss": 2.8065,
      "theoretical_loss": 3.635604689508683,
      "tokens_seen": 1040121856
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003458876629889669,
      "loss": 2.7958,
      "theoretical_loss": 3.6355830485075473,
      "tokens_seen": 1040187392
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003458776328986961,
      "loss": 2.9495,
      "theoretical_loss": 3.6355614092515833,
      "tokens_seen": 1040252928
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003458676028084253,
      "loss": 2.8257,
      "theoretical_loss": 3.6355397717405404,
      "tokens_seen": 1040318464
    },
    {
      "epoch": 0.32,
      "objective/train/advantage_avg": 0.49390941858291626,
      "objective/train/docs_used": 596146,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.828624725341797,
      "objective/train/original_loss": 2.828624725341797,
      "objective/train/theoretical_loss": 3.6355181359741673,
      "objective/train/tokens_used": 1060844000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2440178394317627,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506314039230347,
      "objective/train/weighted_lm_loss": 2.971494436264038,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.04560387134552,
      "theoretical_loss": 3.6355181359741673,
      "tokens_seen": 1040384000
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003458575727181545,
      "loss": 3.1479,
      "theoretical_loss": 3.6355181359741673,
      "tokens_seen": 1040384000
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034584754262788364,
      "loss": 3.0218,
      "theoretical_loss": 3.6354965019522147,
      "tokens_seen": 1040449536
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034583751253761287,
      "loss": 2.8377,
      "theoretical_loss": 3.635474869674431,
      "tokens_seen": 1040515072
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.000345827482447342,
      "loss": 2.9298,
      "theoretical_loss": 3.635453239140566,
      "tokens_seen": 1040580608
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034581745235707123,
      "loss": 2.651,
      "theoretical_loss": 3.63543161035037,
      "tokens_seen": 1040646144
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003458074222668004,
      "loss": 2.8164,
      "theoretical_loss": 3.6354099833035916,
      "tokens_seen": 1040711680
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003457973921765296,
      "loss": 2.842,
      "theoretical_loss": 3.6353883579999815,
      "tokens_seen": 1040777216
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003457873620862588,
      "loss": 2.8291,
      "theoretical_loss": 3.6353667344392884,
      "tokens_seen": 1040842752
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.000345777331995988,
      "loss": 2.9259,
      "theoretical_loss": 3.6353451126212626,
      "tokens_seen": 1040908288
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034576730190571714,
      "loss": 2.8169,
      "theoretical_loss": 3.6353234925456546,
      "tokens_seen": 1040973824
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003457572718154464,
      "loss": 2.7659,
      "theoretical_loss": 3.635301874212213,
      "tokens_seen": 1041039360
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003457472417251755,
      "loss": 2.8818,
      "theoretical_loss": 3.635280257620688,
      "tokens_seen": 1041104896
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034573721163490474,
      "loss": 2.6618,
      "theoretical_loss": 3.6352586427708315,
      "tokens_seen": 1041170432
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003457271815446339,
      "loss": 2.753,
      "theoretical_loss": 3.635237029662391,
      "tokens_seen": 1041235968
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003457171514543631,
      "loss": 2.9099,
      "theoretical_loss": 3.635215418295118,
      "tokens_seen": 1041301504
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003457071213640923,
      "loss": 2.6518,
      "theoretical_loss": 3.6351938086687623,
      "tokens_seen": 1041367040
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034569709127382146,
      "loss": 2.8743,
      "theoretical_loss": 3.6351722007830745,
      "tokens_seen": 1041432576
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034568706118355064,
      "loss": 2.7064,
      "theoretical_loss": 3.635150594637804,
      "tokens_seen": 1041498112
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003456770310932799,
      "loss": 2.8571,
      "theoretical_loss": 3.635128990232702,
      "tokens_seen": 1041563648
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.000345667001003009,
      "loss": 2.6766,
      "theoretical_loss": 3.635107387567518,
      "tokens_seen": 1041629184
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034565697091273824,
      "loss": 2.7017,
      "theoretical_loss": 3.6350857866420037,
      "tokens_seen": 1041694720
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034564694082246737,
      "loss": 3.0708,
      "theoretical_loss": 3.6350641874559084,
      "tokens_seen": 1041760256
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003456369107321966,
      "loss": 2.7023,
      "theoretical_loss": 3.6350425900089833,
      "tokens_seen": 1041825792
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003456268806419258,
      "loss": 2.8028,
      "theoretical_loss": 3.6350209943009784,
      "tokens_seen": 1041891328
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034561685055165496,
      "loss": 3.0595,
      "theoretical_loss": 3.634999400331645,
      "tokens_seen": 1041956864
    },
    {
      "epoch": 0.32,
      "objective/train/advantage_avg": 0.4938527047634125,
      "objective/train/docs_used": 596810,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.758863925933838,
      "objective/train/original_loss": 2.7588634490966797,
      "objective/train/theoretical_loss": 3.6349778081007327,
      "objective/train/tokens_used": 1062482400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24395543336868286,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506254434585571,
      "objective/train/weighted_lm_loss": 2.898405075073242,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0442860126495361,
      "theoretical_loss": 3.6349778081007327,
      "tokens_seen": 1042022400
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034560682046138415,
      "loss": 2.6579,
      "theoretical_loss": 3.6349778081007327,
      "tokens_seen": 1042022400
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003455967903711134,
      "loss": 2.6444,
      "theoretical_loss": 3.6349562176079933,
      "tokens_seen": 1042087936
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003455867602808425,
      "loss": 2.8793,
      "theoretical_loss": 3.634934628853177,
      "tokens_seen": 1042153472
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034557673019057174,
      "loss": 2.7631,
      "theoretical_loss": 3.634913041836035,
      "tokens_seen": 1042219008
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034556670010030087,
      "loss": 2.8076,
      "theoretical_loss": 3.6348914565563186,
      "tokens_seen": 1042284544
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003455566700100301,
      "loss": 2.7474,
      "theoretical_loss": 3.6348698730137774,
      "tokens_seen": 1042350080
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003455466399197593,
      "loss": 2.5696,
      "theoretical_loss": 3.634848291208163,
      "tokens_seen": 1042415616
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034553660982948847,
      "loss": 2.9292,
      "theoretical_loss": 3.6348267111392274,
      "tokens_seen": 1042481152
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034552657973921765,
      "loss": 2.7069,
      "theoretical_loss": 3.6348051328067204,
      "tokens_seen": 1042546688
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034551654964894683,
      "loss": 2.8008,
      "theoretical_loss": 3.6347835562103934,
      "tokens_seen": 1042612224
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.000345506519558676,
      "loss": 2.8279,
      "theoretical_loss": 3.634761981349998,
      "tokens_seen": 1042677760
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034549648946840525,
      "loss": 3.0921,
      "theoretical_loss": 3.634740408225285,
      "tokens_seen": 1042743296
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003454864593781344,
      "loss": 2.8684,
      "theoretical_loss": 3.634718836836006,
      "tokens_seen": 1042808832
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003454764292878636,
      "loss": 2.6973,
      "theoretical_loss": 3.634697267181912,
      "tokens_seen": 1042874368
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034546639919759274,
      "loss": 2.9829,
      "theoretical_loss": 3.634675699262756,
      "tokens_seen": 1042939904
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034545636910732197,
      "loss": 2.8996,
      "theoretical_loss": 3.6346541330782864,
      "tokens_seen": 1043005440
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034544633901705115,
      "loss": 2.8413,
      "theoretical_loss": 3.6346325686282572,
      "tokens_seen": 1043070976
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034543630892678033,
      "loss": 2.8139,
      "theoretical_loss": 3.634611005912419,
      "tokens_seen": 1043136512
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003454262788365095,
      "loss": 2.8935,
      "theoretical_loss": 3.6345894449305236,
      "tokens_seen": 1043202048
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034541624874623875,
      "loss": 2.9296,
      "theoretical_loss": 3.6345678856823227,
      "tokens_seen": 1043267584
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003454062186559679,
      "loss": 2.7544,
      "theoretical_loss": 3.6345463281675676,
      "tokens_seen": 1043333120
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003453961885656971,
      "loss": 2.7873,
      "theoretical_loss": 3.6345247723860106,
      "tokens_seen": 1043398656
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034538615847542624,
      "loss": 3.1202,
      "theoretical_loss": 3.634503218337403,
      "tokens_seen": 1043464192
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003453761283851555,
      "loss": 2.8241,
      "theoretical_loss": 3.6344816660214967,
      "tokens_seen": 1043529728
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034536609829488466,
      "loss": 3.0594,
      "theoretical_loss": 3.6344601154380443,
      "tokens_seen": 1043595264
    },
    {
      "epoch": 0.32,
      "objective/train/advantage_avg": 0.4947175085544586,
      "objective/train/docs_used": 598197,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.8787076473236084,
      "objective/train/original_loss": 2.87870717048645,
      "objective/train/theoretical_loss": 3.6344385665867973,
      "objective/train/tokens_used": 1064120800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24482139945030212,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507162809371948,
      "objective/train/weighted_lm_loss": 3.024608612060547,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0413066148757935,
      "theoretical_loss": 3.6344385665867973,
      "tokens_seen": 1043660800
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034535606820461384,
      "loss": 3.0627,
      "theoretical_loss": 3.6344385665867973,
      "tokens_seen": 1043660800
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.000345346038114343,
      "loss": 2.6434,
      "theoretical_loss": 3.6344170194675076,
      "tokens_seen": 1043726336
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003453360080240722,
      "loss": 2.7493,
      "theoretical_loss": 3.634395474079927,
      "tokens_seen": 1043791872
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034532597793380143,
      "loss": 2.7171,
      "theoretical_loss": 3.6343739304238083,
      "tokens_seen": 1043857408
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003453159478435306,
      "loss": 2.9821,
      "theoretical_loss": 3.634352388498903,
      "tokens_seen": 1043922944
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003453059177532598,
      "loss": 2.8834,
      "theoretical_loss": 3.6343308483049634,
      "tokens_seen": 1043988480
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.000345295887662989,
      "loss": 2.9265,
      "theoretical_loss": 3.6343093098417425,
      "tokens_seen": 1044054016
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003452858575727182,
      "loss": 2.6604,
      "theoretical_loss": 3.634287773108992,
      "tokens_seen": 1044119552
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034527582748244734,
      "loss": 2.986,
      "theoretical_loss": 3.6342662381064645,
      "tokens_seen": 1044185088
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003452657973921766,
      "loss": 3.0415,
      "theoretical_loss": 3.634244704833912,
      "tokens_seen": 1044250624
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003452557673019057,
      "loss": 2.7851,
      "theoretical_loss": 3.6342231732910872,
      "tokens_seen": 1044316160
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034524573721163494,
      "loss": 2.8712,
      "theoretical_loss": 3.6342016434777427,
      "tokens_seen": 1044381696
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003452357071213641,
      "loss": 2.6816,
      "theoretical_loss": 3.6341801153936313,
      "tokens_seen": 1044447232
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003452256770310933,
      "loss": 2.9142,
      "theoretical_loss": 3.634158589038505,
      "tokens_seen": 1044512768
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003452156469408225,
      "loss": 3.023,
      "theoretical_loss": 3.634137064412117,
      "tokens_seen": 1044578304
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034520561685055166,
      "loss": 2.7764,
      "theoretical_loss": 3.6341155415142197,
      "tokens_seen": 1044643840
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034519558676028084,
      "loss": 2.7777,
      "theoretical_loss": 3.634094020344566,
      "tokens_seen": 1044709376
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003451855566700101,
      "loss": 2.8131,
      "theoretical_loss": 3.634072500902909,
      "tokens_seen": 1044774912
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003451755265797392,
      "loss": 2.8102,
      "theoretical_loss": 3.6340509831890015,
      "tokens_seen": 1044840448
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034516549648946844,
      "loss": 2.9261,
      "theoretical_loss": 3.6340294672025957,
      "tokens_seen": 1044905984
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034515546639919757,
      "loss": 2.9743,
      "theoretical_loss": 3.6340079529434455,
      "tokens_seen": 1044971520
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003451454363089268,
      "loss": 2.7665,
      "theoretical_loss": 3.633986440411303,
      "tokens_seen": 1045037056
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.000345135406218656,
      "loss": 2.9649,
      "theoretical_loss": 3.6339649296059227,
      "tokens_seen": 1045102592
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034512537612838517,
      "loss": 2.8788,
      "theoretical_loss": 3.633943420527056,
      "tokens_seen": 1045168128
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034511534603811435,
      "loss": 2.8916,
      "theoretical_loss": 3.6339219131744573,
      "tokens_seen": 1045233664
    },
    {
      "epoch": 0.32,
      "objective/train/advantage_avg": 0.49428799748420715,
      "objective/train/docs_used": 598880,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.747431516647339,
      "objective/train/original_loss": 2.747431516647339,
      "objective/train/theoretical_loss": 3.6339004075478796,
      "objective/train/tokens_used": 1065759200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24438294768333435,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050671100616455,
      "objective/train/weighted_lm_loss": 2.886744260787964,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.044721245765686,
      "theoretical_loss": 3.6339004075478796,
      "tokens_seen": 1045299200
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003451053159478436,
      "loss": 3.0916,
      "theoretical_loss": 3.6339004075478796,
      "tokens_seen": 1045299200
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003450952858575727,
      "loss": 2.8088,
      "theoretical_loss": 3.6338789036470756,
      "tokens_seen": 1045364736
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034508525576730194,
      "loss": 3.1959,
      "theoretical_loss": 3.6338574014717997,
      "tokens_seen": 1045430272
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034507522567703107,
      "loss": 2.7614,
      "theoretical_loss": 3.633835901021804,
      "tokens_seen": 1045495808
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003450651955867603,
      "loss": 2.9016,
      "theoretical_loss": 3.633814402296843,
      "tokens_seen": 1045561344
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003450551654964895,
      "loss": 2.9772,
      "theoretical_loss": 3.6337929052966693,
      "tokens_seen": 1045626880
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034504513540621867,
      "loss": 2.9446,
      "theoretical_loss": 3.633771410021038,
      "tokens_seen": 1045692416
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034503510531594785,
      "loss": 2.8992,
      "theoretical_loss": 3.6337499164697005,
      "tokens_seen": 1045757952
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034502507522567703,
      "loss": 2.7014,
      "theoretical_loss": 3.633728424642412,
      "tokens_seen": 1045823488
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003450150451354062,
      "loss": 2.85,
      "theoretical_loss": 3.6337069345389255,
      "tokens_seen": 1045889024
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034500501504513545,
      "loss": 2.7928,
      "theoretical_loss": 3.633685446158995,
      "tokens_seen": 1045954560
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003449949849548646,
      "loss": 2.9827,
      "theoretical_loss": 3.6336639595023743,
      "tokens_seen": 1046020096
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003449849548645938,
      "loss": 2.8329,
      "theoretical_loss": 3.633642474568817,
      "tokens_seen": 1046085632
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034497492477432294,
      "loss": 3.0235,
      "theoretical_loss": 3.6336209913580775,
      "tokens_seen": 1046151168
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034496489468405217,
      "loss": 2.8706,
      "theoretical_loss": 3.633599509869909,
      "tokens_seen": 1046216704
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034495486459378135,
      "loss": 2.9372,
      "theoretical_loss": 3.6335780301040663,
      "tokens_seen": 1046282240
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034494483450351053,
      "loss": 2.7175,
      "theoretical_loss": 3.633556552060303,
      "tokens_seen": 1046347776
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003449348044132397,
      "loss": 2.8594,
      "theoretical_loss": 3.633535075738373,
      "tokens_seen": 1046413312
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034492477432296895,
      "loss": 2.7632,
      "theoretical_loss": 3.6335136011380307,
      "tokens_seen": 1046478848
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003449147442326981,
      "loss": 2.8433,
      "theoretical_loss": 3.6334921282590305,
      "tokens_seen": 1046544384
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003449047141424273,
      "loss": 3.266,
      "theoretical_loss": 3.633470657101126,
      "tokens_seen": 1046609920
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034489468405215644,
      "loss": 2.9206,
      "theoretical_loss": 3.6334491876640724,
      "tokens_seen": 1046675456
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003448846539618857,
      "loss": 3.0055,
      "theoretical_loss": 3.633427719947623,
      "tokens_seen": 1046740992
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034487462387161486,
      "loss": 2.7515,
      "theoretical_loss": 3.633406253951533,
      "tokens_seen": 1046806528
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034486459378134404,
      "loss": 2.706,
      "theoretical_loss": 3.6333847896755564,
      "tokens_seen": 1046872064
    },
    {
      "epoch": 0.32,
      "objective/train/advantage_avg": 0.49478012323379517,
      "objective/train/docs_used": 599993,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.7842276096343994,
      "objective/train/original_loss": 2.7842280864715576,
      "objective/train/theoretical_loss": 3.633363327119448,
      "objective/train/tokens_used": 1067397600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24487288296222687,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507228374481201,
      "objective/train/weighted_lm_loss": 2.925368070602417,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0253151655197144,
      "theoretical_loss": 3.633363327119448,
      "tokens_seen": 1046937600
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003448545636910732,
      "loss": 2.9597,
      "theoretical_loss": 3.633363327119448,
      "tokens_seen": 1046937600
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003448445336008024,
      "loss": 2.8608,
      "theoretical_loss": 3.6333418662829624,
      "tokens_seen": 1047003136
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003448345035105316,
      "loss": 2.9921,
      "theoretical_loss": 3.633320407165854,
      "tokens_seen": 1047068672
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003448244734202608,
      "loss": 3.0605,
      "theoretical_loss": 3.633298949767877,
      "tokens_seen": 1047134208
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034481444332998994,
      "loss": 2.8099,
      "theoretical_loss": 3.633277494088787,
      "tokens_seen": 1047199744
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003448044132397192,
      "loss": 3.0106,
      "theoretical_loss": 3.6332560401283382,
      "tokens_seen": 1047265280
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034479438314944836,
      "loss": 2.8678,
      "theoretical_loss": 3.6332345878862853,
      "tokens_seen": 1047330816
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034478435305917754,
      "loss": 2.7218,
      "theoretical_loss": 3.6332131373623833,
      "tokens_seen": 1047396352
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003447743229689067,
      "loss": 2.7862,
      "theoretical_loss": 3.633191688556388,
      "tokens_seen": 1047461888
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003447642928786359,
      "loss": 2.9214,
      "theoretical_loss": 3.6331702414680525,
      "tokens_seen": 1047527424
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003447542627883651,
      "loss": 3.0842,
      "theoretical_loss": 3.6331487960971334,
      "tokens_seen": 1047592960
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003447442326980943,
      "loss": 2.7183,
      "theoretical_loss": 3.6331273524433847,
      "tokens_seen": 1047658496
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034473420260782345,
      "loss": 3.0227,
      "theoretical_loss": 3.633105910506562,
      "tokens_seen": 1047724032
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003447241725175527,
      "loss": 2.7972,
      "theoretical_loss": 3.6330844702864202,
      "tokens_seen": 1047789568
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003447141424272818,
      "loss": 2.7216,
      "theoretical_loss": 3.633063031782715,
      "tokens_seen": 1047855104
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034470411233701104,
      "loss": 2.7382,
      "theoretical_loss": 3.633041594995201,
      "tokens_seen": 1047920640
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003446940822467402,
      "loss": 2.744,
      "theoretical_loss": 3.6330201599236345,
      "tokens_seen": 1047986176
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003446840521564694,
      "loss": 2.8779,
      "theoretical_loss": 3.63299872656777,
      "tokens_seen": 1048051712
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003446740220661986,
      "loss": 2.9427,
      "theoretical_loss": 3.6329772949273624,
      "tokens_seen": 1048117248
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034466399197592777,
      "loss": 2.6941,
      "theoretical_loss": 3.6329558650021685,
      "tokens_seen": 1048182784
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034465396188565695,
      "loss": 3.0016,
      "theoretical_loss": 3.6329344367919427,
      "tokens_seen": 1048248320
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003446439317953862,
      "loss": 2.8832,
      "theoretical_loss": 3.6329130102964413,
      "tokens_seen": 1048313856
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003446339017051153,
      "loss": 3.0026,
      "theoretical_loss": 3.6328915855154196,
      "tokens_seen": 1048379392
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034462387161484455,
      "loss": 2.8816,
      "theoretical_loss": 3.632870162448633,
      "tokens_seen": 1048444928
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034461384152457373,
      "loss": 2.8526,
      "theoretical_loss": 3.632848741095838,
      "tokens_seen": 1048510464
    },
    {
      "epoch": 0.32,
      "objective/train/advantage_avg": 0.4950122535228729,
      "objective/train/docs_used": 601088,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.6230123043060303,
      "objective/train/original_loss": 2.623012065887451,
      "objective/train/theoretical_loss": 3.632827321456789,
      "objective/train/tokens_used": 1069036000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24507974088191986,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507471561431885,
      "objective/train/weighted_lm_loss": 2.7560627460479736,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0396549701690674,
      "theoretical_loss": 3.632827321456789,
      "tokens_seen": 1048576000
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003446038114343029,
      "loss": 2.9641,
      "theoretical_loss": 3.632827321456789,
      "tokens_seen": 1048576000
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034459378134403214,
      "loss": 2.8757,
      "theoretical_loss": 3.632805903531243,
      "tokens_seen": 1048641536
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034458375125376127,
      "loss": 2.752,
      "theoretical_loss": 3.6327844873189554,
      "tokens_seen": 1048707072
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003445737211634905,
      "loss": 3.0149,
      "theoretical_loss": 3.632763072819682,
      "tokens_seen": 1048772608
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003445636910732197,
      "loss": 2.7312,
      "theoretical_loss": 3.632741660033179,
      "tokens_seen": 1048838144
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034455366098294887,
      "loss": 2.7695,
      "theoretical_loss": 3.6327202489592025,
      "tokens_seen": 1048903680
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034454363089267805,
      "loss": 2.9142,
      "theoretical_loss": 3.632698839597508,
      "tokens_seen": 1048969216
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034453360080240723,
      "loss": 2.8192,
      "theoretical_loss": 3.632677431947852,
      "tokens_seen": 1049034752
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003445235707121364,
      "loss": 2.7706,
      "theoretical_loss": 3.6326560260099905,
      "tokens_seen": 1049100288
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034451354062186565,
      "loss": 2.7266,
      "theoretical_loss": 3.63263462178368,
      "tokens_seen": 1049165824
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003445035105315948,
      "loss": 2.6513,
      "theoretical_loss": 3.632613219268676,
      "tokens_seen": 1049231360
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.000344493480441324,
      "loss": 3.0647,
      "theoretical_loss": 3.6325918184647366,
      "tokens_seen": 1049296896
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034448345035105314,
      "loss": 2.9309,
      "theoretical_loss": 3.632570419371616,
      "tokens_seen": 1049362432
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034447342026078237,
      "loss": 2.643,
      "theoretical_loss": 3.632549021989071,
      "tokens_seen": 1049427968
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034446339017051155,
      "loss": 2.94,
      "theoretical_loss": 3.6325276263168593,
      "tokens_seen": 1049493504
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034445336008024073,
      "loss": 2.9084,
      "theoretical_loss": 3.6325062323547366,
      "tokens_seen": 1049559040
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003444433299899699,
      "loss": 2.945,
      "theoretical_loss": 3.6324848401024594,
      "tokens_seen": 1049624576
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034443329989969915,
      "loss": 3.1848,
      "theoretical_loss": 3.6324634495597845,
      "tokens_seen": 1049690112
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003444232698094283,
      "loss": 2.819,
      "theoretical_loss": 3.6324420607264685,
      "tokens_seen": 1049755648
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003444132397191575,
      "loss": 2.8245,
      "theoretical_loss": 3.632420673602268,
      "tokens_seen": 1049821184
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034440320962888664,
      "loss": 2.9491,
      "theoretical_loss": 3.6323992881869396,
      "tokens_seen": 1049886720
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003443931795386159,
      "loss": 2.8826,
      "theoretical_loss": 3.6323779044802404,
      "tokens_seen": 1049952256
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034438314944834506,
      "loss": 2.9108,
      "theoretical_loss": 3.632356522481927,
      "tokens_seen": 1050017792
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034437311935807424,
      "loss": 2.7508,
      "theoretical_loss": 3.6323351421917565,
      "tokens_seen": 1050083328
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003443630892678034,
      "loss": 2.9986,
      "theoretical_loss": 3.6323137636094858,
      "tokens_seen": 1050148864
    },
    {
      "epoch": 0.32,
      "objective/train/advantage_avg": 0.4948548972606659,
      "objective/train/docs_used": 601638,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.697977304458618,
      "objective/train/original_loss": 2.697977304458618,
      "objective/train/theoretical_loss": 3.632292386734872,
      "objective/train/tokens_used": 1070674400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24494965374469757,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507307052612305,
      "objective/train/weighted_lm_loss": 2.8348426818847656,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0425469875335693,
      "theoretical_loss": 3.632292386734872,
      "tokens_seen": 1050214400
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003443530591775326,
      "loss": 2.7899,
      "theoretical_loss": 3.632292386734872,
      "tokens_seen": 1050214400
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003443430290872618,
      "loss": 2.7452,
      "theoretical_loss": 3.6322710115676724,
      "tokens_seen": 1050279936
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.000344332998996991,
      "loss": 2.9558,
      "theoretical_loss": 3.632249638107643,
      "tokens_seen": 1050345472
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034432296890672014,
      "loss": 3.1298,
      "theoretical_loss": 3.6322282663545424,
      "tokens_seen": 1050411008
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003443129388164494,
      "loss": 3.0853,
      "theoretical_loss": 3.6322068963081264,
      "tokens_seen": 1050476544
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034430290872617856,
      "loss": 3.0356,
      "theoretical_loss": 3.6321855279681534,
      "tokens_seen": 1050542080
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034429287863590774,
      "loss": 2.8866,
      "theoretical_loss": 3.63216416133438,
      "tokens_seen": 1050607616
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003442828485456369,
      "loss": 3.0387,
      "theoretical_loss": 3.632142796406564,
      "tokens_seen": 1050673152
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003442728184553661,
      "loss": 2.8104,
      "theoretical_loss": 3.6321214331844622,
      "tokens_seen": 1050738688
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003442627883650953,
      "loss": 2.8324,
      "theoretical_loss": 3.6321000716678324,
      "tokens_seen": 1050804224
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003442527582748245,
      "loss": 2.9784,
      "theoretical_loss": 3.6320787118564324,
      "tokens_seen": 1050869760
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034424272818455365,
      "loss": 3.0025,
      "theoretical_loss": 3.6320573537500196,
      "tokens_seen": 1050935296
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003442326980942829,
      "loss": 2.9899,
      "theoretical_loss": 3.6320359973483507,
      "tokens_seen": 1051000832
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.000344222668004012,
      "loss": 2.5865,
      "theoretical_loss": 3.632014642651185,
      "tokens_seen": 1051066368
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034421263791374124,
      "loss": 2.7976,
      "theoretical_loss": 3.6319932896582787,
      "tokens_seen": 1051131904
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003442026078234704,
      "loss": 2.8125,
      "theoretical_loss": 3.6319719383693903,
      "tokens_seen": 1051197440
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003441925777331996,
      "loss": 2.9228,
      "theoretical_loss": 3.631950588784278,
      "tokens_seen": 1051262976
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003441825476429288,
      "loss": 3.0032,
      "theoretical_loss": 3.631929240902698,
      "tokens_seen": 1051328512
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034417251755265797,
      "loss": 2.9173,
      "theoretical_loss": 3.63190789472441,
      "tokens_seen": 1051394048
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034416248746238715,
      "loss": 2.6996,
      "theoretical_loss": 3.6318865502491713,
      "tokens_seen": 1051459584
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003441524573721164,
      "loss": 2.6682,
      "theoretical_loss": 3.6318652074767392,
      "tokens_seen": 1051525120
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003441424272818455,
      "loss": 2.9088,
      "theoretical_loss": 3.6318438664068724,
      "tokens_seen": 1051590656
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034413239719157475,
      "loss": 2.9302,
      "theoretical_loss": 3.631822527039329,
      "tokens_seen": 1051656192
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034412236710130393,
      "loss": 2.8945,
      "theoretical_loss": 3.631801189373867,
      "tokens_seen": 1051721728
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003441123370110331,
      "loss": 2.9922,
      "theoretical_loss": 3.6317798534102455,
      "tokens_seen": 1051787264
    },
    {
      "epoch": 0.32,
      "objective/train/advantage_avg": 0.49300870299339294,
      "objective/train/docs_used": 602388,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.290477991104126,
      "objective/train/original_loss": 2.290477752685547,
      "objective/train/theoretical_loss": 3.631758519148221,
      "objective/train/tokens_used": 1072312800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24312853813171387,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505367517471313,
      "objective/train/weighted_lm_loss": 2.406148910522461,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0432155132293701,
      "theoretical_loss": 3.631758519148221,
      "tokens_seen": 1051852800
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003441023069207623,
      "loss": 2.7539,
      "theoretical_loss": 3.631758519148221,
      "tokens_seen": 1051852800
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034409227683049147,
      "loss": 2.8248,
      "theoretical_loss": 3.6317371865875527,
      "tokens_seen": 1051918336
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034408224674022065,
      "loss": 2.7864,
      "theoretical_loss": 3.631715855727999,
      "tokens_seen": 1051983872
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003440722166499499,
      "loss": 3.2266,
      "theoretical_loss": 3.6316945265693183,
      "tokens_seen": 1052049408
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.000344062186559679,
      "loss": 3.0299,
      "theoretical_loss": 3.631673199111269,
      "tokens_seen": 1052114944
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034405215646940825,
      "loss": 2.7668,
      "theoretical_loss": 3.6316518733536096,
      "tokens_seen": 1052180480
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003440421263791374,
      "loss": 2.7214,
      "theoretical_loss": 3.631630549296098,
      "tokens_seen": 1052246016
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003440320962888666,
      "loss": 3.0201,
      "theoretical_loss": 3.631609226938494,
      "tokens_seen": 1052311552
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003440220661985958,
      "loss": 2.774,
      "theoretical_loss": 3.631587906280555,
      "tokens_seen": 1052377088
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.000344012036108325,
      "loss": 2.7307,
      "theoretical_loss": 3.6315665873220406,
      "tokens_seen": 1052442624
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034400200601805416,
      "loss": 2.9902,
      "theoretical_loss": 3.631545270062709,
      "tokens_seen": 1052508160
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034399197592778334,
      "loss": 2.9405,
      "theoretical_loss": 3.6315239545023195,
      "tokens_seen": 1052573696
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003439819458375125,
      "loss": 2.747,
      "theoretical_loss": 3.63150264064063,
      "tokens_seen": 1052639232
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034397191574724175,
      "loss": 3.0667,
      "theoretical_loss": 3.6314813284774003,
      "tokens_seen": 1052704768
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003439618856569709,
      "loss": 3.2654,
      "theoretical_loss": 3.631460018012389,
      "tokens_seen": 1052770304
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003439518555667001,
      "loss": 2.664,
      "theoretical_loss": 3.631438709245355,
      "tokens_seen": 1052835840
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003439418254764293,
      "loss": 2.8388,
      "theoretical_loss": 3.6314174021760577,
      "tokens_seen": 1052901376
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003439317953861585,
      "loss": 2.8723,
      "theoretical_loss": 3.631396096804256,
      "tokens_seen": 1052966912
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034392176529588766,
      "loss": 2.7216,
      "theoretical_loss": 3.631374793129708,
      "tokens_seen": 1053032448
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034391173520561684,
      "loss": 3.0129,
      "theoretical_loss": 3.6313534911521743,
      "tokens_seen": 1053097984
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.000343901705115346,
      "loss": 3.0621,
      "theoretical_loss": 3.6313321908714133,
      "tokens_seen": 1053163520
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034389167502507526,
      "loss": 2.8913,
      "theoretical_loss": 3.6313108922871855,
      "tokens_seen": 1053229056
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003438816449348044,
      "loss": 2.8126,
      "theoretical_loss": 3.6312895953992483,
      "tokens_seen": 1053294592
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003438716148445336,
      "loss": 2.9543,
      "theoretical_loss": 3.6312683002073625,
      "tokens_seen": 1053360128
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034386158475426275,
      "loss": 2.8167,
      "theoretical_loss": 3.6312470067112868,
      "tokens_seen": 1053425664
    },
    {
      "epoch": 0.32,
      "objective/train/advantage_avg": 0.4930480122566223,
      "objective/train/docs_used": 603451,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.5924291610717773,
      "objective/train/original_loss": 2.5924289226531982,
      "objective/train/theoretical_loss": 3.6312257149107814,
      "objective/train/tokens_used": 1073951200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24316468834877014,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050540804862976,
      "objective/train/weighted_lm_loss": 2.7231204509735107,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0408108234405518,
      "theoretical_loss": 3.6312257149107814,
      "tokens_seen": 1053491200
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.000343851554663992,
      "loss": 3.0278,
      "theoretical_loss": 3.6312257149107814,
      "tokens_seen": 1053491200
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003438415245737212,
      "loss": 2.9603,
      "theoretical_loss": 3.6312044248056052,
      "tokens_seen": 1053556736
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034383149448345034,
      "loss": 2.8956,
      "theoretical_loss": 3.631183136395518,
      "tokens_seen": 1053622272
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003438214643931796,
      "loss": 3.0329,
      "theoretical_loss": 3.631161849680279,
      "tokens_seen": 1053687808
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034381143430290876,
      "loss": 2.9012,
      "theoretical_loss": 3.631140564659648,
      "tokens_seen": 1053753344
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034380140421263794,
      "loss": 3.1661,
      "theoretical_loss": 3.631119281333386,
      "tokens_seen": 1053818880
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003437913741223671,
      "loss": 2.8026,
      "theoretical_loss": 3.631097999701251,
      "tokens_seen": 1053884416
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003437813440320963,
      "loss": 3.0285,
      "theoretical_loss": 3.6310767197630036,
      "tokens_seen": 1053949952
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003437713139418255,
      "loss": 2.9673,
      "theoretical_loss": 3.6310554415184035,
      "tokens_seen": 1054015488
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003437612838515547,
      "loss": 3.0222,
      "theoretical_loss": 3.631034164967211,
      "tokens_seen": 1054081024
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034375125376128385,
      "loss": 2.9747,
      "theoretical_loss": 3.6310128901091856,
      "tokens_seen": 1054146560
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003437412236710131,
      "loss": 2.8468,
      "theoretical_loss": 3.6309916169440872,
      "tokens_seen": 1054212096
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003437311935807422,
      "loss": 2.9532,
      "theoretical_loss": 3.630970345471676,
      "tokens_seen": 1054277632
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034372116349047144,
      "loss": 3.1005,
      "theoretical_loss": 3.6309490756917127,
      "tokens_seen": 1054343168
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003437111334002006,
      "loss": 2.8838,
      "theoretical_loss": 3.630927807603957,
      "tokens_seen": 1054408704
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003437011033099298,
      "loss": 2.6705,
      "theoretical_loss": 3.6309065412081685,
      "tokens_seen": 1054474240
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.000343691073219659,
      "loss": 2.917,
      "theoretical_loss": 3.630885276504108,
      "tokens_seen": 1054539776
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034368104312938817,
      "loss": 2.9951,
      "theoretical_loss": 3.6308640134915366,
      "tokens_seen": 1054605312
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034367101303911735,
      "loss": 2.8157,
      "theoretical_loss": 3.6308427521702136,
      "tokens_seen": 1054670848
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003436609829488466,
      "loss": 2.8373,
      "theoretical_loss": 3.6308214925398996,
      "tokens_seen": 1054736384
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003436509528585757,
      "loss": 2.8611,
      "theoretical_loss": 3.630800234600355,
      "tokens_seen": 1054801920
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034364092276830495,
      "loss": 2.868,
      "theoretical_loss": 3.6307789783513402,
      "tokens_seen": 1054867456
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034363089267803413,
      "loss": 2.9174,
      "theoretical_loss": 3.630757723792616,
      "tokens_seen": 1054932992
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003436208625877633,
      "loss": 3.0829,
      "theoretical_loss": 3.630736470923943,
      "tokens_seen": 1054998528
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003436108324974925,
      "loss": 2.9299,
      "theoretical_loss": 3.6307152197450825,
      "tokens_seen": 1055064064
    },
    {
      "epoch": 0.32,
      "objective/train/advantage_avg": 0.49416765570640564,
      "objective/train/docs_used": 603451,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.7428653240203857,
      "objective/train/original_loss": 2.7428653240203857,
      "objective/train/theoretical_loss": 3.630693970255794,
      "objective/train/tokens_used": 1075589600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2442743331193924,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506584644317627,
      "objective/train/weighted_lm_loss": 2.881625175476074,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0422987937927246,
      "theoretical_loss": 3.630693970255794,
      "tokens_seen": 1055129600
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034360080240722167,
      "loss": 2.9015,
      "theoretical_loss": 3.630693970255794,
      "tokens_seen": 1055129600
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034359077231695085,
      "loss": 3.1842,
      "theoretical_loss": 3.630672722455839,
      "tokens_seen": 1055195136
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003435807422266801,
      "loss": 2.9664,
      "theoretical_loss": 3.630651476344977,
      "tokens_seen": 1055260672
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003435707121364092,
      "loss": 2.8122,
      "theoretical_loss": 3.6306302319229706,
      "tokens_seen": 1055326208
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034356068204613845,
      "loss": 3.1539,
      "theoretical_loss": 3.6306089891895805,
      "tokens_seen": 1055391744
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003435506519558676,
      "loss": 2.8968,
      "theoretical_loss": 3.630587748144566,
      "tokens_seen": 1055457280
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003435406218655968,
      "loss": 3.1734,
      "theoretical_loss": 3.63056650878769,
      "tokens_seen": 1055522816
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.000343530591775326,
      "loss": 2.8081,
      "theoretical_loss": 3.630545271118713,
      "tokens_seen": 1055588352
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003435205616850552,
      "loss": 2.9286,
      "theoretical_loss": 3.6305240351373946,
      "tokens_seen": 1055653888
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034351053159478436,
      "loss": 2.7879,
      "theoretical_loss": 3.630502800843498,
      "tokens_seen": 1055719424
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034350050150451354,
      "loss": 2.8687,
      "theoretical_loss": 3.6304815682367835,
      "tokens_seen": 1055784960
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003434904714142427,
      "loss": 3.0381,
      "theoretical_loss": 3.630460337317012,
      "tokens_seen": 1055850496
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034348044132397195,
      "loss": 3.1209,
      "theoretical_loss": 3.6304391080839453,
      "tokens_seen": 1055916032
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003434704112337011,
      "loss": 2.8258,
      "theoretical_loss": 3.630417880537344,
      "tokens_seen": 1055981568
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003434603811434303,
      "loss": 2.9064,
      "theoretical_loss": 3.6303966546769706,
      "tokens_seen": 1056047104
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003434503510531595,
      "loss": 2.5539,
      "theoretical_loss": 3.6303754305025855,
      "tokens_seen": 1056112640
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003434403209628887,
      "loss": 2.9488,
      "theoretical_loss": 3.630354208013951,
      "tokens_seen": 1056178176
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034343029087261786,
      "loss": 3.0587,
      "theoretical_loss": 3.630332987210828,
      "tokens_seen": 1056243712
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034342026078234704,
      "loss": 2.8866,
      "theoretical_loss": 3.6303117680929784,
      "tokens_seen": 1056309248
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003434102306920762,
      "loss": 3.0654,
      "theoretical_loss": 3.6302905506601633,
      "tokens_seen": 1056374784
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034340020060180546,
      "loss": 3.0161,
      "theoretical_loss": 3.6302693349121453,
      "tokens_seen": 1056440320
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003433901705115346,
      "loss": 2.9731,
      "theoretical_loss": 3.630248120848685,
      "tokens_seen": 1056505856
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003433801404212638,
      "loss": 2.8097,
      "theoretical_loss": 3.6302269084695453,
      "tokens_seen": 1056571392
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034337011033099295,
      "loss": 3.015,
      "theoretical_loss": 3.630205697774487,
      "tokens_seen": 1056636928
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003433600802407222,
      "loss": 3.0405,
      "theoretical_loss": 3.630184488763273,
      "tokens_seen": 1056702464
    },
    {
      "epoch": 0.32,
      "objective/train/advantage_avg": 0.49405616521835327,
      "objective/train/docs_used": 603451,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.912529945373535,
      "objective/train/original_loss": 2.912529230117798,
      "objective/train/theoretical_loss": 3.6301632814356637,
      "objective/train/tokens_used": 1077228000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24419067800045013,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506469011306763,
      "objective/train/weighted_lm_loss": 3.0600781440734863,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0338950157165527,
      "theoretical_loss": 3.6301632814356637,
      "tokens_seen": 1056768000
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034335005015045136,
      "loss": 2.9368,
      "theoretical_loss": 3.6301632814356637,
      "tokens_seen": 1056768000
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034334002006018054,
      "loss": 2.84,
      "theoretical_loss": 3.6301420757914227,
      "tokens_seen": 1056833536
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003433299899699097,
      "loss": 3.1047,
      "theoretical_loss": 3.6301208718303113,
      "tokens_seen": 1056899072
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003433199598796389,
      "loss": 3.0371,
      "theoretical_loss": 3.630099669552091,
      "tokens_seen": 1056964608
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003433099297893681,
      "loss": 2.844,
      "theoretical_loss": 3.630078468956525,
      "tokens_seen": 1057030144
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003432998996990973,
      "loss": 3.0662,
      "theoretical_loss": 3.630057270043374,
      "tokens_seen": 1057095680
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034328986960882645,
      "loss": 3.0318,
      "theoretical_loss": 3.630036072812402,
      "tokens_seen": 1057161216
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003432798395185557,
      "loss": 2.9778,
      "theoretical_loss": 3.63001487726337,
      "tokens_seen": 1057226752
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034326980942828487,
      "loss": 2.7508,
      "theoretical_loss": 3.629993683396041,
      "tokens_seen": 1057292288
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034325977933801405,
      "loss": 2.9577,
      "theoretical_loss": 3.629972491210176,
      "tokens_seen": 1057357824
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034324974924774323,
      "loss": 3.0841,
      "theoretical_loss": 3.629951300705539,
      "tokens_seen": 1057423360
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003432397191574724,
      "loss": 2.8267,
      "theoretical_loss": 3.6299301118818916,
      "tokens_seen": 1057488896
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003432296890672016,
      "loss": 3.154,
      "theoretical_loss": 3.629908924738997,
      "tokens_seen": 1057554432
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003432196589769308,
      "loss": 3.026,
      "theoretical_loss": 3.629887739276616,
      "tokens_seen": 1057619968
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034320962888665995,
      "loss": 2.8828,
      "theoretical_loss": 3.6298665554945133,
      "tokens_seen": 1057685504
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003431995987963892,
      "loss": 3.3478,
      "theoretical_loss": 3.6298453733924507,
      "tokens_seen": 1057751040
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003431895687061183,
      "loss": 3.0931,
      "theoretical_loss": 3.62982419297019,
      "tokens_seen": 1057816576
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034317953861584755,
      "loss": 3.016,
      "theoretical_loss": 3.6298030142274955,
      "tokens_seen": 1057882112
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034316950852557673,
      "loss": 3.0148,
      "theoretical_loss": 3.6297818371641286,
      "tokens_seen": 1057947648
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003431594784353059,
      "loss": 3.0158,
      "theoretical_loss": 3.6297606617798532,
      "tokens_seen": 1058013184
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003431494483450351,
      "loss": 2.7176,
      "theoretical_loss": 3.629739488074432,
      "tokens_seen": 1058078720
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034313941825476433,
      "loss": 3.0355,
      "theoretical_loss": 3.629718316047627,
      "tokens_seen": 1058144256
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034312938816449345,
      "loss": 2.8858,
      "theoretical_loss": 3.629697145699202,
      "tokens_seen": 1058209792
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003431193580742227,
      "loss": 2.9424,
      "theoretical_loss": 3.6296759770289198,
      "tokens_seen": 1058275328
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003431093279839518,
      "loss": 3.1675,
      "theoretical_loss": 3.6296548100365436,
      "tokens_seen": 1058340864
    },
    {
      "epoch": 0.32,
      "objective/train/advantage_avg": 0.4953053891658783,
      "objective/train/docs_used": 604503,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.9056406021118164,
      "objective/train/original_loss": 2.9056413173675537,
      "objective/train/theoretical_loss": 3.629633644721836,
      "objective/train/tokens_used": 1078866400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24540261924266815,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507781505584717,
      "objective/train/weighted_lm_loss": 3.053327798843384,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0363596677780151,
      "theoretical_loss": 3.629633644721836,
      "tokens_seen": 1058406400
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034309929789368105,
      "loss": 2.7971,
      "theoretical_loss": 3.629633644721836,
      "tokens_seen": 1058406400
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003430892678034103,
      "loss": 2.8621,
      "theoretical_loss": 3.6296124810845614,
      "tokens_seen": 1058471936
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003430792377131394,
      "loss": 2.9807,
      "theoretical_loss": 3.629591319124481,
      "tokens_seen": 1058537472
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034306920762286865,
      "loss": 2.9591,
      "theoretical_loss": 3.6295701588413603,
      "tokens_seen": 1058603008
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003430591775325978,
      "loss": 2.9267,
      "theoretical_loss": 3.629549000234961,
      "tokens_seen": 1058668544
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.000343049147442327,
      "loss": 2.9273,
      "theoretical_loss": 3.6295278433050466,
      "tokens_seen": 1058734080
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003430391173520562,
      "loss": 3.1562,
      "theoretical_loss": 3.6295066880513813,
      "tokens_seen": 1058799616
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003430290872617854,
      "loss": 2.8398,
      "theoretical_loss": 3.6294855344737287,
      "tokens_seen": 1058865152
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034301905717151456,
      "loss": 2.8119,
      "theoretical_loss": 3.629464382571851,
      "tokens_seen": 1058930688
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034300902708124374,
      "loss": 3.0054,
      "theoretical_loss": 3.629443232345513,
      "tokens_seen": 1058996224
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003429989969909729,
      "loss": 2.8938,
      "theoretical_loss": 3.629422083794477,
      "tokens_seen": 1059061760
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034298896690070215,
      "loss": 2.7499,
      "theoretical_loss": 3.629400936918508,
      "tokens_seen": 1059127296
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003429789368104313,
      "loss": 2.854,
      "theoretical_loss": 3.629379791717369,
      "tokens_seen": 1059192832
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003429689067201605,
      "loss": 3.0433,
      "theoretical_loss": 3.6293586481908235,
      "tokens_seen": 1059258368
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003429588766298897,
      "loss": 2.8,
      "theoretical_loss": 3.6293375063386364,
      "tokens_seen": 1059323904
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003429488465396189,
      "loss": 3.2319,
      "theoretical_loss": 3.62931636616057,
      "tokens_seen": 1059389440
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034293881644934806,
      "loss": 2.9125,
      "theoretical_loss": 3.6292952276563897,
      "tokens_seen": 1059454976
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034292878635907724,
      "loss": 3.0186,
      "theoretical_loss": 3.6292740908258576,
      "tokens_seen": 1059520512
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003429187562688064,
      "loss": 3.1668,
      "theoretical_loss": 3.62925295566874,
      "tokens_seen": 1059586048
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034290872617853566,
      "loss": 2.998,
      "theoretical_loss": 3.6292318221847992,
      "tokens_seen": 1059651584
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003428986960882648,
      "loss": 2.9648,
      "theoretical_loss": 3.6292106903737995,
      "tokens_seen": 1059717120
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.000342888665997994,
      "loss": 3.046,
      "theoretical_loss": 3.6291895602355053,
      "tokens_seen": 1059782656
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034287863590772315,
      "loss": 3.1079,
      "theoretical_loss": 3.629168431769681,
      "tokens_seen": 1059848192
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003428686058174524,
      "loss": 3.0583,
      "theoretical_loss": 3.6291473049760903,
      "tokens_seen": 1059913728
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034285857572718156,
      "loss": 3.0598,
      "theoretical_loss": 3.629126179854498,
      "tokens_seen": 1059979264
    },
    {
      "epoch": 0.32,
      "objective/train/advantage_avg": 0.4951869249343872,
      "objective/train/docs_used": 605021,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.1251938343048096,
      "objective/train/original_loss": 3.1251940727233887,
      "objective/train/theoretical_loss": 3.6291050564046676,
      "objective/train/tokens_used": 1080504800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2452603429555893,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507655143737793,
      "objective/train/weighted_lm_loss": 3.2837257385253906,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 3.6291050564046676,
      "tokens_seen": 1060044800
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034284854563691074,
      "loss": 3.0065,
      "theoretical_loss": 3.6291050564046676,
      "tokens_seen": 1060044800
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003428385155466399,
      "loss": 2.8664,
      "theoretical_loss": 3.6290839346263644,
      "tokens_seen": 1060110336
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003428284854563691,
      "loss": 3.0111,
      "theoretical_loss": 3.6290628145193526,
      "tokens_seen": 1060175872
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003428184553660983,
      "loss": 2.9575,
      "theoretical_loss": 3.629041696083396,
      "tokens_seen": 1060241408
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003428084252758275,
      "loss": 3.118,
      "theoretical_loss": 3.62902057931826,
      "tokens_seen": 1060306944
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034279839518555665,
      "loss": 2.8248,
      "theoretical_loss": 3.6289994642237087,
      "tokens_seen": 1060372480
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003427883650952859,
      "loss": 2.9683,
      "theoretical_loss": 3.6289783507995064,
      "tokens_seen": 1060438016
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034277833500501507,
      "loss": 2.765,
      "theoretical_loss": 3.628957239045419,
      "tokens_seen": 1060503552
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034276830491474425,
      "loss": 3.012,
      "theoretical_loss": 3.628936128961209,
      "tokens_seen": 1060569088
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034275827482447343,
      "loss": 2.8357,
      "theoretical_loss": 3.6289150205466427,
      "tokens_seen": 1060634624
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003427482447342026,
      "loss": 2.9279,
      "theoretical_loss": 3.628893913801485,
      "tokens_seen": 1060700160
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003427382146439318,
      "loss": 2.7463,
      "theoretical_loss": 3.6288728087255,
      "tokens_seen": 1060765696
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.000342728184553661,
      "loss": 3.1599,
      "theoretical_loss": 3.6288517053184535,
      "tokens_seen": 1060831232
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034271815446339015,
      "loss": 3.038,
      "theoretical_loss": 3.6288306035801092,
      "tokens_seen": 1060896768
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003427081243731194,
      "loss": 3.0947,
      "theoretical_loss": 3.628809503510233,
      "tokens_seen": 1060962304
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003426980942828485,
      "loss": 3.0426,
      "theoretical_loss": 3.6287884051085895,
      "tokens_seen": 1061027840
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034268806419257775,
      "loss": 3.1547,
      "theoretical_loss": 3.628767308374944,
      "tokens_seen": 1061093376
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034267803410230693,
      "loss": 2.9639,
      "theoretical_loss": 3.6287462133090616,
      "tokens_seen": 1061158912
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003426680040120361,
      "loss": 3.1537,
      "theoretical_loss": 3.6287251199107073,
      "tokens_seen": 1061224448
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003426579739217653,
      "loss": 2.7744,
      "theoretical_loss": 3.628704028179647,
      "tokens_seen": 1061289984
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034264794383149453,
      "loss": 2.9944,
      "theoretical_loss": 3.6286829381156447,
      "tokens_seen": 1061355520
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034263791374122366,
      "loss": 3.1097,
      "theoretical_loss": 3.628661849718467,
      "tokens_seen": 1061421056
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003426278836509529,
      "loss": 2.8733,
      "theoretical_loss": 3.628640762987878,
      "tokens_seen": 1061486592
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.000342617853560682,
      "loss": 2.6865,
      "theoretical_loss": 3.6286196779236444,
      "tokens_seen": 1061552128
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034260782347041125,
      "loss": 2.8947,
      "theoretical_loss": 3.628598594525531,
      "tokens_seen": 1061617664
    },
    {
      "epoch": 0.32,
      "objective/train/advantage_avg": 0.4955374300479889,
      "objective/train/docs_used": 606310,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.1696417331695557,
      "objective/train/original_loss": 3.1696414947509766,
      "objective/train/theoretical_loss": 3.628577512793303,
      "objective/train/tokens_used": 1082143200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24562573432922363,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0508023500442505,
      "objective/train/weighted_lm_loss": 3.330638885498047,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0363596677780151,
      "theoretical_loss": 3.628577512793303,
      "tokens_seen": 1061683200
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034259779338014043,
      "loss": 2.9821,
      "theoretical_loss": 3.628577512793303,
      "tokens_seen": 1061683200
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003425877632898696,
      "loss": 3.1363,
      "theoretical_loss": 3.628556432726727,
      "tokens_seen": 1061748736
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003425777331995988,
      "loss": 3.007,
      "theoretical_loss": 3.6285353543255674,
      "tokens_seen": 1061814272
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.000342567703109328,
      "loss": 2.8054,
      "theoretical_loss": 3.6285142775895904,
      "tokens_seen": 1061879808
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034255767301905716,
      "loss": 2.7311,
      "theoretical_loss": 3.6284932025185617,
      "tokens_seen": 1061945344
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003425476429287864,
      "loss": 3.1688,
      "theoretical_loss": 3.6284721291122466,
      "tokens_seen": 1062010880
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003425376128385155,
      "loss": 3.2151,
      "theoretical_loss": 3.6284510573704116,
      "tokens_seen": 1062076416
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034252758274824476,
      "loss": 2.8116,
      "theoretical_loss": 3.6284299872928223,
      "tokens_seen": 1062141952
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003425175526579739,
      "loss": 3.0275,
      "theoretical_loss": 3.6284089188792445,
      "tokens_seen": 1062207488
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003425075225677031,
      "loss": 2.8667,
      "theoretical_loss": 3.6283878521294444,
      "tokens_seen": 1062273024
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003424974924774323,
      "loss": 3.0243,
      "theoretical_loss": 3.6283667870431873,
      "tokens_seen": 1062338560
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003424874623871615,
      "loss": 3.0942,
      "theoretical_loss": 3.6283457236202397,
      "tokens_seen": 1062404096
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034247743229689066,
      "loss": 3.079,
      "theoretical_loss": 3.6283246618603675,
      "tokens_seen": 1062469632
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003424674022066199,
      "loss": 2.981,
      "theoretical_loss": 3.628303601763337,
      "tokens_seen": 1062535168
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.000342457372116349,
      "loss": 2.9482,
      "theoretical_loss": 3.6282825433289148,
      "tokens_seen": 1062600704
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034244734202607826,
      "loss": 3.023,
      "theoretical_loss": 3.628261486556866,
      "tokens_seen": 1062666240
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003424373119358074,
      "loss": 2.9817,
      "theoretical_loss": 3.6282404314469576,
      "tokens_seen": 1062731776
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003424272818455366,
      "loss": 2.978,
      "theoretical_loss": 3.6282193779989558,
      "tokens_seen": 1062797312
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003424172517552658,
      "loss": 2.9282,
      "theoretical_loss": 3.628198326212627,
      "tokens_seen": 1062862848
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.000342407221664995,
      "loss": 2.9409,
      "theoretical_loss": 3.6281772760877375,
      "tokens_seen": 1062928384
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034239719157472416,
      "loss": 3.1481,
      "theoretical_loss": 3.6281562276240535,
      "tokens_seen": 1062993920
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034238716148445335,
      "loss": 3.1778,
      "theoretical_loss": 3.6281351808213422,
      "tokens_seen": 1063059456
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003423771313941825,
      "loss": 3.2105,
      "theoretical_loss": 3.6281141356793696,
      "tokens_seen": 1063124992
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034236710130391176,
      "loss": 3.1505,
      "theoretical_loss": 3.6280930921979024,
      "tokens_seen": 1063190528
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003423570712136409,
      "loss": 3.0591,
      "theoretical_loss": 3.6280720503767077,
      "tokens_seen": 1063256064
    },
    {
      "epoch": 0.32,
      "objective/train/advantage_avg": 0.49469804763793945,
      "objective/train/docs_used": 606853,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.8382372856140137,
      "objective/train/original_loss": 2.838237762451172,
      "objective/train/theoretical_loss": 3.628051010215551,
      "objective/train/tokens_used": 1083781600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2447831928730011,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507140159606934,
      "objective/train/weighted_lm_loss": 2.982318162918091,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0425469875335693,
      "theoretical_loss": 3.628051010215551,
      "tokens_seen": 1063321600
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003423470411233701,
      "loss": 3.0208,
      "theoretical_loss": 3.628051010215551,
      "tokens_seen": 1063321600
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003423370110330993,
      "loss": 2.8421,
      "theoretical_loss": 3.6280299717142004,
      "tokens_seen": 1063387136
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003423269809428285,
      "loss": 3.0476,
      "theoretical_loss": 3.6280089348724216,
      "tokens_seen": 1063452672
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003423169508525577,
      "loss": 2.7499,
      "theoretical_loss": 3.627987899689982,
      "tokens_seen": 1063518208
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034230692076228685,
      "loss": 2.9797,
      "theoretical_loss": 3.6279668661666484,
      "tokens_seen": 1063583744
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003422968906720161,
      "loss": 3.1256,
      "theoretical_loss": 3.6279458343021878,
      "tokens_seen": 1063649280
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034228686058174527,
      "loss": 2.8016,
      "theoretical_loss": 3.6279248040963674,
      "tokens_seen": 1063714816
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034227683049147445,
      "loss": 3.152,
      "theoretical_loss": 3.627903775548953,
      "tokens_seen": 1063780352
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034226680040120363,
      "loss": 2.9093,
      "theoretical_loss": 3.6278827486597134,
      "tokens_seen": 1063845888
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003422567703109328,
      "loss": 3.1958,
      "theoretical_loss": 3.6278617234284143,
      "tokens_seen": 1063911424
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.000342246740220662,
      "loss": 3.0184,
      "theoretical_loss": 3.627840699854824,
      "tokens_seen": 1063976960
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003422367101303912,
      "loss": 2.9278,
      "theoretical_loss": 3.6278196779387084,
      "tokens_seen": 1064042496
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034222668004012035,
      "loss": 2.8868,
      "theoretical_loss": 3.627798657679836,
      "tokens_seen": 1064108032
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003422166499498496,
      "loss": 3.045,
      "theoretical_loss": 3.6277776390779732,
      "tokens_seen": 1064173568
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003422066198595787,
      "loss": 2.8644,
      "theoretical_loss": 3.627756622132888,
      "tokens_seen": 1064239104
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034219658976930795,
      "loss": 3.0912,
      "theoretical_loss": 3.627735606844347,
      "tokens_seen": 1064304640
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034218655967903713,
      "loss": 3.0714,
      "theoretical_loss": 3.6277145932121186,
      "tokens_seen": 1064370176
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003421765295887663,
      "loss": 3.1393,
      "theoretical_loss": 3.6276935812359703,
      "tokens_seen": 1064435712
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003421664994984955,
      "loss": 3.1099,
      "theoretical_loss": 3.627672570915668,
      "tokens_seen": 1064501248
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034215646940822473,
      "loss": 3.0217,
      "theoretical_loss": 3.6276515622509815,
      "tokens_seen": 1064566784
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034214643931795386,
      "loss": 3.0134,
      "theoretical_loss": 3.627630555241677,
      "tokens_seen": 1064632320
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003421364092276831,
      "loss": 2.8741,
      "theoretical_loss": 3.6276095498875227,
      "tokens_seen": 1064697856
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003421263791374122,
      "loss": 3.1289,
      "theoretical_loss": 3.627588546188286,
      "tokens_seen": 1064763392
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034211634904714145,
      "loss": 3.0369,
      "theoretical_loss": 3.6275675441437345,
      "tokens_seen": 1064828928
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034210631895687063,
      "loss": 2.9125,
      "theoretical_loss": 3.627546543753637,
      "tokens_seen": 1064894464
    },
    {
      "epoch": 0.32,
      "objective/train/advantage_avg": 0.494121253490448,
      "objective/train/docs_used": 607789,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.776517629623413,
      "objective/train/original_loss": 2.776517629623413,
      "objective/train/theoretical_loss": 3.62752554501776,
      "objective/train/tokens_used": 1085420000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24422381818294525,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506535768508911,
      "objective/train/weighted_lm_loss": 2.917219638824463,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0338950157165527,
      "theoretical_loss": 3.62752554501776,
      "tokens_seen": 1064960000
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003420962888665998,
      "loss": 2.9987,
      "theoretical_loss": 3.62752554501776,
      "tokens_seen": 1064960000
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.000342086258776329,
      "loss": 2.9279,
      "theoretical_loss": 3.6275045479358727,
      "tokens_seen": 1065025536
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003420762286860582,
      "loss": 2.9968,
      "theoretical_loss": 3.6274835525077425,
      "tokens_seen": 1065091072
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034206619859578736,
      "loss": 3.0795,
      "theoretical_loss": 3.6274625587331375,
      "tokens_seen": 1065156608
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003420561685055166,
      "loss": 2.8971,
      "theoretical_loss": 3.627441566611825,
      "tokens_seen": 1065222144
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003420461384152457,
      "loss": 3.1102,
      "theoretical_loss": 3.6274205761435745,
      "tokens_seen": 1065287680
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034203610832497496,
      "loss": 3.1657,
      "theoretical_loss": 3.627399587328153,
      "tokens_seen": 1065353216
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003420260782347041,
      "loss": 2.8603,
      "theoretical_loss": 3.6273786001653288,
      "tokens_seen": 1065418752
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003420160481444333,
      "loss": 2.9214,
      "theoretical_loss": 3.627357614654871,
      "tokens_seen": 1065484288
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003420060180541625,
      "loss": 3.013,
      "theoretical_loss": 3.6273366307965467,
      "tokens_seen": 1065549824
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003419959879638917,
      "loss": 2.6022,
      "theoretical_loss": 3.627315648590125,
      "tokens_seen": 1065615360
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034198595787362086,
      "loss": 2.885,
      "theoretical_loss": 3.627294668035374,
      "tokens_seen": 1065680896
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003419759277833501,
      "loss": 3.1409,
      "theoretical_loss": 3.6272736891320627,
      "tokens_seen": 1065746432
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003419658976930792,
      "loss": 2.7925,
      "theoretical_loss": 3.627252711879959,
      "tokens_seen": 1065811968
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034195586760280846,
      "loss": 2.9345,
      "theoretical_loss": 3.627231736278831,
      "tokens_seen": 1065877504
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003419458375125376,
      "loss": 3.025,
      "theoretical_loss": 3.627210762328448,
      "tokens_seen": 1065943040
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003419358074222668,
      "loss": 3.0592,
      "theoretical_loss": 3.6271897900285786,
      "tokens_seen": 1066008576
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.000341925777331996,
      "loss": 3.0566,
      "theoretical_loss": 3.6271688193789915,
      "tokens_seen": 1066074112
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003419157472417252,
      "loss": 3.2165,
      "theoretical_loss": 3.6271478503794548,
      "tokens_seen": 1066139648
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034190571715145436,
      "loss": 2.9741,
      "theoretical_loss": 3.627126883029737,
      "tokens_seen": 1066205184
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034189568706118355,
      "loss": 2.8348,
      "theoretical_loss": 3.627105917329608,
      "tokens_seen": 1066270720
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003418856569709127,
      "loss": 2.9943,
      "theoretical_loss": 3.6270849532788363,
      "tokens_seen": 1066336256
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034187562688064196,
      "loss": 3.1285,
      "theoretical_loss": 3.6270639908771907,
      "tokens_seen": 1066401792
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003418655967903711,
      "loss": 2.8222,
      "theoretical_loss": 3.62704303012444,
      "tokens_seen": 1066467328
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003418555667001003,
      "loss": 3.0691,
      "theoretical_loss": 3.627022071020353,
      "tokens_seen": 1066532864
    },
    {
      "epoch": 0.32,
      "objective/train/advantage_avg": 0.49631085991859436,
      "objective/train/docs_used": 608475,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.030895948410034,
      "objective/train/original_loss": 3.030895233154297,
      "objective/train/theoretical_loss": 3.627001113564699,
      "objective/train/tokens_used": 1087058400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2463710606098175,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0508835315704346,
      "objective/train/weighted_lm_loss": 3.185084581375122,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0432155132293701,
      "theoretical_loss": 3.627001113564699,
      "tokens_seen": 1066598400
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034184553660982945,
      "loss": 3.0849,
      "theoretical_loss": 3.627001113564699,
      "tokens_seen": 1066598400
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003418355065195587,
      "loss": 3.0507,
      "theoretical_loss": 3.626980157757247,
      "tokens_seen": 1066663936
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034182547642928787,
      "loss": 2.7927,
      "theoretical_loss": 3.6269592035977665,
      "tokens_seen": 1066729472
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034181544633901705,
      "loss": 2.8931,
      "theoretical_loss": 3.6269382510860257,
      "tokens_seen": 1066795008
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034180541624874623,
      "loss": 2.9712,
      "theoretical_loss": 3.6269173002217956,
      "tokens_seen": 1066860544
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034179538615847547,
      "loss": 2.96,
      "theoretical_loss": 3.6268963510048433,
      "tokens_seen": 1066926080
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003417853560682046,
      "loss": 2.9279,
      "theoretical_loss": 3.6268754034349397,
      "tokens_seen": 1066991616
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034177532597793383,
      "loss": 2.996,
      "theoretical_loss": 3.626854457511853,
      "tokens_seen": 1067057152
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034176529588766295,
      "loss": 3.0797,
      "theoretical_loss": 3.6268335132353533,
      "tokens_seen": 1067122688
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003417552657973922,
      "loss": 3.0026,
      "theoretical_loss": 3.6268125706052103,
      "tokens_seen": 1067188224
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034174523570712137,
      "loss": 2.9507,
      "theoretical_loss": 3.6267916296211933,
      "tokens_seen": 1067253760
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034173520561685055,
      "loss": 2.8991,
      "theoretical_loss": 3.626770690283071,
      "tokens_seen": 1067319296
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034172517552657973,
      "loss": 3.0645,
      "theoretical_loss": 3.6267497525906145,
      "tokens_seen": 1067384832
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003417151454363089,
      "loss": 3.0831,
      "theoretical_loss": 3.6267288165435922,
      "tokens_seen": 1067450368
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003417051153460381,
      "loss": 2.7327,
      "theoretical_loss": 3.626707882141774,
      "tokens_seen": 1067515904
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034169508525576733,
      "loss": 2.9423,
      "theoretical_loss": 3.62668694938493,
      "tokens_seen": 1067581440
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034168505516549646,
      "loss": 3.0286,
      "theoretical_loss": 3.6266660182728296,
      "tokens_seen": 1067646976
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003416750250752257,
      "loss": 2.8116,
      "theoretical_loss": 3.626645088805243,
      "tokens_seen": 1067712512
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003416649949849548,
      "loss": 3.0254,
      "theoretical_loss": 3.62662416098194,
      "tokens_seen": 1067778048
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034165496489468406,
      "loss": 3.0009,
      "theoretical_loss": 3.62660323480269,
      "tokens_seen": 1067843584
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034164493480441324,
      "loss": 2.8949,
      "theoretical_loss": 3.6265823102672634,
      "tokens_seen": 1067909120
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003416349047141424,
      "loss": 2.8159,
      "theoretical_loss": 3.62656138737543,
      "tokens_seen": 1067974656
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003416248746238716,
      "loss": 2.9519,
      "theoretical_loss": 3.6265404661269605,
      "tokens_seen": 1068040192
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034161484453360083,
      "loss": 2.9264,
      "theoretical_loss": 3.6265195465216244,
      "tokens_seen": 1068105728
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034160481444332996,
      "loss": 3.0323,
      "theoretical_loss": 3.6264986285591916,
      "tokens_seen": 1068171264
    },
    {
      "epoch": 0.32,
      "objective/train/advantage_avg": 0.49358248710632324,
      "objective/train/docs_used": 609657,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.2417101860046387,
      "objective/train/original_loss": 3.2417094707489014,
      "objective/train/theoretical_loss": 3.6264777122394327,
      "objective/train/tokens_used": 1088696800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24372783303260803,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050597071647644,
      "objective/train/weighted_lm_loss": 3.405820369720459,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 3.6264777122394327,
      "tokens_seen": 1068236800
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003415947843530592,
      "loss": 3.3864,
      "theoretical_loss": 3.6264777122394327,
      "tokens_seen": 1068236800
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003415847542627884,
      "loss": 3.029,
      "theoretical_loss": 3.626456797562118,
      "tokens_seen": 1068302336
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034157472417251756,
      "loss": 2.8052,
      "theoretical_loss": 3.6264358845270177,
      "tokens_seen": 1068367872
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003415646940822468,
      "loss": 3.073,
      "theoretical_loss": 3.626414973133902,
      "tokens_seen": 1068433408
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003415546639919759,
      "loss": 2.909,
      "theoretical_loss": 3.626394063382541,
      "tokens_seen": 1068498944
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034154463390170516,
      "loss": 2.7652,
      "theoretical_loss": 3.626373155272706,
      "tokens_seen": 1068564480
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003415346038114343,
      "loss": 2.9936,
      "theoretical_loss": 3.6263522488041664,
      "tokens_seen": 1068630016
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003415245737211635,
      "loss": 3.1536,
      "theoretical_loss": 3.6263313439766938,
      "tokens_seen": 1068695552
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003415145436308927,
      "loss": 3.0629,
      "theoretical_loss": 3.626310440790058,
      "tokens_seen": 1068761088
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003415045135406219,
      "loss": 3.1306,
      "theoretical_loss": 3.6262895392440297,
      "tokens_seen": 1068826624
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034149448345035106,
      "loss": 3.2031,
      "theoretical_loss": 3.626268639338379,
      "tokens_seen": 1068892160
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003414844533600803,
      "loss": 2.8992,
      "theoretical_loss": 3.6262477410728784,
      "tokens_seen": 1068957696
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003414744232698094,
      "loss": 3.1212,
      "theoretical_loss": 3.626226844447297,
      "tokens_seen": 1069023232
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034146439317953866,
      "loss": 3.3053,
      "theoretical_loss": 3.626205949461406,
      "tokens_seen": 1069088768
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003414543630892678,
      "loss": 3.0082,
      "theoretical_loss": 3.6261850561149767,
      "tokens_seen": 1069154304
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.000341444332998997,
      "loss": 3.1731,
      "theoretical_loss": 3.626164164407779,
      "tokens_seen": 1069219840
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003414343029087262,
      "loss": 2.9627,
      "theoretical_loss": 3.626143274339585,
      "tokens_seen": 1069285376
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003414242728184554,
      "loss": 3.1332,
      "theoretical_loss": 3.626122385910165,
      "tokens_seen": 1069350912
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034141424272818456,
      "loss": 2.7539,
      "theoretical_loss": 3.62610149911929,
      "tokens_seen": 1069416448
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034140421263791375,
      "loss": 3.0759,
      "theoretical_loss": 3.6260806139667316,
      "tokens_seen": 1069481984
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034139418254764293,
      "loss": 3.109,
      "theoretical_loss": 3.62605973045226,
      "tokens_seen": 1069547520
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034138415245737216,
      "loss": 2.8715,
      "theoretical_loss": 3.6260388485756474,
      "tokens_seen": 1069613056
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003413741223671013,
      "loss": 3.0606,
      "theoretical_loss": 3.626017968336664,
      "tokens_seen": 1069678592
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003413640922768305,
      "loss": 2.8465,
      "theoretical_loss": 3.6259970897350815,
      "tokens_seen": 1069744128
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034135406218655965,
      "loss": 2.9986,
      "theoretical_loss": 3.625976212770672,
      "tokens_seen": 1069809664
    },
    {
      "epoch": 0.32,
      "objective/train/advantage_avg": 0.4955526292324066,
      "objective/train/docs_used": 610408,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.51613187789917,
      "objective/train/original_loss": 2.51613187789917,
      "objective/train/theoretical_loss": 3.625955337443205,
      "objective/train/tokens_used": 1090335200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2456391155719757,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0508038997650146,
      "objective/train/weighted_lm_loss": 2.643848180770874,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0363596677780151,
      "theoretical_loss": 3.625955337443205,
      "tokens_seen": 1069875200
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003413440320962889,
      "loss": 3.0016,
      "theoretical_loss": 3.625955337443205,
      "tokens_seen": 1069875200
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034133400200601807,
      "loss": 2.8859,
      "theoretical_loss": 3.6259344637524533,
      "tokens_seen": 1069940736
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034132397191574725,
      "loss": 3.0551,
      "theoretical_loss": 3.6259135916981884,
      "tokens_seen": 1070006272
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034131394182547643,
      "loss": 3.0874,
      "theoretical_loss": 3.6258927212801813,
      "tokens_seen": 1070071808
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034130391173520567,
      "loss": 3.0883,
      "theoretical_loss": 3.625871852498203,
      "tokens_seen": 1070137344
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003412938816449348,
      "loss": 3.047,
      "theoretical_loss": 3.6258509853520264,
      "tokens_seen": 1070202880
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034128385155466403,
      "loss": 3.0268,
      "theoretical_loss": 3.625830119841422,
      "tokens_seen": 1070268416
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034127382146439315,
      "loss": 2.9233,
      "theoretical_loss": 3.625809255966162,
      "tokens_seen": 1070333952
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003412637913741224,
      "loss": 2.7773,
      "theoretical_loss": 3.6257883937260176,
      "tokens_seen": 1070399488
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034125376128385157,
      "loss": 3.0247,
      "theoretical_loss": 3.6257675331207615,
      "tokens_seen": 1070465024
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034124373119358075,
      "loss": 3.0004,
      "theoretical_loss": 3.6257466741501645,
      "tokens_seen": 1070530560
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034123370110330993,
      "loss": 2.7287,
      "theoretical_loss": 3.6257258168139987,
      "tokens_seen": 1070596096
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003412236710130391,
      "loss": 2.8795,
      "theoretical_loss": 3.625704961112037,
      "tokens_seen": 1070661632
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003412136409227683,
      "loss": 2.9707,
      "theoretical_loss": 3.6256841070440498,
      "tokens_seen": 1070727168
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034120361083249753,
      "loss": 2.8483,
      "theoretical_loss": 3.62566325460981,
      "tokens_seen": 1070792704
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034119358074222666,
      "loss": 2.9649,
      "theoretical_loss": 3.6256424038090893,
      "tokens_seen": 1070858240
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003411835506519559,
      "loss": 3.1571,
      "theoretical_loss": 3.6256215546416595,
      "tokens_seen": 1070923776
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.000341173520561685,
      "loss": 2.982,
      "theoretical_loss": 3.625600707107294,
      "tokens_seen": 1070989312
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034116349047141426,
      "loss": 3.048,
      "theoretical_loss": 3.6255798612057633,
      "tokens_seen": 1071054848
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034115346038114344,
      "loss": 3.027,
      "theoretical_loss": 3.6255590169368404,
      "tokens_seen": 1071120384
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003411434302908726,
      "loss": 3.118,
      "theoretical_loss": 3.6255381743002983,
      "tokens_seen": 1071185920
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003411334002006018,
      "loss": 3.0412,
      "theoretical_loss": 3.6255173332959076,
      "tokens_seen": 1071251456
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034112337011033103,
      "loss": 2.8441,
      "theoretical_loss": 3.6254964939234426,
      "tokens_seen": 1071316992
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034111334002006016,
      "loss": 3.0852,
      "theoretical_loss": 3.625475656182674,
      "tokens_seen": 1071382528
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003411033099297894,
      "loss": 2.8981,
      "theoretical_loss": 3.625454820073375,
      "tokens_seen": 1071448064
    },
    {
      "epoch": 0.32,
      "objective/train/advantage_avg": 0.4949130117893219,
      "objective/train/docs_used": 611059,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.7532122135162354,
      "objective/train/original_loss": 2.7532126903533936,
      "objective/train/theoretical_loss": 3.6254339855953184,
      "objective/train/tokens_used": 1091973600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2450019270181656,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050736665725708,
      "objective/train/weighted_lm_loss": 2.8927154541015625,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0417574644088745,
      "theoretical_loss": 3.6254339855953184,
      "tokens_seen": 1071513600
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003410932798395185,
      "loss": 2.9432,
      "theoretical_loss": 3.6254339855953184,
      "tokens_seen": 1071513600
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034108324974924776,
      "loss": 2.8549,
      "theoretical_loss": 3.625413152748276,
      "tokens_seen": 1071579136
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034107321965897694,
      "loss": 2.925,
      "theoretical_loss": 3.625392321532021,
      "tokens_seen": 1071644672
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003410631895687061,
      "loss": 3.1705,
      "theoretical_loss": 3.6253714919463254,
      "tokens_seen": 1071710208
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003410531594784353,
      "loss": 2.9674,
      "theoretical_loss": 3.6253506639909627,
      "tokens_seen": 1071775744
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003410431293881645,
      "loss": 3.0921,
      "theoretical_loss": 3.6253298376657046,
      "tokens_seen": 1071841280
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034103309929789366,
      "loss": 3.0693,
      "theoretical_loss": 3.625309012970325,
      "tokens_seen": 1071906816
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003410230692076229,
      "loss": 3.0378,
      "theoretical_loss": 3.6252881899045954,
      "tokens_seen": 1071972352
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.000341013039117352,
      "loss": 2.7825,
      "theoretical_loss": 3.62526736846829,
      "tokens_seen": 1072037888
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034100300902708126,
      "loss": 3.147,
      "theoretical_loss": 3.625246548661181,
      "tokens_seen": 1072103424
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034099297893681044,
      "loss": 3.1544,
      "theoretical_loss": 3.625225730483041,
      "tokens_seen": 1072168960
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003409829488465396,
      "loss": 3.1724,
      "theoretical_loss": 3.625204913933644,
      "tokens_seen": 1072234496
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003409729187562688,
      "loss": 2.8688,
      "theoretical_loss": 3.625184099012762,
      "tokens_seen": 1072300032
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.000340962888665998,
      "loss": 3.0865,
      "theoretical_loss": 3.6251632857201685,
      "tokens_seen": 1072365568
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.00034095285857572717,
      "loss": 3.0,
      "theoretical_loss": 3.625142474055637,
      "tokens_seen": 1072431104
    },
    {
      "epoch": 0.32,
      "learning_rate": 0.0003409428284854564,
      "loss": 3.0207,
      "theoretical_loss": 3.62512166401894,
      "tokens_seen": 1072496640
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00034093279839518553,
      "loss": 3.1946,
      "theoretical_loss": 3.625100855609851,
      "tokens_seen": 1072562176
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00034092276830491477,
      "loss": 2.8989,
      "theoretical_loss": 3.6250800488281434,
      "tokens_seen": 1072627712
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003409127382146439,
      "loss": 2.9702,
      "theoretical_loss": 3.6250592436735904,
      "tokens_seen": 1072693248
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00034090270812437313,
      "loss": 3.1337,
      "theoretical_loss": 3.625038440145965,
      "tokens_seen": 1072758784
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003408926780341023,
      "loss": 3.0763,
      "theoretical_loss": 3.6250176382450414,
      "tokens_seen": 1072824320
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003408826479438315,
      "loss": 3.2599,
      "theoretical_loss": 3.624996837970593,
      "tokens_seen": 1072889856
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00034087261785356067,
      "loss": 2.9877,
      "theoretical_loss": 3.6249760393223918,
      "tokens_seen": 1072955392
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00034086258776328985,
      "loss": 3.132,
      "theoretical_loss": 3.624955242300213,
      "tokens_seen": 1073020928
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00034085255767301903,
      "loss": 3.0408,
      "theoretical_loss": 3.6249344469038296,
      "tokens_seen": 1073086464
    },
    {
      "epoch": 0.33,
      "objective/train/advantage_avg": 0.493761271238327,
      "objective/train/docs_used": 612309,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.07370924949646,
      "objective/train/original_loss": 3.07370924949646,
      "objective/train/theoretical_loss": 3.624913653133015,
      "objective/train/tokens_used": 1093612000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24387811124324799,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050615906715393,
      "objective/train/weighted_lm_loss": 3.229201555252075,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0396549701690674,
      "theoretical_loss": 3.624913653133015,
      "tokens_seen": 1073152000
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00034084252758274827,
      "loss": 3.1123,
      "theoretical_loss": 3.624913653133015,
      "tokens_seen": 1073152000
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00034083249749247745,
      "loss": 3.0821,
      "theoretical_loss": 3.6248928609875435,
      "tokens_seen": 1073217536
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00034082246740220663,
      "loss": 2.9583,
      "theoretical_loss": 3.6248720704671875,
      "tokens_seen": 1073283072
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00034081243731193587,
      "loss": 3.0355,
      "theoretical_loss": 3.624851281571723,
      "tokens_seen": 1073348608
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.000340802407221665,
      "loss": 2.8067,
      "theoretical_loss": 3.6248304943009213,
      "tokens_seen": 1073414144
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00034079237713139423,
      "loss": 3.1293,
      "theoretical_loss": 3.624809708654558,
      "tokens_seen": 1073479680
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00034078234704112335,
      "loss": 2.7418,
      "theoretical_loss": 3.624788924632406,
      "tokens_seen": 1073545216
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003407723169508526,
      "loss": 2.8864,
      "theoretical_loss": 3.62476814223424,
      "tokens_seen": 1073610752
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00034076228686058177,
      "loss": 3.0991,
      "theoretical_loss": 3.624747361459834,
      "tokens_seen": 1073676288
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00034075225677031095,
      "loss": 3.1487,
      "theoretical_loss": 3.624726582308961,
      "tokens_seen": 1073741824
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00034074222668004013,
      "loss": 2.9888,
      "theoretical_loss": 3.624705804781396,
      "tokens_seen": 1073807360
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003407321965897693,
      "loss": 3.2434,
      "theoretical_loss": 3.624685028876913,
      "tokens_seen": 1073872896
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003407221664994985,
      "loss": 2.9669,
      "theoretical_loss": 3.6246642545952854,
      "tokens_seen": 1073938432
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00034071213640922773,
      "loss": 2.9483,
      "theoretical_loss": 3.624643481936289,
      "tokens_seen": 1074003968
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00034070210631895686,
      "loss": 2.7618,
      "theoretical_loss": 3.6246227108996965,
      "tokens_seen": 1074069504
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003406920762286861,
      "loss": 3.1466,
      "theoretical_loss": 3.624601941485283,
      "tokens_seen": 1074135040
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003406820461384152,
      "loss": 3.0549,
      "theoretical_loss": 3.624581173692823,
      "tokens_seen": 1074200576
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00034067201604814446,
      "loss": 3.1599,
      "theoretical_loss": 3.6245604075220896,
      "tokens_seen": 1074266112
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00034066198595787364,
      "loss": 2.9468,
      "theoretical_loss": 3.6245396429728594,
      "tokens_seen": 1074331648
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003406519558676028,
      "loss": 2.8558,
      "theoretical_loss": 3.6245188800449046,
      "tokens_seen": 1074397184
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.000340641925777332,
      "loss": 2.9438,
      "theoretical_loss": 3.6244981187380017,
      "tokens_seen": 1074462720
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00034063189568706123,
      "loss": 2.9705,
      "theoretical_loss": 3.624477359051924,
      "tokens_seen": 1074528256
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00034062186559679036,
      "loss": 2.8746,
      "theoretical_loss": 3.6244566009864463,
      "tokens_seen": 1074593792
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003406118355065196,
      "loss": 2.884,
      "theoretical_loss": 3.6244358445413436,
      "tokens_seen": 1074659328
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003406018054162487,
      "loss": 2.9438,
      "theoretical_loss": 3.6244150897163907,
      "tokens_seen": 1074724864
    },
    {
      "epoch": 0.33,
      "objective/train/advantage_avg": 0.4950747787952423,
      "objective/train/docs_used": 613082,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.7145726680755615,
      "objective/train/original_loss": 2.7145724296569824,
      "objective/train/theoretical_loss": 3.624394336511362,
      "objective/train/tokens_used": 1095250400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2451566606760025,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507537126541138,
      "objective/train/weighted_lm_loss": 2.8523995876312256,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0408108234405518,
      "theoretical_loss": 3.624394336511362,
      "tokens_seen": 1074790400
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00034059177532597796,
      "loss": 2.9917,
      "theoretical_loss": 3.624394336511362,
      "tokens_seen": 1074790400
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00034058174523570714,
      "loss": 2.9381,
      "theoretical_loss": 3.6243735849260323,
      "tokens_seen": 1074855936
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003405717151454363,
      "loss": 2.7369,
      "theoretical_loss": 3.624352834960177,
      "tokens_seen": 1074921472
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003405616850551655,
      "loss": 2.9269,
      "theoretical_loss": 3.6243320866135704,
      "tokens_seen": 1074987008
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003405516549648947,
      "loss": 3.0669,
      "theoretical_loss": 3.6243113398859874,
      "tokens_seen": 1075052544
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00034054162487462386,
      "loss": 2.9412,
      "theoretical_loss": 3.6242905947772033,
      "tokens_seen": 1075118080
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003405315947843531,
      "loss": 3.0039,
      "theoretical_loss": 3.624269851286993,
      "tokens_seen": 1075183616
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003405215646940822,
      "loss": 2.8359,
      "theoretical_loss": 3.624249109415132,
      "tokens_seen": 1075249152
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00034051153460381146,
      "loss": 2.7399,
      "theoretical_loss": 3.6242283691613943,
      "tokens_seen": 1075314688
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00034050150451354064,
      "loss": 3.0222,
      "theoretical_loss": 3.624207630525556,
      "tokens_seen": 1075380224
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003404914744232698,
      "loss": 2.9833,
      "theoretical_loss": 3.6241868935073924,
      "tokens_seen": 1075445760
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.000340481444332999,
      "loss": 2.9516,
      "theoretical_loss": 3.6241661581066777,
      "tokens_seen": 1075511296
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003404714142427282,
      "loss": 2.6946,
      "theoretical_loss": 3.6241454243231885,
      "tokens_seen": 1075576832
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00034046138415245737,
      "loss": 3.0192,
      "theoretical_loss": 3.624124692156699,
      "tokens_seen": 1075642368
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003404513540621866,
      "loss": 2.8376,
      "theoretical_loss": 3.624103961606986,
      "tokens_seen": 1075707904
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00034044132397191573,
      "loss": 2.9627,
      "theoretical_loss": 3.624083232673823,
      "tokens_seen": 1075773440
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00034043129388164497,
      "loss": 2.7753,
      "theoretical_loss": 3.6240625053569873,
      "tokens_seen": 1075838976
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003404212637913741,
      "loss": 2.747,
      "theoretical_loss": 3.6240417796562534,
      "tokens_seen": 1075904512
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00034041123370110333,
      "loss": 2.9197,
      "theoretical_loss": 3.624021055571397,
      "tokens_seen": 1075970048
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003404012036108325,
      "loss": 2.9193,
      "theoretical_loss": 3.624000333102194,
      "tokens_seen": 1076035584
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003403911735205617,
      "loss": 3.042,
      "theoretical_loss": 3.62397961224842,
      "tokens_seen": 1076101120
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00034038114343029087,
      "loss": 2.724,
      "theoretical_loss": 3.62395889300985,
      "tokens_seen": 1076166656
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00034037111334002005,
      "loss": 2.8543,
      "theoretical_loss": 3.623938175386261,
      "tokens_seen": 1076232192
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00034036108324974923,
      "loss": 3.0139,
      "theoretical_loss": 3.623917459377428,
      "tokens_seen": 1076297728
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00034035105315947847,
      "loss": 2.9847,
      "theoretical_loss": 3.623896744983127,
      "tokens_seen": 1076363264
    },
    {
      "epoch": 0.33,
      "objective/train/advantage_avg": 0.495092511177063,
      "objective/train/docs_used": 614240,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.7287771701812744,
      "objective/train/original_loss": 2.7287769317626953,
      "objective/train/theoretical_loss": 3.6238760322031336,
      "objective/train/tokens_used": 1096888800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2451663464307785,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507556200027466,
      "objective/train/weighted_lm_loss": 2.8673255443573,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.044721245765686,
      "theoretical_loss": 3.6238760322031336,
      "tokens_seen": 1076428800
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003403410230692076,
      "loss": 3.067,
      "theoretical_loss": 3.6238760322031336,
      "tokens_seen": 1076428800
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00034033099297893683,
      "loss": 2.7648,
      "theoretical_loss": 3.6238553210372237,
      "tokens_seen": 1076494336
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.000340320962888666,
      "loss": 3.0262,
      "theoretical_loss": 3.6238346114851736,
      "tokens_seen": 1076559872
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003403109327983952,
      "loss": 3.2467,
      "theoretical_loss": 3.6238139035467594,
      "tokens_seen": 1076625408
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003403009027081244,
      "loss": 2.9473,
      "theoretical_loss": 3.6237931972217567,
      "tokens_seen": 1076690944
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00034029087261785356,
      "loss": 2.9617,
      "theoretical_loss": 3.6237724925099424,
      "tokens_seen": 1076756480
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00034028084252758274,
      "loss": 3.1593,
      "theoretical_loss": 3.6237517894110916,
      "tokens_seen": 1076822016
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00034027081243731197,
      "loss": 2.8942,
      "theoretical_loss": 3.6237310879249813,
      "tokens_seen": 1076887552
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003402607823470411,
      "loss": 2.8843,
      "theoretical_loss": 3.6237103880513875,
      "tokens_seen": 1076953088
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00034025075225677033,
      "loss": 2.828,
      "theoretical_loss": 3.623689689790086,
      "tokens_seen": 1077018624
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00034024072216649946,
      "loss": 3.0477,
      "theoretical_loss": 3.6236689931408543,
      "tokens_seen": 1077084160
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003402306920762287,
      "loss": 3.15,
      "theoretical_loss": 3.6236482981034674,
      "tokens_seen": 1077149696
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003402206619859579,
      "loss": 2.9432,
      "theoretical_loss": 3.6236276046777025,
      "tokens_seen": 1077215232
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00034021063189568706,
      "loss": 2.7773,
      "theoretical_loss": 3.6236069128633357,
      "tokens_seen": 1077280768
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00034020060180541624,
      "loss": 3.0698,
      "theoretical_loss": 3.623586222660144,
      "tokens_seen": 1077346304
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003401905717151454,
      "loss": 3.0913,
      "theoretical_loss": 3.6235655340679034,
      "tokens_seen": 1077411840
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003401805416248746,
      "loss": 3.0936,
      "theoretical_loss": 3.623544847086391,
      "tokens_seen": 1077477376
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00034017051153460384,
      "loss": 2.6507,
      "theoretical_loss": 3.623524161715383,
      "tokens_seen": 1077542912
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00034016048144433296,
      "loss": 2.8691,
      "theoretical_loss": 3.623503477954656,
      "tokens_seen": 1077608448
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003401504513540622,
      "loss": 3.1378,
      "theoretical_loss": 3.6234827958039872,
      "tokens_seen": 1077673984
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003401404212637914,
      "loss": 2.8147,
      "theoretical_loss": 3.6234621152631536,
      "tokens_seen": 1077739520
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00034013039117352056,
      "loss": 3.0676,
      "theoretical_loss": 3.623441436331931,
      "tokens_seen": 1077805056
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00034012036108324974,
      "loss": 3.0636,
      "theoretical_loss": 3.623420759010097,
      "tokens_seen": 1077870592
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003401103309929789,
      "loss": 2.7927,
      "theoretical_loss": 3.6234000832974282,
      "tokens_seen": 1077936128
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003401003009027081,
      "loss": 3.086,
      "theoretical_loss": 3.623379409193702,
      "tokens_seen": 1078001664
    },
    {
      "epoch": 0.33,
      "objective/train/advantage_avg": 0.49604180455207825,
      "objective/train/docs_used": 615029,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.7373600006103516,
      "objective/train/original_loss": 2.7373600006103516,
      "objective/train/theoretical_loss": 3.6233587366986946,
      "objective/train/tokens_used": 1098527200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2460985630750656,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050855278968811,
      "objective/train/weighted_lm_loss": 2.8765134811401367,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0437887907028198,
      "theoretical_loss": 3.6233587366986946,
      "tokens_seen": 1078067200
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00034009027081243734,
      "loss": 3.1716,
      "theoretical_loss": 3.6233587366986946,
      "tokens_seen": 1078067200
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003400802407221665,
      "loss": 2.8896,
      "theoretical_loss": 3.6233380658121837,
      "tokens_seen": 1078132736
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003400702106318957,
      "loss": 2.9817,
      "theoretical_loss": 3.6233173965339462,
      "tokens_seen": 1078198272
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003400601805416249,
      "loss": 3.038,
      "theoretical_loss": 3.6232967288637594,
      "tokens_seen": 1078263808
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00034005015045135406,
      "loss": 3.1537,
      "theoretical_loss": 3.6232760628014002,
      "tokens_seen": 1078329344
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003400401203610833,
      "loss": 3.1063,
      "theoretical_loss": 3.623255398346646,
      "tokens_seen": 1078394880
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003400300902708124,
      "loss": 2.7863,
      "theoretical_loss": 3.6232347354992736,
      "tokens_seen": 1078460416
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00034002006018054166,
      "loss": 3.076,
      "theoretical_loss": 3.6232140742590606,
      "tokens_seen": 1078525952
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00034001003009027084,
      "loss": 2.9942,
      "theoretical_loss": 3.623193414625785,
      "tokens_seen": 1078591488
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00034,
      "loss": 3.1917,
      "theoretical_loss": 3.6231727565992227,
      "tokens_seen": 1078657024
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003399899699097292,
      "loss": 2.8686,
      "theoretical_loss": 3.623152100179153,
      "tokens_seen": 1078722560
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003399799398194584,
      "loss": 3.1354,
      "theoretical_loss": 3.6231314453653516,
      "tokens_seen": 1078788096
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033996990972918757,
      "loss": 2.9431,
      "theoretical_loss": 3.6231107921575973,
      "tokens_seen": 1078853632
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003399598796389168,
      "loss": 2.9908,
      "theoretical_loss": 3.623090140555667,
      "tokens_seen": 1078919168
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033994984954864593,
      "loss": 3.0363,
      "theoretical_loss": 3.623069490559339,
      "tokens_seen": 1078984704
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033993981945837517,
      "loss": 2.8576,
      "theoretical_loss": 3.62304884216839,
      "tokens_seen": 1079050240
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003399297893681043,
      "loss": 2.9213,
      "theoretical_loss": 3.6230281953825987,
      "tokens_seen": 1079115776
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033991975927783353,
      "loss": 3.0591,
      "theoretical_loss": 3.623007550201742,
      "tokens_seen": 1079181312
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003399097291875627,
      "loss": 3.0357,
      "theoretical_loss": 3.6229869066255977,
      "tokens_seen": 1079246848
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003398996990972919,
      "loss": 3.0696,
      "theoretical_loss": 3.6229662646539444,
      "tokens_seen": 1079312384
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033988966900702107,
      "loss": 2.8854,
      "theoretical_loss": 3.6229456242865594,
      "tokens_seen": 1079377920
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033987963891675025,
      "loss": 2.9419,
      "theoretical_loss": 3.6229249855232206,
      "tokens_seen": 1079443456
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033986960882647943,
      "loss": 2.9969,
      "theoretical_loss": 3.6229043483637065,
      "tokens_seen": 1079508992
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033985957873620867,
      "loss": 2.8449,
      "theoretical_loss": 3.622883712807795,
      "tokens_seen": 1079574528
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003398495486459378,
      "loss": 2.9103,
      "theoretical_loss": 3.6228630788552634,
      "tokens_seen": 1079640064
    },
    {
      "epoch": 0.33,
      "objective/train/advantage_avg": 0.49528974294662476,
      "objective/train/docs_used": 616160,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.386021852493286,
      "objective/train/original_loss": 3.3860223293304443,
      "objective/train/theoretical_loss": 3.6228424465058904,
      "objective/train/tokens_used": 1100165600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2453775852918625,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507763624191284,
      "objective/train/weighted_lm_loss": 3.557960271835327,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0363596677780151,
      "theoretical_loss": 3.6228424465058904,
      "tokens_seen": 1079705600
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033983951855566703,
      "loss": 3.0172,
      "theoretical_loss": 3.6228424465058904,
      "tokens_seen": 1079705600
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003398294884653962,
      "loss": 2.7953,
      "theoretical_loss": 3.622821815759454,
      "tokens_seen": 1079771136
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003398194583751254,
      "loss": 2.9039,
      "theoretical_loss": 3.6228011866157326,
      "tokens_seen": 1079836672
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003398094282848546,
      "loss": 3.0137,
      "theoretical_loss": 3.6227805590745037,
      "tokens_seen": 1079902208
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033979939819458376,
      "loss": 3.0874,
      "theoretical_loss": 3.622759933135547,
      "tokens_seen": 1079967744
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033978936810431294,
      "loss": 3.1043,
      "theoretical_loss": 3.6227393087986393,
      "tokens_seen": 1080033280
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033977933801404217,
      "loss": 2.6934,
      "theoretical_loss": 3.6227186860635596,
      "tokens_seen": 1080098816
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003397693079237713,
      "loss": 3.0445,
      "theoretical_loss": 3.622698064930087,
      "tokens_seen": 1080164352
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033975927783350053,
      "loss": 2.7746,
      "theoretical_loss": 3.622677445397999,
      "tokens_seen": 1080229888
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033974924774322966,
      "loss": 3.1531,
      "theoretical_loss": 3.6226568274670745,
      "tokens_seen": 1080295424
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003397392176529589,
      "loss": 2.9281,
      "theoretical_loss": 3.6226362111370918,
      "tokens_seen": 1080360960
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003397291875626881,
      "loss": 2.9355,
      "theoretical_loss": 3.6226155964078295,
      "tokens_seen": 1080426496
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033971915747241726,
      "loss": 3.0073,
      "theoretical_loss": 3.6225949832790665,
      "tokens_seen": 1080492032
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033970912738214644,
      "loss": 2.8904,
      "theoretical_loss": 3.6225743717505816,
      "tokens_seen": 1080557568
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003396990972918756,
      "loss": 2.995,
      "theoretical_loss": 3.622553761822153,
      "tokens_seen": 1080623104
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003396890672016048,
      "loss": 2.9164,
      "theoretical_loss": 3.6225331534935594,
      "tokens_seen": 1080688640
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033967903711133404,
      "loss": 3.146,
      "theoretical_loss": 3.6225125467645807,
      "tokens_seen": 1080754176
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033966900702106316,
      "loss": 3.1074,
      "theoretical_loss": 3.622491941634994,
      "tokens_seen": 1080819712
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003396589769307924,
      "loss": 2.9816,
      "theoretical_loss": 3.62247133810458,
      "tokens_seen": 1080885248
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003396489468405216,
      "loss": 2.8525,
      "theoretical_loss": 3.622450736173116,
      "tokens_seen": 1080950784
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033963891675025076,
      "loss": 3.0379,
      "theoretical_loss": 3.6224301358403825,
      "tokens_seen": 1081016320
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033962888665997994,
      "loss": 2.9947,
      "theoretical_loss": 3.622409537106158,
      "tokens_seen": 1081081856
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003396188565697091,
      "loss": 2.7842,
      "theoretical_loss": 3.6223889399702207,
      "tokens_seen": 1081147392
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003396088264794383,
      "loss": 2.8621,
      "theoretical_loss": 3.6223683444323505,
      "tokens_seen": 1081212928
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033959879638916754,
      "loss": 2.902,
      "theoretical_loss": 3.6223477504923265,
      "tokens_seen": 1081278464
    },
    {
      "epoch": 0.33,
      "objective/train/advantage_avg": 0.49573010206222534,
      "objective/train/docs_used": 616786,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.1175262928009033,
      "objective/train/original_loss": 3.1175265312194824,
      "objective/train/theoretical_loss": 3.622327158149928,
      "objective/train/tokens_used": 1101804000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2457893341779709,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050822377204895,
      "objective/train/weighted_lm_loss": 3.2760298252105713,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0457526445388794,
      "theoretical_loss": 3.622327158149928,
      "tokens_seen": 1081344000
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033958876629889667,
      "loss": 3.0118,
      "theoretical_loss": 3.622327158149928,
      "tokens_seen": 1081344000
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003395787362086259,
      "loss": 2.999,
      "theoretical_loss": 3.622306567404934,
      "tokens_seen": 1081409536
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033956870611835503,
      "loss": 2.9403,
      "theoretical_loss": 3.622285978257124,
      "tokens_seen": 1081475072
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033955867602808426,
      "loss": 2.9962,
      "theoretical_loss": 3.6222653907062776,
      "tokens_seen": 1081540608
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033954864593781345,
      "loss": 2.9313,
      "theoretical_loss": 3.622244804752173,
      "tokens_seen": 1081606144
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003395386158475426,
      "loss": 2.8428,
      "theoretical_loss": 3.6222242203945907,
      "tokens_seen": 1081671680
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003395285857572718,
      "loss": 2.8697,
      "theoretical_loss": 3.6222036376333104,
      "tokens_seen": 1081737216
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033951855566700104,
      "loss": 3.051,
      "theoretical_loss": 3.622183056468111,
      "tokens_seen": 1081802752
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033950852557673017,
      "loss": 2.8733,
      "theoretical_loss": 3.6221624768987715,
      "tokens_seen": 1081868288
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003394984954864594,
      "loss": 2.9685,
      "theoretical_loss": 3.622141898925072,
      "tokens_seen": 1081933824
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033948846539618853,
      "loss": 3.0097,
      "theoretical_loss": 3.622121322546793,
      "tokens_seen": 1081999360
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033947843530591777,
      "loss": 2.9088,
      "theoretical_loss": 3.6221007477637137,
      "tokens_seen": 1082064896
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033946840521564695,
      "loss": 3.192,
      "theoretical_loss": 3.622080174575613,
      "tokens_seen": 1082130432
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033945837512537613,
      "loss": 3.2376,
      "theoretical_loss": 3.6220596029822714,
      "tokens_seen": 1082195968
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003394483450351053,
      "loss": 3.002,
      "theoretical_loss": 3.6220390329834684,
      "tokens_seen": 1082261504
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003394383149448345,
      "loss": 2.9757,
      "theoretical_loss": 3.622018464578985,
      "tokens_seen": 1082327040
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003394282848545637,
      "loss": 2.9482,
      "theoretical_loss": 3.621997897768599,
      "tokens_seen": 1082392576
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003394182547642929,
      "loss": 2.9678,
      "theoretical_loss": 3.621977332552092,
      "tokens_seen": 1082458112
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033940822467402204,
      "loss": 2.8877,
      "theoretical_loss": 3.621956768929243,
      "tokens_seen": 1082523648
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033939819458375127,
      "loss": 2.734,
      "theoretical_loss": 3.621936206899833,
      "tokens_seen": 1082589184
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003393881644934804,
      "loss": 2.8636,
      "theoretical_loss": 3.6219156464636413,
      "tokens_seen": 1082654720
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033937813440320963,
      "loss": 3.0619,
      "theoretical_loss": 3.621895087620449,
      "tokens_seen": 1082720256
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003393681043129388,
      "loss": 2.891,
      "theoretical_loss": 3.621874530370035,
      "tokens_seen": 1082785792
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.000339358074222668,
      "loss": 3.0244,
      "theoretical_loss": 3.62185397471218,
      "tokens_seen": 1082851328
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003393480441323972,
      "loss": 3.052,
      "theoretical_loss": 3.6218334206466647,
      "tokens_seen": 1082916864
    },
    {
      "epoch": 0.33,
      "objective/train/advantage_avg": 0.4932069480419159,
      "objective/train/docs_used": 617451,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.103405714035034,
      "objective/train/original_loss": 3.103405714035034,
      "objective/train/theoretical_loss": 3.6218128681732686,
      "objective/train/tokens_used": 1103442400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24334865808486938,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505576133728027,
      "objective/train/weighted_lm_loss": 3.2605350017547607,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0363596677780151,
      "theoretical_loss": 3.6218128681732686,
      "tokens_seen": 1082982400
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003393380140421264,
      "loss": 3.2151,
      "theoretical_loss": 3.6218128681732686,
      "tokens_seen": 1082982400
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003393279839518556,
      "loss": 3.0469,
      "theoretical_loss": 3.621792317291773,
      "tokens_seen": 1083047936
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003393179538615848,
      "loss": 2.8435,
      "theoretical_loss": 3.621771768001958,
      "tokens_seen": 1083113472
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033930792377131396,
      "loss": 2.9792,
      "theoretical_loss": 3.6217512203036026,
      "tokens_seen": 1083179008
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033929789368104314,
      "loss": 2.9669,
      "theoretical_loss": 3.621730674196489,
      "tokens_seen": 1083244544
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033928786359077237,
      "loss": 2.9664,
      "theoretical_loss": 3.6217101296803977,
      "tokens_seen": 1083310080
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003392778335005015,
      "loss": 3.1006,
      "theoretical_loss": 3.6216895867551084,
      "tokens_seen": 1083375616
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033926780341023073,
      "loss": 3.0637,
      "theoretical_loss": 3.621669045420402,
      "tokens_seen": 1083441152
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033925777331995986,
      "loss": 2.6478,
      "theoretical_loss": 3.6216485056760592,
      "tokens_seen": 1083506688
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003392477432296891,
      "loss": 2.7649,
      "theoretical_loss": 3.621627967521861,
      "tokens_seen": 1083572224
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003392377131394183,
      "loss": 3.1581,
      "theoretical_loss": 3.621607430957587,
      "tokens_seen": 1083637760
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033922768304914746,
      "loss": 2.8966,
      "theoretical_loss": 3.621586895983019,
      "tokens_seen": 1083703296
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033921765295887664,
      "loss": 2.7839,
      "theoretical_loss": 3.621566362597938,
      "tokens_seen": 1083768832
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003392076228686058,
      "loss": 2.9454,
      "theoretical_loss": 3.621545830802124,
      "tokens_seen": 1083834368
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.000339197592778335,
      "loss": 3.0127,
      "theoretical_loss": 3.621525300595359,
      "tokens_seen": 1083899904
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033918756268806424,
      "loss": 2.8966,
      "theoretical_loss": 3.6215047719774227,
      "tokens_seen": 1083965440
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033917753259779336,
      "loss": 3.1175,
      "theoretical_loss": 3.621484244948097,
      "tokens_seen": 1084030976
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003391675025075226,
      "loss": 2.8148,
      "theoretical_loss": 3.6214637195071626,
      "tokens_seen": 1084096512
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003391574724172518,
      "loss": 3.0425,
      "theoretical_loss": 3.6214431956544004,
      "tokens_seen": 1084162048
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033914744232698096,
      "loss": 2.8042,
      "theoretical_loss": 3.621422673389592,
      "tokens_seen": 1084227584
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033913741223671014,
      "loss": 3.1133,
      "theoretical_loss": 3.621402152712518,
      "tokens_seen": 1084293120
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003391273821464393,
      "loss": 3.094,
      "theoretical_loss": 3.62138163362296,
      "tokens_seen": 1084358656
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003391173520561685,
      "loss": 3.0381,
      "theoretical_loss": 3.621361116120699,
      "tokens_seen": 1084424192
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033910732196589774,
      "loss": 3.0892,
      "theoretical_loss": 3.621340600205517,
      "tokens_seen": 1084489728
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033909729187562687,
      "loss": 3.0564,
      "theoretical_loss": 3.621320085877194,
      "tokens_seen": 1084555264
    },
    {
      "epoch": 0.33,
      "objective/train/advantage_avg": 0.4942246675491333,
      "objective/train/docs_used": 618647,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.8410749435424805,
      "objective/train/original_loss": 2.8410751819610596,
      "objective/train/theoretical_loss": 3.621299573135513,
      "objective/train/tokens_used": 1105080800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24431510269641876,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506644248962402,
      "objective/train/weighted_lm_loss": 2.985311508178711,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0408108234405518,
      "theoretical_loss": 3.621299573135513,
      "tokens_seen": 1084620800
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003390872617853561,
      "loss": 2.8753,
      "theoretical_loss": 3.621299573135513,
      "tokens_seen": 1084620800
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033907723169508523,
      "loss": 2.7868,
      "theoretical_loss": 3.621279061980254,
      "tokens_seen": 1084686336
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033906720160481446,
      "loss": 3.0312,
      "theoretical_loss": 3.6212585524111995,
      "tokens_seen": 1084751872
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033905717151454365,
      "loss": 2.8851,
      "theoretical_loss": 3.6212380444281305,
      "tokens_seen": 1084817408
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033904714142427283,
      "loss": 2.9009,
      "theoretical_loss": 3.6212175380308285,
      "tokens_seen": 1084882944
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.000339037111334002,
      "loss": 3.0553,
      "theoretical_loss": 3.621197033219075,
      "tokens_seen": 1084948480
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033902708124373124,
      "loss": 3.0371,
      "theoretical_loss": 3.621176529992652,
      "tokens_seen": 1085014016
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033901705115346037,
      "loss": 2.3388,
      "theoretical_loss": 3.6211560283513418,
      "tokens_seen": 1085079552
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003390070210631896,
      "loss": 2.9296,
      "theoretical_loss": 3.6211355282949245,
      "tokens_seen": 1085145088
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033899699097291873,
      "loss": 2.963,
      "theoretical_loss": 3.6211150298231827,
      "tokens_seen": 1085210624
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033898696088264797,
      "loss": 2.8959,
      "theoretical_loss": 3.6210945329358992,
      "tokens_seen": 1085276160
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033897693079237715,
      "loss": 3.0161,
      "theoretical_loss": 3.6210740376328543,
      "tokens_seen": 1085341696
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033896690070210633,
      "loss": 2.7724,
      "theoretical_loss": 3.6210535439138303,
      "tokens_seen": 1085407232
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003389568706118355,
      "loss": 3.0015,
      "theoretical_loss": 3.6210330517786096,
      "tokens_seen": 1085472768
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003389468405215647,
      "loss": 2.8831,
      "theoretical_loss": 3.6210125612269746,
      "tokens_seen": 1085538304
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003389368104312939,
      "loss": 2.5947,
      "theoretical_loss": 3.620992072258706,
      "tokens_seen": 1085603840
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003389267803410231,
      "loss": 3.1089,
      "theoretical_loss": 3.620971584873587,
      "tokens_seen": 1085669376
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033891675025075224,
      "loss": 2.8555,
      "theoretical_loss": 3.620951099071399,
      "tokens_seen": 1085734912
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033890672016048147,
      "loss": 3.2484,
      "theoretical_loss": 3.6209306148519245,
      "tokens_seen": 1085800448
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003388966900702106,
      "loss": 3.0061,
      "theoretical_loss": 3.620910132214946,
      "tokens_seen": 1085865984
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033888665997993983,
      "loss": 2.8191,
      "theoretical_loss": 3.620889651160245,
      "tokens_seen": 1085931520
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.000338876629889669,
      "loss": 2.9259,
      "theoretical_loss": 3.620869171687604,
      "tokens_seen": 1085997056
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003388665997993982,
      "loss": 3.1446,
      "theoretical_loss": 3.620848693796806,
      "tokens_seen": 1086062592
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003388565697091274,
      "loss": 3.0944,
      "theoretical_loss": 3.6208282174876327,
      "tokens_seen": 1086128128
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003388465396188566,
      "loss": 2.8607,
      "theoretical_loss": 3.620807742759867,
      "tokens_seen": 1086193664
    },
    {
      "epoch": 0.33,
      "objective/train/advantage_avg": 0.49413469433784485,
      "objective/train/docs_used": 619458,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.641779899597168,
      "objective/train/original_loss": 2.641779899597168,
      "objective/train/theoretical_loss": 3.620787269613291,
      "objective/train/tokens_used": 1106719200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24424579739570618,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506550073623657,
      "objective/train/weighted_lm_loss": 2.7756171226501465,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 3.620787269613291,
      "tokens_seen": 1086259200
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033883650952858574,
      "loss": 2.921,
      "theoretical_loss": 3.620787269613291,
      "tokens_seen": 1086259200
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.000338826479438315,
      "loss": 2.7936,
      "theoretical_loss": 3.6207667980476868,
      "tokens_seen": 1086324736
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003388164493480441,
      "loss": 3.0804,
      "theoretical_loss": 3.620746328062838,
      "tokens_seen": 1086390272
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033880641925777334,
      "loss": 2.7834,
      "theoretical_loss": 3.6207258596585263,
      "tokens_seen": 1086455808
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003387963891675025,
      "loss": 2.8626,
      "theoretical_loss": 3.6207053928345347,
      "tokens_seen": 1086521344
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003387863590772317,
      "loss": 2.961,
      "theoretical_loss": 3.620684927590646,
      "tokens_seen": 1086586880
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003387763289869609,
      "loss": 2.868,
      "theoretical_loss": 3.6206644639266425,
      "tokens_seen": 1086652416
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033876629889669006,
      "loss": 2.7736,
      "theoretical_loss": 3.620644001842308,
      "tokens_seen": 1086717952
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033875626880641924,
      "loss": 2.4058,
      "theoretical_loss": 3.620623541337424,
      "tokens_seen": 1086783488
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003387462387161485,
      "loss": 3.0392,
      "theoretical_loss": 3.6206030824117734,
      "tokens_seen": 1086849024
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003387362086258776,
      "loss": 2.9046,
      "theoretical_loss": 3.6205826250651403,
      "tokens_seen": 1086914560
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033872617853560684,
      "loss": 3.0704,
      "theoretical_loss": 3.6205621692973065,
      "tokens_seen": 1086980096
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033871614844533597,
      "loss": 2.8589,
      "theoretical_loss": 3.6205417151080557,
      "tokens_seen": 1087045632
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003387061183550652,
      "loss": 2.9073,
      "theoretical_loss": 3.6205212624971708,
      "tokens_seen": 1087111168
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003386960882647944,
      "loss": 3.0212,
      "theoretical_loss": 3.620500811464434,
      "tokens_seen": 1087176704
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033868605817452356,
      "loss": 2.9475,
      "theoretical_loss": 3.62048036200963,
      "tokens_seen": 1087242240
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033867602808425275,
      "loss": 2.6346,
      "theoretical_loss": 3.62045991413254,
      "tokens_seen": 1087307776
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.000338665997993982,
      "loss": 3.055,
      "theoretical_loss": 3.620439467832949,
      "tokens_seen": 1087373312
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003386559679037111,
      "loss": 3.0785,
      "theoretical_loss": 3.620419023110639,
      "tokens_seen": 1087438848
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033864593781344034,
      "loss": 2.9818,
      "theoretical_loss": 3.620398579965394,
      "tokens_seen": 1087504384
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033863590772316947,
      "loss": 2.9165,
      "theoretical_loss": 3.620378138396997,
      "tokens_seen": 1087569920
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003386258776328987,
      "loss": 2.8814,
      "theoretical_loss": 3.6203576984052317,
      "tokens_seen": 1087635456
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003386158475426279,
      "loss": 3.0513,
      "theoretical_loss": 3.6203372599898804,
      "tokens_seen": 1087700992
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033860581745235707,
      "loss": 2.8822,
      "theoretical_loss": 3.620316823150728,
      "tokens_seen": 1087766528
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033859578736208625,
      "loss": 2.9689,
      "theoretical_loss": 3.6202963878875574,
      "tokens_seen": 1087832064
    },
    {
      "epoch": 0.33,
      "objective/train/advantage_avg": 0.49565160274505615,
      "objective/train/docs_used": 620460,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.7528347969055176,
      "objective/train/original_loss": 2.7528347969055176,
      "objective/train/theoretical_loss": 3.620275954200152,
      "objective/train/tokens_used": 1108357600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24575240910053253,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0508143901824951,
      "objective/train/weighted_lm_loss": 2.8927319049835205,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0363596677780151,
      "theoretical_loss": 3.620275954200152,
      "tokens_seen": 1087897600
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033858575727181543,
      "loss": 2.878,
      "theoretical_loss": 3.620275954200152,
      "tokens_seen": 1087897600
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033857572718154467,
      "loss": 2.805,
      "theoretical_loss": 3.6202555220882955,
      "tokens_seen": 1087963136
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033856569709127385,
      "loss": 3.0527,
      "theoretical_loss": 3.6202350915517716,
      "tokens_seen": 1088028672
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033855566700100303,
      "loss": 2.9636,
      "theoretical_loss": 3.6202146625903637,
      "tokens_seen": 1088094208
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003385456369107322,
      "loss": 3.0476,
      "theoretical_loss": 3.620194235203856,
      "tokens_seen": 1088159744
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033853560682046144,
      "loss": 3.0032,
      "theoretical_loss": 3.620173809392032,
      "tokens_seen": 1088225280
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033852557673019057,
      "loss": 3.0865,
      "theoretical_loss": 3.6201533851546746,
      "tokens_seen": 1088290816
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003385155466399198,
      "loss": 2.9416,
      "theoretical_loss": 3.620132962491569,
      "tokens_seen": 1088356352
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033850551654964893,
      "loss": 2.9488,
      "theoretical_loss": 3.6201125414024986,
      "tokens_seen": 1088421888
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033849548645937817,
      "loss": 3.0672,
      "theoretical_loss": 3.620092121887247,
      "tokens_seen": 1088487424
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033848545636910735,
      "loss": 2.9903,
      "theoretical_loss": 3.6200717039455985,
      "tokens_seen": 1088552960
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033847542627883653,
      "loss": 3.0695,
      "theoretical_loss": 3.620051287577337,
      "tokens_seen": 1088618496
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003384653961885657,
      "loss": 2.9829,
      "theoretical_loss": 3.620030872782247,
      "tokens_seen": 1088684032
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003384553660982949,
      "loss": 2.661,
      "theoretical_loss": 3.6200104595601115,
      "tokens_seen": 1088749568
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003384453360080241,
      "loss": 3.001,
      "theoretical_loss": 3.619990047910716,
      "tokens_seen": 1088815104
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003384353059177533,
      "loss": 2.9381,
      "theoretical_loss": 3.6199696378338433,
      "tokens_seen": 1088880640
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033842527582748244,
      "loss": 3.0289,
      "theoretical_loss": 3.6199492293292783,
      "tokens_seen": 1088946176
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033841524573721167,
      "loss": 2.8953,
      "theoretical_loss": 3.6199288223968056,
      "tokens_seen": 1089011712
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003384052156469408,
      "loss": 3.0108,
      "theoretical_loss": 3.619908417036209,
      "tokens_seen": 1089077248
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033839518555667003,
      "loss": 2.9426,
      "theoretical_loss": 3.619888013247273,
      "tokens_seen": 1089142784
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003383851554663992,
      "loss": 3.0129,
      "theoretical_loss": 3.619867611029782,
      "tokens_seen": 1089208320
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003383751253761284,
      "loss": 2.9994,
      "theoretical_loss": 3.6198472103835204,
      "tokens_seen": 1089273856
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003383650952858576,
      "loss": 2.8097,
      "theoretical_loss": 3.6198268113082728,
      "tokens_seen": 1089339392
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003383550651955868,
      "loss": 3.1057,
      "theoretical_loss": 3.6198064138038237,
      "tokens_seen": 1089404928
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033834503510531594,
      "loss": 2.8437,
      "theoretical_loss": 3.619786017869957,
      "tokens_seen": 1089470464
    },
    {
      "epoch": 0.33,
      "objective/train/advantage_avg": 0.4935101568698883,
      "objective/train/docs_used": 620976,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.0405521392822266,
      "objective/train/original_loss": 3.0405516624450684,
      "objective/train/theoretical_loss": 3.619765623506458,
      "objective/train/tokens_used": 1109996000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24375854432582855,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505900382995605,
      "objective/train/weighted_lm_loss": 3.1945815086364746,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0338950157165527,
      "theoretical_loss": 3.619765623506458,
      "tokens_seen": 1089536000
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003383350050150452,
      "loss": 2.949,
      "theoretical_loss": 3.619765623506458,
      "tokens_seen": 1089536000
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003383249749247743,
      "loss": 2.8953,
      "theoretical_loss": 3.6197452307131117,
      "tokens_seen": 1089601536
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033831494483450354,
      "loss": 2.9561,
      "theoretical_loss": 3.619724839489702,
      "tokens_seen": 1089667072
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003383049147442327,
      "loss": 2.7177,
      "theoretical_loss": 3.6197044498360142,
      "tokens_seen": 1089732608
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003382948846539619,
      "loss": 2.7906,
      "theoretical_loss": 3.6196840617518324,
      "tokens_seen": 1089798144
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003382848545636911,
      "loss": 2.9087,
      "theoretical_loss": 3.6196636752369415,
      "tokens_seen": 1089863680
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033827482447342026,
      "loss": 2.7274,
      "theoretical_loss": 3.619643290291127,
      "tokens_seen": 1089929216
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033826479438314944,
      "loss": 3.1717,
      "theoretical_loss": 3.6196229069141737,
      "tokens_seen": 1089994752
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003382547642928787,
      "loss": 2.7539,
      "theoretical_loss": 3.6196025251058663,
      "tokens_seen": 1090060288
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003382447342026078,
      "loss": 2.8165,
      "theoretical_loss": 3.6195821448659897,
      "tokens_seen": 1090125824
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033823470411233704,
      "loss": 2.7196,
      "theoretical_loss": 3.6195617661943285,
      "tokens_seen": 1090191360
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033822467402206617,
      "loss": 2.6593,
      "theoretical_loss": 3.619541389090669,
      "tokens_seen": 1090256896
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003382146439317954,
      "loss": 3.0696,
      "theoretical_loss": 3.6195210135547953,
      "tokens_seen": 1090322432
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003382046138415246,
      "loss": 3.1425,
      "theoretical_loss": 3.6195006395864926,
      "tokens_seen": 1090387968
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033819458375125376,
      "loss": 2.9552,
      "theoretical_loss": 3.619480267185547,
      "tokens_seen": 1090453504
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033818455366098295,
      "loss": 2.9997,
      "theoretical_loss": 3.619459896351742,
      "tokens_seen": 1090519040
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003381745235707122,
      "loss": 3.0022,
      "theoretical_loss": 3.6194395270848645,
      "tokens_seen": 1090584576
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003381644934804413,
      "loss": 3.156,
      "theoretical_loss": 3.6194191593847,
      "tokens_seen": 1090650112
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033815446339017054,
      "loss": 2.8879,
      "theoretical_loss": 3.6193987932510323,
      "tokens_seen": 1090715648
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033814443329989967,
      "loss": 2.761,
      "theoretical_loss": 3.619378428683648,
      "tokens_seen": 1090781184
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003381344032096289,
      "loss": 3.2199,
      "theoretical_loss": 3.6193580656823316,
      "tokens_seen": 1090846720
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003381243731193581,
      "loss": 3.0663,
      "theoretical_loss": 3.6193377042468695,
      "tokens_seen": 1090912256
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033811434302908727,
      "loss": 2.9404,
      "theoretical_loss": 3.6193173443770474,
      "tokens_seen": 1090977792
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033810431293881645,
      "loss": 2.5918,
      "theoretical_loss": 3.6192969860726496,
      "tokens_seen": 1091043328
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033809428284854563,
      "loss": 2.821,
      "theoretical_loss": 3.619276629333463,
      "tokens_seen": 1091108864
    },
    {
      "epoch": 0.33,
      "objective/train/advantage_avg": 0.4947642683982849,
      "objective/train/docs_used": 621707,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.739819288253784,
      "objective/train/original_loss": 2.739818811416626,
      "objective/train/theoretical_loss": 3.6192562741592726,
      "objective/train/tokens_used": 1111634400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2448539286851883,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507210493087769,
      "objective/train/weighted_lm_loss": 2.878462314605713,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0461136102676392,
      "theoretical_loss": 3.6192562741592726,
      "tokens_seen": 1091174400
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003380842527582748,
      "loss": 2.8552,
      "theoretical_loss": 3.6192562741592726,
      "tokens_seen": 1091174400
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033807422266800405,
      "loss": 2.9605,
      "theoretical_loss": 3.6192359205498645,
      "tokens_seen": 1091239936
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003380641925777332,
      "loss": 2.8531,
      "theoretical_loss": 3.619215568505024,
      "tokens_seen": 1091305472
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003380541624874624,
      "loss": 2.9665,
      "theoretical_loss": 3.6191952180245375,
      "tokens_seen": 1091371008
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033804413239719154,
      "loss": 2.9922,
      "theoretical_loss": 3.61917486910819,
      "tokens_seen": 1091436544
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033803410230692077,
      "loss": 2.8851,
      "theoretical_loss": 3.6191545217557675,
      "tokens_seen": 1091502080
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033802407221664995,
      "loss": 2.9673,
      "theoretical_loss": 3.6191341759670568,
      "tokens_seen": 1091567616
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033801404212637913,
      "loss": 2.9164,
      "theoretical_loss": 3.619113831741843,
      "tokens_seen": 1091633152
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003380040120361083,
      "loss": 2.8078,
      "theoretical_loss": 3.619093489079913,
      "tokens_seen": 1091698688
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033799398194583755,
      "loss": 3.0965,
      "theoretical_loss": 3.6190731479810516,
      "tokens_seen": 1091764224
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003379839518555667,
      "loss": 2.7683,
      "theoretical_loss": 3.619052808445045,
      "tokens_seen": 1091829760
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003379739217652959,
      "loss": 3.0348,
      "theoretical_loss": 3.619032470471681,
      "tokens_seen": 1091895296
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033796389167502504,
      "loss": 3.0256,
      "theoretical_loss": 3.6190121340607435,
      "tokens_seen": 1091960832
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003379538615847543,
      "loss": 2.8763,
      "theoretical_loss": 3.6189917992120204,
      "tokens_seen": 1092026368
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033794383149448346,
      "loss": 2.9212,
      "theoretical_loss": 3.618971465925297,
      "tokens_seen": 1092091904
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033793380140421264,
      "loss": 3.0793,
      "theoretical_loss": 3.61895113420036,
      "tokens_seen": 1092157440
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003379237713139418,
      "loss": 2.7875,
      "theoretical_loss": 3.618930804036996,
      "tokens_seen": 1092222976
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.000337913741223671,
      "loss": 2.7835,
      "theoretical_loss": 3.618910475434991,
      "tokens_seen": 1092288512
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003379037111334002,
      "loss": 2.8234,
      "theoretical_loss": 3.618890148394131,
      "tokens_seen": 1092354048
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003378936810431294,
      "loss": 3.1087,
      "theoretical_loss": 3.6188698229142036,
      "tokens_seen": 1092419584
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033788365095285854,
      "loss": 2.9394,
      "theoretical_loss": 3.618849498994994,
      "tokens_seen": 1092485120
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003378736208625878,
      "loss": 2.9159,
      "theoretical_loss": 3.61882917663629,
      "tokens_seen": 1092550656
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003378635907723169,
      "loss": 3.0371,
      "theoretical_loss": 3.618808855837877,
      "tokens_seen": 1092616192
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033785356068204614,
      "loss": 3.0024,
      "theoretical_loss": 3.618788536599543,
      "tokens_seen": 1092681728
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003378435305917753,
      "loss": 2.8843,
      "theoretical_loss": 3.6187682189210726,
      "tokens_seen": 1092747264
    },
    {
      "epoch": 0.33,
      "objective/train/advantage_avg": 0.4943099021911621,
      "objective/train/docs_used": 622698,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.953923463821411,
      "objective/train/original_loss": 2.9539241790771484,
      "objective/train/theoretical_loss": 3.6187479028022547,
      "objective/train/tokens_used": 1113272800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24440418183803558,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506733655929565,
      "objective/train/weighted_lm_loss": 3.1035335063934326,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0425469875335693,
      "theoretical_loss": 3.6187479028022547,
      "tokens_seen": 1092812800
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003378335005015045,
      "loss": 3.2123,
      "theoretical_loss": 3.6187479028022547,
      "tokens_seen": 1092812800
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033782347041123374,
      "loss": 2.714,
      "theoretical_loss": 3.6187275882428747,
      "tokens_seen": 1092878336
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003378134403209629,
      "loss": 2.8643,
      "theoretical_loss": 3.61870727524272,
      "tokens_seen": 1092943872
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003378034102306921,
      "loss": 3.2466,
      "theoretical_loss": 3.6186869638015775,
      "tokens_seen": 1093009408
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003377933801404213,
      "loss": 2.9513,
      "theoretical_loss": 3.6186666539192336,
      "tokens_seen": 1093074944
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033778335005015046,
      "loss": 2.7297,
      "theoretical_loss": 3.618646345595475,
      "tokens_seen": 1093140480
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033777331995987964,
      "loss": 2.9512,
      "theoretical_loss": 3.6186260388300897,
      "tokens_seen": 1093206016
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003377632898696089,
      "loss": 3.1982,
      "theoretical_loss": 3.6186057336228643,
      "tokens_seen": 1093271552
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.000337753259779338,
      "loss": 2.9185,
      "theoretical_loss": 3.618585429973585,
      "tokens_seen": 1093337088
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033774322968906724,
      "loss": 2.9902,
      "theoretical_loss": 3.6185651278820403,
      "tokens_seen": 1093402624
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033773319959879637,
      "loss": 3.1372,
      "theoretical_loss": 3.618544827348016,
      "tokens_seen": 1093468160
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003377231695085256,
      "loss": 3.0276,
      "theoretical_loss": 3.6185245283713003,
      "tokens_seen": 1093533696
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003377131394182548,
      "loss": 3.2943,
      "theoretical_loss": 3.6185042309516797,
      "tokens_seen": 1093599232
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033770310932798396,
      "loss": 3.0455,
      "theoretical_loss": 3.6184839350889417,
      "tokens_seen": 1093664768
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033769307923771315,
      "loss": 2.8922,
      "theoretical_loss": 3.618463640782874,
      "tokens_seen": 1093730304
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003376830491474424,
      "loss": 2.949,
      "theoretical_loss": 3.618443348033263,
      "tokens_seen": 1093795840
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003376730190571715,
      "loss": 2.9329,
      "theoretical_loss": 3.618423056839897,
      "tokens_seen": 1093861376
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033766298896690074,
      "loss": 3.1348,
      "theoretical_loss": 3.6184027672025634,
      "tokens_seen": 1093926912
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033765295887662987,
      "loss": 3.0814,
      "theoretical_loss": 3.618382479121049,
      "tokens_seen": 1093992448
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003376429287863591,
      "loss": 2.8476,
      "theoretical_loss": 3.618362192595142,
      "tokens_seen": 1094057984
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003376328986960883,
      "loss": 2.9459,
      "theoretical_loss": 3.6183419076246293,
      "tokens_seen": 1094123520
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033762286860581747,
      "loss": 2.812,
      "theoretical_loss": 3.618321624209299,
      "tokens_seen": 1094189056
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033761283851554665,
      "loss": 3.0313,
      "theoretical_loss": 3.6183013423489383,
      "tokens_seen": 1094254592
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033760280842527583,
      "loss": 2.8566,
      "theoretical_loss": 3.618281062043335,
      "tokens_seen": 1094320128
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.000337592778335005,
      "loss": 3.2674,
      "theoretical_loss": 3.6182607832922775,
      "tokens_seen": 1094385664
    },
    {
      "epoch": 0.33,
      "objective/train/advantage_avg": 0.49554604291915894,
      "objective/train/docs_used": 623416,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.6778464317321777,
      "objective/train/original_loss": 2.6778461933135986,
      "objective/train/theoretical_loss": 3.6182405060955523,
      "objective/train/tokens_used": 1114911200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24561844766139984,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0508031845092773,
      "objective/train/weighted_lm_loss": 2.8137426376342773,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0425469875335693,
      "theoretical_loss": 3.6182405060955523,
      "tokens_seen": 1094451200
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033758274824473425,
      "loss": 2.8259,
      "theoretical_loss": 3.6182405060955523,
      "tokens_seen": 1094451200
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003375727181544634,
      "loss": 3.086,
      "theoretical_loss": 3.6182202304529483,
      "tokens_seen": 1094516736
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003375626880641926,
      "loss": 3.0518,
      "theoretical_loss": 3.6181999563642533,
      "tokens_seen": 1094582272
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033755265797392174,
      "loss": 3.0882,
      "theoretical_loss": 3.618179683829254,
      "tokens_seen": 1094647808
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033754262788365097,
      "loss": 3.0637,
      "theoretical_loss": 3.6181594128477395,
      "tokens_seen": 1094713344
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033753259779338015,
      "loss": 2.9123,
      "theoretical_loss": 3.618139143419498,
      "tokens_seen": 1094778880
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033752256770310933,
      "loss": 2.9661,
      "theoretical_loss": 3.618118875544316,
      "tokens_seen": 1094844416
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003375125376128385,
      "loss": 2.9785,
      "theoretical_loss": 3.618098609221983,
      "tokens_seen": 1094909952
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033750250752256775,
      "loss": 2.5378,
      "theoretical_loss": 3.618078344452286,
      "tokens_seen": 1094975488
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003374924774322969,
      "loss": 2.9256,
      "theoretical_loss": 3.6180580812350143,
      "tokens_seen": 1095041024
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003374824473420261,
      "loss": 3.221,
      "theoretical_loss": 3.618037819569955,
      "tokens_seen": 1095106560
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033747241725175524,
      "loss": 3.0524,
      "theoretical_loss": 3.6180175594568973,
      "tokens_seen": 1095172096
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003374623871614845,
      "loss": 2.9328,
      "theoretical_loss": 3.6179973008956283,
      "tokens_seen": 1095237632
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033745235707121366,
      "loss": 3.0318,
      "theoretical_loss": 3.6179770438859373,
      "tokens_seen": 1095303168
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033744232698094284,
      "loss": 2.9568,
      "theoretical_loss": 3.6179567884276125,
      "tokens_seen": 1095368704
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.000337432296890672,
      "loss": 2.7783,
      "theoretical_loss": 3.6179365345204415,
      "tokens_seen": 1095434240
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003374222668004012,
      "loss": 2.8691,
      "theoretical_loss": 3.617916282164214,
      "tokens_seen": 1095499776
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003374122367101304,
      "loss": 3.0388,
      "theoretical_loss": 3.6178960313587174,
      "tokens_seen": 1095565312
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003374022066198596,
      "loss": 3.0779,
      "theoretical_loss": 3.61787578210374,
      "tokens_seen": 1095630848
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033739217652958874,
      "loss": 2.6928,
      "theoretical_loss": 3.6178555343990713,
      "tokens_seen": 1095696384
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.000337382146439318,
      "loss": 3.007,
      "theoretical_loss": 3.6178352882444997,
      "tokens_seen": 1095761920
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003373721163490471,
      "loss": 2.9999,
      "theoretical_loss": 3.6178150436398138,
      "tokens_seen": 1095827456
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033736208625877634,
      "loss": 2.8929,
      "theoretical_loss": 3.6177948005848015,
      "tokens_seen": 1095892992
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003373520561685055,
      "loss": 2.9983,
      "theoretical_loss": 3.617774559079253,
      "tokens_seen": 1095958528
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003373420260782347,
      "loss": 3.165,
      "theoretical_loss": 3.6177543191229553,
      "tokens_seen": 1096024064
    },
    {
      "epoch": 0.33,
      "objective/train/advantage_avg": 0.49322599172592163,
      "objective/train/docs_used": 624708,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.514512777328491,
      "objective/train/original_loss": 2.514512538909912,
      "objective/train/theoretical_loss": 3.617734080715698,
      "objective/train/tokens_used": 1116549600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24336163699626923,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050559639930725,
      "objective/train/weighted_lm_loss": 2.641916036605835,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 3.617734080715698,
      "tokens_seen": 1096089600
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003373319959879639,
      "loss": 2.9288,
      "theoretical_loss": 3.617734080715698,
      "tokens_seen": 1096089600
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003373219658976931,
      "loss": 2.9951,
      "theoretical_loss": 3.6177138438572705,
      "tokens_seen": 1096155136
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033731193580742225,
      "loss": 2.9747,
      "theoretical_loss": 3.617693608547461,
      "tokens_seen": 1096220672
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003373019057171515,
      "loss": 2.7211,
      "theoretical_loss": 3.6176733747860585,
      "tokens_seen": 1096286208
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003372918756268806,
      "loss": 3.118,
      "theoretical_loss": 3.617653142572852,
      "tokens_seen": 1096351744
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033728184553660984,
      "loss": 2.8327,
      "theoretical_loss": 3.6176329119076307,
      "tokens_seen": 1096417280
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.000337271815446339,
      "loss": 3.1201,
      "theoretical_loss": 3.6176126827901838,
      "tokens_seen": 1096482816
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003372617853560682,
      "loss": 2.9077,
      "theoretical_loss": 3.6175924552203,
      "tokens_seen": 1096548352
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003372517552657974,
      "loss": 2.8597,
      "theoretical_loss": 3.6175722291977683,
      "tokens_seen": 1096613888
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033724172517552657,
      "loss": 2.9692,
      "theoretical_loss": 3.617552004722378,
      "tokens_seen": 1096679424
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033723169508525575,
      "loss": 2.8301,
      "theoretical_loss": 3.617531781793919,
      "tokens_seen": 1096744960
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.000337221664994985,
      "loss": 2.9373,
      "theoretical_loss": 3.6175115604121793,
      "tokens_seen": 1096810496
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003372116349047141,
      "loss": 2.9653,
      "theoretical_loss": 3.617491340576949,
      "tokens_seen": 1096876032
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033720160481444335,
      "loss": 3.0116,
      "theoretical_loss": 3.6174711222880176,
      "tokens_seen": 1096941568
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003371915747241725,
      "loss": 3.029,
      "theoretical_loss": 3.617450905545174,
      "tokens_seen": 1097007104
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003371815446339017,
      "loss": 2.9322,
      "theoretical_loss": 3.6174306903482076,
      "tokens_seen": 1097072640
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003371715145436309,
      "loss": 2.8824,
      "theoretical_loss": 3.617410476696908,
      "tokens_seen": 1097138176
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033716148445336007,
      "loss": 2.8694,
      "theoretical_loss": 3.617390264591065,
      "tokens_seen": 1097203712
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033715145436308925,
      "loss": 3.0574,
      "theoretical_loss": 3.617370054030468,
      "tokens_seen": 1097269248
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003371414242728185,
      "loss": 2.7327,
      "theoretical_loss": 3.617349845014906,
      "tokens_seen": 1097334784
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003371313941825476,
      "loss": 3.0921,
      "theoretical_loss": 3.6173296375441697,
      "tokens_seen": 1097400320
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033712136409227685,
      "loss": 2.9021,
      "theoretical_loss": 3.6173094316180476,
      "tokens_seen": 1097465856
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.000337111334002006,
      "loss": 2.8803,
      "theoretical_loss": 3.6172892272363306,
      "tokens_seen": 1097531392
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003371013039117352,
      "loss": 3.0505,
      "theoretical_loss": 3.617269024398807,
      "tokens_seen": 1097596928
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003370912738214644,
      "loss": 2.8932,
      "theoretical_loss": 3.6172488231052675,
      "tokens_seen": 1097662464
    },
    {
      "epoch": 0.33,
      "objective/train/advantage_avg": 0.49657732248306274,
      "objective/train/docs_used": 625328,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.002473831176758,
      "objective/train/original_loss": 3.0024735927581787,
      "objective/train/theoretical_loss": 3.617228623355502,
      "objective/train/tokens_used": 1118188000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24662140011787415,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0509114265441895,
      "objective/train/weighted_lm_loss": 3.1552772521972656,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0467102527618408,
      "theoretical_loss": 3.617228623355502,
      "tokens_seen": 1097728000
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003370812437311936,
      "loss": 3.0305,
      "theoretical_loss": 3.617228623355502,
      "tokens_seen": 1097728000
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003370712136409228,
      "loss": 3.0302,
      "theoretical_loss": 3.6172084251493004,
      "tokens_seen": 1097793536
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033706118355065194,
      "loss": 2.8145,
      "theoretical_loss": 3.6171882284864525,
      "tokens_seen": 1097859072
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033705115346038117,
      "loss": 2.7007,
      "theoretical_loss": 3.617168033366748,
      "tokens_seen": 1097924608
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033704112337011035,
      "loss": 2.9125,
      "theoretical_loss": 3.617147839789977,
      "tokens_seen": 1097990144
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033703109327983953,
      "loss": 2.7832,
      "theoretical_loss": 3.6171276477559293,
      "tokens_seen": 1098055680
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003370210631895687,
      "loss": 3.0728,
      "theoretical_loss": 3.6171074572643955,
      "tokens_seen": 1098121216
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033701103309929795,
      "loss": 2.9469,
      "theoretical_loss": 3.6170872683151654,
      "tokens_seen": 1098186752
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003370010030090271,
      "loss": 2.9128,
      "theoretical_loss": 3.6170670809080296,
      "tokens_seen": 1098252288
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003369909729187563,
      "loss": 3.0203,
      "theoretical_loss": 3.617046895042778,
      "tokens_seen": 1098317824
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033698094282848544,
      "loss": 2.862,
      "theoretical_loss": 3.6170267107192,
      "tokens_seen": 1098383360
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003369709127382147,
      "loss": 3.0904,
      "theoretical_loss": 3.617006527937087,
      "tokens_seen": 1098448896
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033696088264794386,
      "loss": 2.8997,
      "theoretical_loss": 3.6169863466962293,
      "tokens_seen": 1098514432
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033695085255767304,
      "loss": 2.928,
      "theoretical_loss": 3.6169661669964164,
      "tokens_seen": 1098579968
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003369408224674022,
      "loss": 2.8325,
      "theoretical_loss": 3.61694598883744,
      "tokens_seen": 1098645504
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003369307923771314,
      "loss": 2.9524,
      "theoretical_loss": 3.6169258122190895,
      "tokens_seen": 1098711040
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003369207622868606,
      "loss": 2.8629,
      "theoretical_loss": 3.6169056371411554,
      "tokens_seen": 1098776576
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003369107321965898,
      "loss": 3.1414,
      "theoretical_loss": 3.6168854636034284,
      "tokens_seen": 1098842112
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033690070210631894,
      "loss": 2.9053,
      "theoretical_loss": 3.6168652916056994,
      "tokens_seen": 1098907648
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003368906720160482,
      "loss": 3.1744,
      "theoretical_loss": 3.616845121147759,
      "tokens_seen": 1098973184
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003368806419257773,
      "loss": 3.107,
      "theoretical_loss": 3.6168249522293974,
      "tokens_seen": 1099038720
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033687061183550654,
      "loss": 2.9562,
      "theoretical_loss": 3.6168047848504052,
      "tokens_seen": 1099104256
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003368605817452357,
      "loss": 2.7288,
      "theoretical_loss": 3.6167846190105744,
      "tokens_seen": 1099169792
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003368505516549649,
      "loss": 2.9697,
      "theoretical_loss": 3.616764454709694,
      "tokens_seen": 1099235328
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003368405215646941,
      "loss": 2.7465,
      "theoretical_loss": 3.6167442919475556,
      "tokens_seen": 1099300864
    },
    {
      "epoch": 0.33,
      "objective/train/advantage_avg": 0.49405384063720703,
      "objective/train/docs_used": 626712,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.5394864082336426,
      "objective/train/original_loss": 2.5394864082336426,
      "objective/train/theoretical_loss": 3.616724130723951,
      "objective/train/tokens_used": 1119826400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24416863918304443,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506465435028076,
      "objective/train/weighted_lm_loss": 2.668053150177002,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 3.616724130723951,
      "tokens_seen": 1099366400
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003368304914744233,
      "loss": 2.9128,
      "theoretical_loss": 3.616724130723951,
      "tokens_seen": 1099366400
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033682046138415245,
      "loss": 2.9548,
      "theoretical_loss": 3.6167039710386693,
      "tokens_seen": 1099431936
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003368104312938817,
      "loss": 2.8721,
      "theoretical_loss": 3.6166838128915026,
      "tokens_seen": 1099497472
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003368004012036108,
      "loss": 2.9019,
      "theoretical_loss": 3.616663656282242,
      "tokens_seen": 1099563008
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033679037111334004,
      "loss": 3.0352,
      "theoretical_loss": 3.6166435012106777,
      "tokens_seen": 1099628544
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003367803410230692,
      "loss": 3.038,
      "theoretical_loss": 3.616623347676601,
      "tokens_seen": 1099694080
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003367703109327984,
      "loss": 3.0641,
      "theoretical_loss": 3.6166031956798035,
      "tokens_seen": 1099759616
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003367602808425276,
      "loss": 3.0959,
      "theoretical_loss": 3.616583045220076,
      "tokens_seen": 1099825152
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033675025075225677,
      "loss": 2.7999,
      "theoretical_loss": 3.61656289629721,
      "tokens_seen": 1099890688
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033674022066198595,
      "loss": 3.0168,
      "theoretical_loss": 3.6165427489109963,
      "tokens_seen": 1099956224
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003367301905717152,
      "loss": 2.9024,
      "theoretical_loss": 3.6165226030612256,
      "tokens_seen": 1100021760
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003367201604814443,
      "loss": 2.8717,
      "theoretical_loss": 3.6165024587476906,
      "tokens_seen": 1100087296
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033671013039117355,
      "loss": 2.5805,
      "theoretical_loss": 3.616482315970182,
      "tokens_seen": 1100152832
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033670010030090273,
      "loss": 3.0218,
      "theoretical_loss": 3.6164621747284906,
      "tokens_seen": 1100218368
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003366900702106319,
      "loss": 2.8675,
      "theoretical_loss": 3.616442035022409,
      "tokens_seen": 1100283904
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003366800401203611,
      "loss": 2.9444,
      "theoretical_loss": 3.616421896851728,
      "tokens_seen": 1100349440
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033667001003009027,
      "loss": 2.9223,
      "theoretical_loss": 3.616401760216238,
      "tokens_seen": 1100414976
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033665997993981945,
      "loss": 2.9656,
      "theoretical_loss": 3.616381625115733,
      "tokens_seen": 1100480512
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003366499498495487,
      "loss": 2.8581,
      "theoretical_loss": 3.616361491550003,
      "tokens_seen": 1100546048
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003366399197592778,
      "loss": 3.0543,
      "theoretical_loss": 3.61634135951884,
      "tokens_seen": 1100611584
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033662988966900705,
      "loss": 2.9344,
      "theoretical_loss": 3.616321229022035,
      "tokens_seen": 1100677120
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003366198595787362,
      "loss": 2.9601,
      "theoretical_loss": 3.6163011000593803,
      "tokens_seen": 1100742656
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003366098294884654,
      "loss": 3.1119,
      "theoretical_loss": 3.616280972630668,
      "tokens_seen": 1100808192
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003365997993981946,
      "loss": 2.9199,
      "theoretical_loss": 3.6162608467356896,
      "tokens_seen": 1100873728
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003365897693079238,
      "loss": 2.9695,
      "theoretical_loss": 3.6162407223742368,
      "tokens_seen": 1100939264
    },
    {
      "epoch": 0.33,
      "objective/train/advantage_avg": 0.495417982339859,
      "objective/train/docs_used": 627263,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.261798858642578,
      "objective/train/original_loss": 3.2617993354797363,
      "objective/train/theoretical_loss": 3.616220599546101,
      "objective/train/tokens_used": 1121464800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24549810588359833,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507898330688477,
      "objective/train/weighted_lm_loss": 3.4275171756744385,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 3.616220599546101,
      "tokens_seen": 1101004800
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033657973921765295,
      "loss": 3.0582,
      "theoretical_loss": 3.616220599546101,
      "tokens_seen": 1101004800
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033656970912738214,
      "loss": 3.1261,
      "theoretical_loss": 3.6162004782510753,
      "tokens_seen": 1101070336
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003365596790371113,
      "loss": 3.0415,
      "theoretical_loss": 3.616180358488951,
      "tokens_seen": 1101135872
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033654964894684055,
      "loss": 3.2878,
      "theoretical_loss": 3.6161602402595197,
      "tokens_seen": 1101201408
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003365396188565697,
      "loss": 3.0635,
      "theoretical_loss": 3.616140123562574,
      "tokens_seen": 1101266944
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003365295887662989,
      "loss": 2.8652,
      "theoretical_loss": 3.616120008397906,
      "tokens_seen": 1101332480
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003365195586760281,
      "loss": 3.1579,
      "theoretical_loss": 3.6160998947653074,
      "tokens_seen": 1101398016
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003365095285857573,
      "loss": 3.0047,
      "theoretical_loss": 3.616079782664571,
      "tokens_seen": 1101463552
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033649949849548646,
      "loss": 2.8581,
      "theoretical_loss": 3.6160596720954885,
      "tokens_seen": 1101529088
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033648946840521564,
      "loss": 2.7942,
      "theoretical_loss": 3.6160395630578517,
      "tokens_seen": 1101594624
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003364794383149448,
      "loss": 3.1661,
      "theoretical_loss": 3.616019455551454,
      "tokens_seen": 1101660160
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033646940822467406,
      "loss": 3.0302,
      "theoretical_loss": 3.6159993495760867,
      "tokens_seen": 1101725696
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003364593781344032,
      "loss": 2.9093,
      "theoretical_loss": 3.615979245131543,
      "tokens_seen": 1101791232
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003364493480441324,
      "loss": 3.0928,
      "theoretical_loss": 3.6159591422176147,
      "tokens_seen": 1101856768
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033643931795386154,
      "loss": 2.786,
      "theoretical_loss": 3.6159390408340943,
      "tokens_seen": 1101922304
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003364292878635908,
      "loss": 2.7357,
      "theoretical_loss": 3.615918940980775,
      "tokens_seen": 1101987840
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033641925777331996,
      "loss": 2.8365,
      "theoretical_loss": 3.615898842657448,
      "tokens_seen": 1102053376
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033640922768304914,
      "loss": 2.9827,
      "theoretical_loss": 3.615878745863907,
      "tokens_seen": 1102118912
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003363991975927783,
      "loss": 3.1447,
      "theoretical_loss": 3.615858650599944,
      "tokens_seen": 1102184448
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003363891675025075,
      "loss": 3.2728,
      "theoretical_loss": 3.615838556865352,
      "tokens_seen": 1102249984
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003363791374122367,
      "loss": 2.6247,
      "theoretical_loss": 3.615818464659924,
      "tokens_seen": 1102315520
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003363691073219659,
      "loss": 2.8802,
      "theoretical_loss": 3.615798373983451,
      "tokens_seen": 1102381056
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033635907723169505,
      "loss": 2.6476,
      "theoretical_loss": 3.615778284835728,
      "tokens_seen": 1102446592
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003363490471414243,
      "loss": 2.7121,
      "theoretical_loss": 3.6157581972165467,
      "tokens_seen": 1102512128
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033633901705115346,
      "loss": 2.8856,
      "theoretical_loss": 3.6157381111257,
      "tokens_seen": 1102577664
    },
    {
      "epoch": 0.33,
      "objective/train/advantage_avg": 0.4939133822917938,
      "objective/train/docs_used": 628470,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.6915979385375977,
      "objective/train/original_loss": 2.6915972232818604,
      "objective/train/theoretical_loss": 3.6157180265629814,
      "objective/train/tokens_used": 1123103200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24401219189167023,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506317615509033,
      "objective/train/weighted_lm_loss": 2.8278520107269287,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0442860126495361,
      "theoretical_loss": 3.6157180265629814,
      "tokens_seen": 1102643200
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033632898696088265,
      "loss": 2.8504,
      "theoretical_loss": 3.6157180265629814,
      "tokens_seen": 1102643200
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003363189568706119,
      "loss": 2.8088,
      "theoretical_loss": 3.6156979435281826,
      "tokens_seen": 1102708736
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.000336308926780341,
      "loss": 2.7395,
      "theoretical_loss": 3.6156778620210974,
      "tokens_seen": 1102774272
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033629889669007024,
      "loss": 2.8017,
      "theoretical_loss": 3.6156577820415188,
      "tokens_seen": 1102839808
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003362888665997994,
      "loss": 2.8641,
      "theoretical_loss": 3.6156377035892397,
      "tokens_seen": 1102905344
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003362788365095286,
      "loss": 2.8547,
      "theoretical_loss": 3.6156176266640534,
      "tokens_seen": 1102970880
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003362688064192578,
      "loss": 2.9005,
      "theoretical_loss": 3.6155975512657523,
      "tokens_seen": 1103036416
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033625877632898697,
      "loss": 3.04,
      "theoretical_loss": 3.6155774773941305,
      "tokens_seen": 1103101952
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033624874623871615,
      "loss": 2.7004,
      "theoretical_loss": 3.615557405048981,
      "tokens_seen": 1103167488
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003362387161484454,
      "loss": 2.9081,
      "theoretical_loss": 3.6155373342300967,
      "tokens_seen": 1103233024
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003362286860581745,
      "loss": 2.6969,
      "theoretical_loss": 3.615517264937271,
      "tokens_seen": 1103298560
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033621865596790375,
      "loss": 2.7695,
      "theoretical_loss": 3.6154971971702974,
      "tokens_seen": 1103364096
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033620862587763293,
      "loss": 2.9412,
      "theoretical_loss": 3.615477130928969,
      "tokens_seen": 1103429632
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003361985957873621,
      "loss": 2.9347,
      "theoretical_loss": 3.6154570662130796,
      "tokens_seen": 1103495168
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003361885656970913,
      "loss": 2.7747,
      "theoretical_loss": 3.6154370030224223,
      "tokens_seen": 1103560704
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033617853560682047,
      "loss": 2.8688,
      "theoretical_loss": 3.6154169413567905,
      "tokens_seen": 1103626240
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033616850551654965,
      "loss": 3.0339,
      "theoretical_loss": 3.615396881215978,
      "tokens_seen": 1103691776
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003361584754262789,
      "loss": 2.8788,
      "theoretical_loss": 3.615376822599779,
      "tokens_seen": 1103757312
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.000336148445336008,
      "loss": 2.8927,
      "theoretical_loss": 3.6153567655079852,
      "tokens_seen": 1103822848
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033613841524573725,
      "loss": 2.8848,
      "theoretical_loss": 3.6153367099403924,
      "tokens_seen": 1103888384
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003361283851554664,
      "loss": 2.8245,
      "theoretical_loss": 3.615316655896793,
      "tokens_seen": 1103953920
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003361183550651956,
      "loss": 2.9783,
      "theoretical_loss": 3.615296603376981,
      "tokens_seen": 1104019456
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003361083249749248,
      "loss": 3.029,
      "theoretical_loss": 3.6152765523807506,
      "tokens_seen": 1104084992
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.000336098294884654,
      "loss": 3.0236,
      "theoretical_loss": 3.615256502907896,
      "tokens_seen": 1104150528
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033608826479438316,
      "loss": 2.9575,
      "theoretical_loss": 3.6152364549582092,
      "tokens_seen": 1104216064
    },
    {
      "epoch": 0.33,
      "objective/train/advantage_avg": 0.4960314631462097,
      "objective/train/docs_used": 629051,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.8878097534179688,
      "objective/train/original_loss": 2.887810230255127,
      "objective/train/theoretical_loss": 3.6152164085314853,
      "objective/train/tokens_used": 1124741600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2460937798023224,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0508543252944946,
      "objective/train/weighted_lm_loss": 3.0346784591674805,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0408108234405518,
      "theoretical_loss": 3.6152164085314853,
      "tokens_seen": 1104281600
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033607823470411234,
      "loss": 2.7968,
      "theoretical_loss": 3.6152164085314853,
      "tokens_seen": 1104281600
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003360682046138415,
      "loss": 3.0357,
      "theoretical_loss": 3.615196363627519,
      "tokens_seen": 1104347136
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033605817452357075,
      "loss": 2.8847,
      "theoretical_loss": 3.615176320246103,
      "tokens_seen": 1104412672
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003360481444332999,
      "loss": 3.073,
      "theoretical_loss": 3.6151562783870315,
      "tokens_seen": 1104478208
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003360381143430291,
      "loss": 2.8993,
      "theoretical_loss": 3.615136238050099,
      "tokens_seen": 1104543744
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003360280842527583,
      "loss": 2.9863,
      "theoretical_loss": 3.6151161992350995,
      "tokens_seen": 1104609280
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003360180541624875,
      "loss": 2.9866,
      "theoretical_loss": 3.615096161941827,
      "tokens_seen": 1104674816
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033600802407221666,
      "loss": 2.938,
      "theoretical_loss": 3.6150761261700755,
      "tokens_seen": 1104740352
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033599799398194584,
      "loss": 2.96,
      "theoretical_loss": 3.61505609191964,
      "tokens_seen": 1104805888
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.000335987963891675,
      "loss": 3.0713,
      "theoretical_loss": 3.6150360591903135,
      "tokens_seen": 1104871424
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033597793380140426,
      "loss": 2.8815,
      "theoretical_loss": 3.6150160279818913,
      "tokens_seen": 1104936960
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003359679037111334,
      "loss": 2.8582,
      "theoretical_loss": 3.6149959982941677,
      "tokens_seen": 1105002496
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003359578736208626,
      "loss": 2.9623,
      "theoretical_loss": 3.6149759701269364,
      "tokens_seen": 1105068032
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033594784353059174,
      "loss": 3.1634,
      "theoretical_loss": 3.614955943479992,
      "tokens_seen": 1105133568
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.000335937813440321,
      "loss": 3.1804,
      "theoretical_loss": 3.6149359183531296,
      "tokens_seen": 1105199104
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033592778335005016,
      "loss": 3.0488,
      "theoretical_loss": 3.614915894746143,
      "tokens_seen": 1105264640
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.00033591775325977934,
      "loss": 2.7237,
      "theoretical_loss": 3.614895872658827,
      "tokens_seen": 1105330176
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003359077231695085,
      "loss": 2.8538,
      "theoretical_loss": 3.6148758520909765,
      "tokens_seen": 1105395712
    },
    {
      "epoch": 0.33,
      "learning_rate": 0.0003358976930792377,
      "loss": 3.1429,
      "theoretical_loss": 3.6148558330423852,
      "tokens_seen": 1105461248
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003358876629889669,
      "loss": 3.028,
      "theoretical_loss": 3.6148358155128486,
      "tokens_seen": 1105526784
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003358776328986961,
      "loss": 2.9089,
      "theoretical_loss": 3.6148157995021615,
      "tokens_seen": 1105592320
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033586760280842525,
      "loss": 2.9364,
      "theoretical_loss": 3.614795785010118,
      "tokens_seen": 1105657856
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003358575727181545,
      "loss": 2.9151,
      "theoretical_loss": 3.614775772036513,
      "tokens_seen": 1105723392
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033584754262788366,
      "loss": 2.9721,
      "theoretical_loss": 3.6147557605811413,
      "tokens_seen": 1105788928
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033583751253761285,
      "loss": 2.9691,
      "theoretical_loss": 3.6147357506437983,
      "tokens_seen": 1105854464
    },
    {
      "epoch": 0.34,
      "objective/train/advantage_avg": 0.49149277806282043,
      "objective/train/docs_used": 630354,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.838923931121826,
      "objective/train/original_loss": 2.838923931121826,
      "objective/train/theoretical_loss": 3.614715742224278,
      "objective/train/tokens_used": 1126380000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2416970133781433,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0503777265548706,
      "objective/train/weighted_lm_loss": 2.982062578201294,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0253151655197144,
      "theoretical_loss": 3.614715742224278,
      "tokens_seen": 1105920000
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.000335827482447342,
      "loss": 3.1094,
      "theoretical_loss": 3.614715742224278,
      "tokens_seen": 1105920000
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003358174523570712,
      "loss": 3.0273,
      "theoretical_loss": 3.614695735322376,
      "tokens_seen": 1105985536
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003358074222668004,
      "loss": 3.005,
      "theoretical_loss": 3.614675729937887,
      "tokens_seen": 1106051072
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003357973921765296,
      "loss": 2.9933,
      "theoretical_loss": 3.6146557260706063,
      "tokens_seen": 1106116608
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033578736208625875,
      "loss": 2.9255,
      "theoretical_loss": 3.614635723720329,
      "tokens_seen": 1106182144
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.000335777331995988,
      "loss": 2.8886,
      "theoretical_loss": 3.614615722886849,
      "tokens_seen": 1106247680
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003357673019057171,
      "loss": 2.9633,
      "theoretical_loss": 3.6145957235699635,
      "tokens_seen": 1106313216
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033575727181544635,
      "loss": 2.8556,
      "theoretical_loss": 3.614575725769466,
      "tokens_seen": 1106378752
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033574724172517553,
      "loss": 2.9623,
      "theoretical_loss": 3.6145557294851525,
      "tokens_seen": 1106444288
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003357372116349047,
      "loss": 2.9589,
      "theoretical_loss": 3.6145357347168177,
      "tokens_seen": 1106509824
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003357271815446339,
      "loss": 2.6107,
      "theoretical_loss": 3.614515741464257,
      "tokens_seen": 1106575360
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033571715145436313,
      "loss": 3.0493,
      "theoretical_loss": 3.6144957497272667,
      "tokens_seen": 1106640896
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033570712136409225,
      "loss": 3.0252,
      "theoretical_loss": 3.614475759505641,
      "tokens_seen": 1106706432
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003356970912738215,
      "loss": 2.7041,
      "theoretical_loss": 3.6144557707991756,
      "tokens_seen": 1106771968
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003356870611835506,
      "loss": 3.0106,
      "theoretical_loss": 3.614435783607666,
      "tokens_seen": 1106837504
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033567703109327985,
      "loss": 2.9627,
      "theoretical_loss": 3.614415797930908,
      "tokens_seen": 1106903040
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033566700100300903,
      "loss": 2.7053,
      "theoretical_loss": 3.614395813768697,
      "tokens_seen": 1106968576
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003356569709127382,
      "loss": 3.1808,
      "theoretical_loss": 3.614375831120828,
      "tokens_seen": 1107034112
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003356469408224674,
      "loss": 3.0399,
      "theoretical_loss": 3.6143558499870974,
      "tokens_seen": 1107099648
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003356369107321966,
      "loss": 2.797,
      "theoretical_loss": 3.6143358703673005,
      "tokens_seen": 1107165184
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033562688064192576,
      "loss": 2.763,
      "theoretical_loss": 3.6143158922612333,
      "tokens_seen": 1107230720
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.000335616850551655,
      "loss": 2.9398,
      "theoretical_loss": 3.614295915668691,
      "tokens_seen": 1107296256
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003356068204613841,
      "loss": 3.0198,
      "theoretical_loss": 3.6142759405894695,
      "tokens_seen": 1107361792
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033559679037111336,
      "loss": 2.9737,
      "theoretical_loss": 3.6142559670233645,
      "tokens_seen": 1107427328
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003355867602808425,
      "loss": 2.8077,
      "theoretical_loss": 3.6142359949701723,
      "tokens_seen": 1107492864
    },
    {
      "epoch": 0.34,
      "objective/train/advantage_avg": 0.49450618028640747,
      "objective/train/docs_used": 630918,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.619603157043457,
      "objective/train/original_loss": 2.619602918624878,
      "objective/train/theoretical_loss": 3.6142160244296884,
      "objective/train/tokens_used": 1128018400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24459275603294373,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506938695907593,
      "objective/train/weighted_lm_loss": 2.752271890640259,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0417574644088745,
      "theoretical_loss": 3.6142160244296884,
      "tokens_seen": 1107558400
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003355767301905717,
      "loss": 3.0589,
      "theoretical_loss": 3.6142160244296884,
      "tokens_seen": 1107558400
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033556670010030095,
      "loss": 2.9237,
      "theoretical_loss": 3.614196055401709,
      "tokens_seen": 1107623936
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003355566700100301,
      "loss": 3.0132,
      "theoretical_loss": 3.61417608788603,
      "tokens_seen": 1107689472
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003355466399197593,
      "loss": 3.0772,
      "theoretical_loss": 3.6141561218824476,
      "tokens_seen": 1107755008
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003355366098294885,
      "loss": 2.9186,
      "theoretical_loss": 3.6141361573907567,
      "tokens_seen": 1107820544
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003355265797392177,
      "loss": 2.9498,
      "theoretical_loss": 3.614116194410755,
      "tokens_seen": 1107886080
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033551654964894686,
      "loss": 3.0139,
      "theoretical_loss": 3.614096232942238,
      "tokens_seen": 1107951616
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033550651955867604,
      "loss": 2.7827,
      "theoretical_loss": 3.6140762729850016,
      "tokens_seen": 1108017152
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003354964894684052,
      "loss": 2.94,
      "theoretical_loss": 3.614056314538842,
      "tokens_seen": 1108082688
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033548645937813446,
      "loss": 2.9748,
      "theoretical_loss": 3.6140363576035552,
      "tokens_seen": 1108148224
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003354764292878636,
      "loss": 3.0414,
      "theoretical_loss": 3.6140164021789385,
      "tokens_seen": 1108213760
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003354663991975928,
      "loss": 3.0482,
      "theoretical_loss": 3.6139964482647873,
      "tokens_seen": 1108279296
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033545636910732195,
      "loss": 2.9561,
      "theoretical_loss": 3.613976495860898,
      "tokens_seen": 1108344832
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003354463390170512,
      "loss": 2.7132,
      "theoretical_loss": 3.6139565449670683,
      "tokens_seen": 1108410368
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033543630892678036,
      "loss": 2.994,
      "theoretical_loss": 3.613936595583093,
      "tokens_seen": 1108475904
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033542627883650954,
      "loss": 2.8984,
      "theoretical_loss": 3.613916647708769,
      "tokens_seen": 1108541440
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003354162487462387,
      "loss": 3.0144,
      "theoretical_loss": 3.613896701343893,
      "tokens_seen": 1108606976
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003354062186559679,
      "loss": 3.1177,
      "theoretical_loss": 3.6138767564882617,
      "tokens_seen": 1108672512
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003353961885656971,
      "loss": 2.8607,
      "theoretical_loss": 3.613856813141672,
      "tokens_seen": 1108738048
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003353861584754263,
      "loss": 2.9521,
      "theoretical_loss": 3.6138368713039197,
      "tokens_seen": 1108803584
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033537612838515545,
      "loss": 2.9423,
      "theoretical_loss": 3.613816930974802,
      "tokens_seen": 1108869120
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003353660982948847,
      "loss": 2.8207,
      "theoretical_loss": 3.613796992154115,
      "tokens_seen": 1108934656
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033535606820461386,
      "loss": 2.8056,
      "theoretical_loss": 3.6137770548416563,
      "tokens_seen": 1109000192
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033534603811434305,
      "loss": 2.8746,
      "theoretical_loss": 3.6137571190372224,
      "tokens_seen": 1109065728
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003353360080240722,
      "loss": 2.8058,
      "theoretical_loss": 3.613737184740611,
      "tokens_seen": 1109131264
    },
    {
      "epoch": 0.34,
      "objective/train/advantage_avg": 0.495446115732193,
      "objective/train/docs_used": 632203,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.913756847381592,
      "objective/train/original_loss": 2.9137563705444336,
      "objective/train/theoretical_loss": 3.6137172519516163,
      "objective/train/tokens_used": 1129656800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24553903937339783,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507928133010864,
      "objective/train/weighted_lm_loss": 3.0620152950286865,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0304546356201172,
      "theoretical_loss": 3.6137172519516163,
      "tokens_seen": 1109196800
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003353259779338014,
      "loss": 2.9741,
      "theoretical_loss": 3.6137172519516163,
      "tokens_seen": 1109196800
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003353159478435306,
      "loss": 2.9369,
      "theoretical_loss": 3.6136973206700382,
      "tokens_seen": 1109262336
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003353059177532598,
      "loss": 3.1252,
      "theoretical_loss": 3.6136773908956723,
      "tokens_seen": 1109327872
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033529588766298895,
      "loss": 2.7837,
      "theoretical_loss": 3.613657462628315,
      "tokens_seen": 1109393408
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003352858575727182,
      "loss": 2.8228,
      "theoretical_loss": 3.613637535867765,
      "tokens_seen": 1109458944
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003352758274824473,
      "loss": 2.8976,
      "theoretical_loss": 3.613617610613818,
      "tokens_seen": 1109524480
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033526579739217655,
      "loss": 2.9538,
      "theoretical_loss": 3.613597686866271,
      "tokens_seen": 1109590016
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033525576730190573,
      "loss": 3.1448,
      "theoretical_loss": 3.6135777646249223,
      "tokens_seen": 1109655552
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003352457372116349,
      "loss": 2.852,
      "theoretical_loss": 3.613557843889568,
      "tokens_seen": 1109721088
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003352357071213641,
      "loss": 2.9896,
      "theoretical_loss": 3.6135379246600063,
      "tokens_seen": 1109786624
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033522567703109333,
      "loss": 3.0991,
      "theoretical_loss": 3.613518006936034,
      "tokens_seen": 1109852160
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033521564694082245,
      "loss": 2.8979,
      "theoretical_loss": 3.613498090717448,
      "tokens_seen": 1109917696
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003352056168505517,
      "loss": 3.0413,
      "theoretical_loss": 3.613478176004046,
      "tokens_seen": 1109983232
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003351955867602808,
      "loss": 3.0443,
      "theoretical_loss": 3.6134582627956258,
      "tokens_seen": 1110048768
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033518555667001005,
      "loss": 2.8003,
      "theoretical_loss": 3.6134383510919843,
      "tokens_seen": 1110114304
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033517552657973923,
      "loss": 2.8461,
      "theoretical_loss": 3.613418440892919,
      "tokens_seen": 1110179840
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003351654964894684,
      "loss": 2.8595,
      "theoretical_loss": 3.6133985321982274,
      "tokens_seen": 1110245376
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003351554663991976,
      "loss": 3.0823,
      "theoretical_loss": 3.613378625007707,
      "tokens_seen": 1110310912
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003351454363089268,
      "loss": 2.8521,
      "theoretical_loss": 3.613358719321156,
      "tokens_seen": 1110376448
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033513540621865596,
      "loss": 3.0626,
      "theoretical_loss": 3.613338815138371,
      "tokens_seen": 1110441984
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003351253761283852,
      "loss": 2.9406,
      "theoretical_loss": 3.6133189124591505,
      "tokens_seen": 1110507520
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003351153460381143,
      "loss": 2.9452,
      "theoretical_loss": 3.613299011283292,
      "tokens_seen": 1110573056
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033510531594784356,
      "loss": 2.8983,
      "theoretical_loss": 3.613279111610593,
      "tokens_seen": 1110638592
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003350952858575727,
      "loss": 2.9984,
      "theoretical_loss": 3.613259213440851,
      "tokens_seen": 1110704128
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003350852557673019,
      "loss": 2.6925,
      "theoretical_loss": 3.613239316773865,
      "tokens_seen": 1110769664
    },
    {
      "epoch": 0.34,
      "objective/train/advantage_avg": 0.4953152537345886,
      "objective/train/docs_used": 632721,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.328181505203247,
      "objective/train/original_loss": 2.328181266784668,
      "objective/train/theoretical_loss": 3.6132194216094313,
      "objective/train/tokens_used": 1131295200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24541977047920227,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050779104232788,
      "objective/train/weighted_lm_loss": 2.446483850479126,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.044721245765686,
      "theoretical_loss": 3.6132194216094313,
      "tokens_seen": 1110835200
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003350752256770311,
      "loss": 2.8528,
      "theoretical_loss": 3.6132194216094313,
      "tokens_seen": 1110835200
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003350651955867603,
      "loss": 2.9636,
      "theoretical_loss": 3.6131995279473488,
      "tokens_seen": 1110900736
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033505516549648946,
      "loss": 3.2235,
      "theoretical_loss": 3.6131796357874157,
      "tokens_seen": 1110966272
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003350451354062187,
      "loss": 3.054,
      "theoretical_loss": 3.613159745129429,
      "tokens_seen": 1111031808
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003350351053159478,
      "loss": 2.7071,
      "theoretical_loss": 3.6131398559731873,
      "tokens_seen": 1111097344
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033502507522567706,
      "loss": 2.7452,
      "theoretical_loss": 3.6131199683184887,
      "tokens_seen": 1111162880
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003350150451354062,
      "loss": 2.6585,
      "theoretical_loss": 3.613100082165131,
      "tokens_seen": 1111228416
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003350050150451354,
      "loss": 2.7039,
      "theoretical_loss": 3.6130801975129123,
      "tokens_seen": 1111293952
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003349949849548646,
      "loss": 3.0354,
      "theoretical_loss": 3.613060314361631,
      "tokens_seen": 1111359488
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003349849548645938,
      "loss": 2.7827,
      "theoretical_loss": 3.6130404327110854,
      "tokens_seen": 1111425024
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033497492477432296,
      "loss": 3.0903,
      "theoretical_loss": 3.613020552561074,
      "tokens_seen": 1111490560
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033496489468405215,
      "loss": 3.0707,
      "theoretical_loss": 3.613000673911394,
      "tokens_seen": 1111556096
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003349548645937813,
      "loss": 2.8833,
      "theoretical_loss": 3.6129807967618444,
      "tokens_seen": 1111621632
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033494483450351056,
      "loss": 3.0848,
      "theoretical_loss": 3.6129609211122236,
      "tokens_seen": 1111687168
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003349348044132397,
      "loss": 2.7603,
      "theoretical_loss": 3.61294104696233,
      "tokens_seen": 1111752704
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003349247743229689,
      "loss": 3.0975,
      "theoretical_loss": 3.6129211743119622,
      "tokens_seen": 1111818240
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033491474423269805,
      "loss": 2.8449,
      "theoretical_loss": 3.6129013031609185,
      "tokens_seen": 1111883776
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003349047141424273,
      "loss": 3.1295,
      "theoretical_loss": 3.612881433508997,
      "tokens_seen": 1111949312
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033489468405215647,
      "loss": 2.8597,
      "theoretical_loss": 3.612861565355997,
      "tokens_seen": 1112014848
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033488465396188565,
      "loss": 3.03,
      "theoretical_loss": 3.6128416987017165,
      "tokens_seen": 1112080384
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033487462387161483,
      "loss": 2.7448,
      "theoretical_loss": 3.6128218335459543,
      "tokens_seen": 1112145920
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033486459378134406,
      "loss": 2.7136,
      "theoretical_loss": 3.612801969888509,
      "tokens_seen": 1112211456
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003348545636910732,
      "loss": 3.1142,
      "theoretical_loss": 3.6127821077291804,
      "tokens_seen": 1112276992
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003348445336008024,
      "loss": 3.0081,
      "theoretical_loss": 3.6127622470677654,
      "tokens_seen": 1112342528
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003348345035105316,
      "loss": 3.0821,
      "theoretical_loss": 3.6127423879040643,
      "tokens_seen": 1112408064
    },
    {
      "epoch": 0.34,
      "objective/train/advantage_avg": 0.49618011713027954,
      "objective/train/docs_used": 634025,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.7505431175231934,
      "objective/train/original_loss": 2.7505431175231934,
      "objective/train/theoretical_loss": 3.612722530237875,
      "objective/train/tokens_used": 1132933600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.246237650513649,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0508698225021362,
      "objective/train/weighted_lm_loss": 2.8904869556427,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0396549701690674,
      "theoretical_loss": 3.612722530237875,
      "tokens_seen": 1112473600
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003348244734202608,
      "loss": 2.9762,
      "theoretical_loss": 3.612722530237875,
      "tokens_seen": 1112473600
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033481444332999,
      "loss": 2.8689,
      "theoretical_loss": 3.6127026740689967,
      "tokens_seen": 1112539136
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033480441323971915,
      "loss": 2.8201,
      "theoretical_loss": 3.6126828193972287,
      "tokens_seen": 1112604672
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003347943831494484,
      "loss": 3.0169,
      "theoretical_loss": 3.612662966222369,
      "tokens_seen": 1112670208
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003347843530591775,
      "loss": 2.7585,
      "theoretical_loss": 3.612643114544218,
      "tokens_seen": 1112735744
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033477432296890675,
      "loss": 2.6975,
      "theoretical_loss": 3.612623264362573,
      "tokens_seen": 1112801280
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033476429287863593,
      "loss": 3.1896,
      "theoretical_loss": 3.612603415677234,
      "tokens_seen": 1112866816
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003347542627883651,
      "loss": 3.062,
      "theoretical_loss": 3.6125835684880006,
      "tokens_seen": 1112932352
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003347442326980943,
      "loss": 2.6375,
      "theoretical_loss": 3.612563722794671,
      "tokens_seen": 1112997888
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033473420260782353,
      "loss": 2.9808,
      "theoretical_loss": 3.612543878597045,
      "tokens_seen": 1113063424
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033472417251755265,
      "loss": 2.7537,
      "theoretical_loss": 3.6125240358949213,
      "tokens_seen": 1113128960
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003347141424272819,
      "loss": 2.9065,
      "theoretical_loss": 3.6125041946881,
      "tokens_seen": 1113194496
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.000334704112337011,
      "loss": 2.8561,
      "theoretical_loss": 3.612484354976379,
      "tokens_seen": 1113260032
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033469408224674025,
      "loss": 2.8437,
      "theoretical_loss": 3.612464516759559,
      "tokens_seen": 1113325568
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033468405215646943,
      "loss": 2.9402,
      "theoretical_loss": 3.612444680037439,
      "tokens_seen": 1113391104
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003346740220661986,
      "loss": 2.7891,
      "theoretical_loss": 3.6124248448098175,
      "tokens_seen": 1113456640
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003346639919759278,
      "loss": 2.9214,
      "theoretical_loss": 3.612405011076495,
      "tokens_seen": 1113522176
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.000334653961885657,
      "loss": 2.8294,
      "theoretical_loss": 3.612385178837271,
      "tokens_seen": 1113587712
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033464393179538616,
      "loss": 2.9748,
      "theoretical_loss": 3.612365348091945,
      "tokens_seen": 1113653248
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003346339017051154,
      "loss": 3.2485,
      "theoretical_loss": 3.612345518840316,
      "tokens_seen": 1113718784
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003346238716148445,
      "loss": 3.1496,
      "theoretical_loss": 3.6123256910821837,
      "tokens_seen": 1113784320
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033461384152457376,
      "loss": 2.9405,
      "theoretical_loss": 3.612305864817348,
      "tokens_seen": 1113849856
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003346038114343029,
      "loss": 2.7815,
      "theoretical_loss": 3.6122860400456087,
      "tokens_seen": 1113915392
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003345937813440321,
      "loss": 2.8849,
      "theoretical_loss": 3.612266216766765,
      "tokens_seen": 1113980928
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003345837512537613,
      "loss": 2.8915,
      "theoretical_loss": 3.612246394980617,
      "tokens_seen": 1114046464
    },
    {
      "epoch": 0.34,
      "objective/train/advantage_avg": 0.4944785237312317,
      "objective/train/docs_used": 634689,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.172600030899048,
      "objective/train/original_loss": 2.172600269317627,
      "objective/train/theoretical_loss": 3.6122265746869653,
      "objective/train/tokens_used": 1134572000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24460627138614655,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506912469863892,
      "objective/train/weighted_lm_loss": 2.2826924324035645,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0304546356201172,
      "theoretical_loss": 3.6122265746869653,
      "tokens_seen": 1114112000
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003345737211634905,
      "loss": 2.7927,
      "theoretical_loss": 3.6122265746869653,
      "tokens_seen": 1114112000
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033456369107321966,
      "loss": 2.8587,
      "theoretical_loss": 3.612206755885608,
      "tokens_seen": 1114177536
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003345536609829489,
      "loss": 2.9214,
      "theoretical_loss": 3.6121869385763463,
      "tokens_seen": 1114243072
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.000334543630892678,
      "loss": 3.0524,
      "theoretical_loss": 3.6121671227589798,
      "tokens_seen": 1114308608
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033453360080240726,
      "loss": 2.6951,
      "theoretical_loss": 3.6121473084333084,
      "tokens_seen": 1114374144
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003345235707121364,
      "loss": 2.8145,
      "theoretical_loss": 3.6121274955991316,
      "tokens_seen": 1114439680
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003345135406218656,
      "loss": 3.118,
      "theoretical_loss": 3.6121076842562507,
      "tokens_seen": 1114505216
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003345035105315948,
      "loss": 2.8717,
      "theoretical_loss": 3.612087874404464,
      "tokens_seen": 1114570752
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.000334493480441324,
      "loss": 3.2185,
      "theoretical_loss": 3.6120680660435736,
      "tokens_seen": 1114636288
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033448345035105316,
      "loss": 2.9744,
      "theoretical_loss": 3.612048259173378,
      "tokens_seen": 1114701824
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033447342026078235,
      "loss": 2.9844,
      "theoretical_loss": 3.612028453793678,
      "tokens_seen": 1114767360
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003344633901705115,
      "loss": 3.319,
      "theoretical_loss": 3.6120086499042747,
      "tokens_seen": 1114832896
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033445336008024076,
      "loss": 2.6964,
      "theoretical_loss": 3.6119888475049664,
      "tokens_seen": 1114898432
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003344433299899699,
      "loss": 2.9237,
      "theoretical_loss": 3.611969046595555,
      "tokens_seen": 1114963968
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003344332998996991,
      "loss": 3.0933,
      "theoretical_loss": 3.6119492471758403,
      "tokens_seen": 1115029504
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033442326980942825,
      "loss": 3.0325,
      "theoretical_loss": 3.611929449245623,
      "tokens_seen": 1115095040
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003344132397191575,
      "loss": 3.1151,
      "theoretical_loss": 3.6119096528047026,
      "tokens_seen": 1115160576
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033440320962888667,
      "loss": 2.8559,
      "theoretical_loss": 3.611889857852881,
      "tokens_seen": 1115226112
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033439317953861585,
      "loss": 2.702,
      "theoretical_loss": 3.611870064389957,
      "tokens_seen": 1115291648
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033438314944834503,
      "loss": 2.6318,
      "theoretical_loss": 3.611850272415732,
      "tokens_seen": 1115357184
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033437311935807427,
      "loss": 3.0134,
      "theoretical_loss": 3.6118304819300073,
      "tokens_seen": 1115422720
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003343630892678034,
      "loss": 3.0207,
      "theoretical_loss": 3.6118106929325826,
      "tokens_seen": 1115488256
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033435305917753263,
      "loss": 2.8239,
      "theoretical_loss": 3.611790905423259,
      "tokens_seen": 1115553792
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033434302908726175,
      "loss": 3.1385,
      "theoretical_loss": 3.611771119401836,
      "tokens_seen": 1115619328
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.000334332998996991,
      "loss": 3.0483,
      "theoretical_loss": 3.6117513348681163,
      "tokens_seen": 1115684864
    },
    {
      "epoch": 0.34,
      "objective/train/advantage_avg": 0.49426841735839844,
      "objective/train/docs_used": 636068,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.4985594749450684,
      "objective/train/original_loss": 2.4985594749450684,
      "objective/train/theoretical_loss": 3.611731551821899,
      "objective/train/tokens_used": 1136210400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.244376078248024,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506690740585327,
      "objective/train/weighted_lm_loss": 2.624910831451416,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0425469875335693,
      "theoretical_loss": 3.611731551821899,
      "tokens_seen": 1115750400
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033432296890672017,
      "loss": 2.7278,
      "theoretical_loss": 3.611731551821899,
      "tokens_seen": 1115750400
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033431293881644935,
      "loss": 3.1787,
      "theoretical_loss": 3.6117117702629855,
      "tokens_seen": 1115815936
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033430290872617853,
      "loss": 2.811,
      "theoretical_loss": 3.6116919901911766,
      "tokens_seen": 1115881472
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003342928786359077,
      "loss": 3.075,
      "theoretical_loss": 3.6116722116062734,
      "tokens_seen": 1115947008
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003342828485456369,
      "loss": 3.0279,
      "theoretical_loss": 3.611652434508077,
      "tokens_seen": 1116012544
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033427281845536613,
      "loss": 3.0374,
      "theoretical_loss": 3.6116326588963874,
      "tokens_seen": 1116078080
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033426278836509526,
      "loss": 2.8983,
      "theoretical_loss": 3.6116128847710063,
      "tokens_seen": 1116143616
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003342527582748245,
      "loss": 2.8253,
      "theoretical_loss": 3.6115931121317346,
      "tokens_seen": 1116209152
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003342427281845536,
      "loss": 2.7589,
      "theoretical_loss": 3.6115733409783735,
      "tokens_seen": 1116274688
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033423269809428285,
      "loss": 2.8705,
      "theoretical_loss": 3.611553571310724,
      "tokens_seen": 1116340224
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033422266800401204,
      "loss": 3.0805,
      "theoretical_loss": 3.6115338031285873,
      "tokens_seen": 1116405760
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003342126379137412,
      "loss": 2.8855,
      "theoretical_loss": 3.611514036431764,
      "tokens_seen": 1116471296
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003342026078234704,
      "loss": 2.9007,
      "theoretical_loss": 3.6114942712200566,
      "tokens_seen": 1116536832
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033419257773319963,
      "loss": 2.872,
      "theoretical_loss": 3.6114745074932655,
      "tokens_seen": 1116602368
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033418254764292876,
      "loss": 3.0584,
      "theoretical_loss": 3.6114547452511916,
      "tokens_seen": 1116667904
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.000334172517552658,
      "loss": 2.6513,
      "theoretical_loss": 3.611434984493637,
      "tokens_seen": 1116733440
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003341624874623871,
      "loss": 2.9328,
      "theoretical_loss": 3.611415225220403,
      "tokens_seen": 1116798976
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033415245737211636,
      "loss": 3.0207,
      "theoretical_loss": 3.61139546743129,
      "tokens_seen": 1116864512
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033414242728184554,
      "loss": 3.0636,
      "theoretical_loss": 3.611375711126101,
      "tokens_seen": 1116930048
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003341323971915747,
      "loss": 2.8687,
      "theoretical_loss": 3.611355956304637,
      "tokens_seen": 1116995584
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003341223671013039,
      "loss": 2.9252,
      "theoretical_loss": 3.6113362029666987,
      "tokens_seen": 1117061120
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003341123370110331,
      "loss": 2.9929,
      "theoretical_loss": 3.6113164511120885,
      "tokens_seen": 1117126656
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033410230692076226,
      "loss": 2.7476,
      "theoretical_loss": 3.6112967007406076,
      "tokens_seen": 1117192192
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003340922768304915,
      "loss": 2.9924,
      "theoretical_loss": 3.6112769518520578,
      "tokens_seen": 1117257728
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003340822467402207,
      "loss": 2.9662,
      "theoretical_loss": 3.611257204446241,
      "tokens_seen": 1117323264
    },
    {
      "epoch": 0.34,
      "objective/train/advantage_avg": 0.49554890394210815,
      "objective/train/docs_used": 636687,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.7588961124420166,
      "objective/train/original_loss": 2.7588961124420166,
      "objective/train/theoretical_loss": 3.6112374585229583,
      "objective/train/tokens_used": 1137848800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24561628699302673,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050803542137146,
      "objective/train/weighted_lm_loss": 2.899052381515503,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0451053380966187,
      "theoretical_loss": 3.6112374585229583,
      "tokens_seen": 1117388800
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033407221664994986,
      "loss": 3.0675,
      "theoretical_loss": 3.6112374585229583,
      "tokens_seen": 1117388800
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003340621865596791,
      "loss": 2.8162,
      "theoretical_loss": 3.611217714082012,
      "tokens_seen": 1117454336
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003340521564694082,
      "loss": 3.0866,
      "theoretical_loss": 3.6111979711232043,
      "tokens_seen": 1117519872
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033404212637913746,
      "loss": 2.8724,
      "theoretical_loss": 3.611178229646336,
      "tokens_seen": 1117585408
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003340320962888666,
      "loss": 2.9385,
      "theoretical_loss": 3.61115848965121,
      "tokens_seen": 1117650944
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003340220661985958,
      "loss": 2.7707,
      "theoretical_loss": 3.611138751137627,
      "tokens_seen": 1117716480
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.000334012036108325,
      "loss": 3.0193,
      "theoretical_loss": 3.6111190141053893,
      "tokens_seen": 1117782016
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003340020060180542,
      "loss": 2.9801,
      "theoretical_loss": 3.6110992785543,
      "tokens_seen": 1117847552
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033399197592778336,
      "loss": 2.8644,
      "theoretical_loss": 3.61107954448416,
      "tokens_seen": 1117913088
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033398194583751255,
      "loss": 2.9851,
      "theoretical_loss": 3.611059811894772,
      "tokens_seen": 1117978624
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003339719157472417,
      "loss": 3.0221,
      "theoretical_loss": 3.611040080785937,
      "tokens_seen": 1118044160
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033396188565697096,
      "loss": 2.8748,
      "theoretical_loss": 3.6110203511574586,
      "tokens_seen": 1118109696
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003339518555667001,
      "loss": 3.0381,
      "theoretical_loss": 3.611000623009138,
      "tokens_seen": 1118175232
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003339418254764293,
      "loss": 2.7914,
      "theoretical_loss": 3.6109808963407777,
      "tokens_seen": 1118240768
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033393179538615845,
      "loss": 2.9237,
      "theoretical_loss": 3.61096117115218,
      "tokens_seen": 1118306304
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003339217652958877,
      "loss": 2.9242,
      "theoretical_loss": 3.610941447443147,
      "tokens_seen": 1118371840
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033391173520561687,
      "loss": 2.8993,
      "theoretical_loss": 3.6109217252134815,
      "tokens_seen": 1118437376
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033390170511534605,
      "loss": 2.8895,
      "theoretical_loss": 3.6109020044629854,
      "tokens_seen": 1118502912
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033389167502507523,
      "loss": 2.9617,
      "theoretical_loss": 3.610882285191461,
      "tokens_seen": 1118568448
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033388164493480447,
      "loss": 3.0594,
      "theoretical_loss": 3.610862567398711,
      "tokens_seen": 1118633984
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003338716148445336,
      "loss": 3.1514,
      "theoretical_loss": 3.6108428510845383,
      "tokens_seen": 1118699520
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033386158475426283,
      "loss": 2.8924,
      "theoretical_loss": 3.6108231362487446,
      "tokens_seen": 1118765056
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033385155466399195,
      "loss": 2.953,
      "theoretical_loss": 3.6108034228911334,
      "tokens_seen": 1118830592
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003338415245737212,
      "loss": 3.1048,
      "theoretical_loss": 3.610783711011506,
      "tokens_seen": 1118896128
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033383149448345037,
      "loss": 2.7879,
      "theoretical_loss": 3.6107640006096657,
      "tokens_seen": 1118961664
    },
    {
      "epoch": 0.34,
      "objective/train/advantage_avg": 0.49313297867774963,
      "objective/train/docs_used": 638066,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.6653244495391846,
      "objective/train/original_loss": 2.6653239727020264,
      "objective/train/theoretical_loss": 3.6107442916854158,
      "objective/train/tokens_used": 1139487200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24333544075489044,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505502223968506,
      "objective/train/weighted_lm_loss": 2.800039529800415,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0338950157165527,
      "theoretical_loss": 3.6107442916854158,
      "tokens_seen": 1119027200
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033382146439317955,
      "loss": 2.9178,
      "theoretical_loss": 3.6107442916854158,
      "tokens_seen": 1119027200
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033381143430290873,
      "loss": 2.8666,
      "theoretical_loss": 3.6107245842385582,
      "tokens_seen": 1119092736
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003338014042126379,
      "loss": 2.8781,
      "theoretical_loss": 3.610704878268896,
      "tokens_seen": 1119158272
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003337913741223671,
      "loss": 2.8947,
      "theoretical_loss": 3.6106851737762318,
      "tokens_seen": 1119223808
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033378134403209633,
      "loss": 2.8863,
      "theoretical_loss": 3.6106654707603685,
      "tokens_seen": 1119289344
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033377131394182546,
      "loss": 2.8445,
      "theoretical_loss": 3.610645769221109,
      "tokens_seen": 1119354880
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003337612838515547,
      "loss": 3.1021,
      "theoretical_loss": 3.610626069158256,
      "tokens_seen": 1119420416
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003337512537612838,
      "loss": 2.65,
      "theoretical_loss": 3.6106063705716123,
      "tokens_seen": 1119485952
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033374122367101306,
      "loss": 3.0469,
      "theoretical_loss": 3.6105866734609817,
      "tokens_seen": 1119551488
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033373119358074224,
      "loss": 2.9838,
      "theoretical_loss": 3.610566977826166,
      "tokens_seen": 1119617024
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003337211634904714,
      "loss": 2.8016,
      "theoretical_loss": 3.6105472836669694,
      "tokens_seen": 1119682560
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003337111334002006,
      "loss": 2.5364,
      "theoretical_loss": 3.610527590983194,
      "tokens_seen": 1119748096
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033370110330992983,
      "loss": 2.8519,
      "theoretical_loss": 3.6105078997746443,
      "tokens_seen": 1119813632
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033369107321965896,
      "loss": 2.7579,
      "theoretical_loss": 3.6104882100411215,
      "tokens_seen": 1119879168
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003336810431293882,
      "loss": 2.9639,
      "theoretical_loss": 3.6104685217824306,
      "tokens_seen": 1119944704
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003336710130391173,
      "loss": 2.6695,
      "theoretical_loss": 3.6104488349983743,
      "tokens_seen": 1120010240
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033366098294884656,
      "loss": 2.8929,
      "theoretical_loss": 3.610429149688755,
      "tokens_seen": 1120075776
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033365095285857574,
      "loss": 3.0029,
      "theoretical_loss": 3.610409465853377,
      "tokens_seen": 1120141312
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003336409227683049,
      "loss": 3.1508,
      "theoretical_loss": 3.6103897834920433,
      "tokens_seen": 1120206848
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003336308926780341,
      "loss": 3.0699,
      "theoretical_loss": 3.6103701026045574,
      "tokens_seen": 1120272384
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003336208625877633,
      "loss": 3.0313,
      "theoretical_loss": 3.610350423190723,
      "tokens_seen": 1120337920
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033361083249749246,
      "loss": 2.8801,
      "theoretical_loss": 3.6103307452503426,
      "tokens_seen": 1120403456
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003336008024072217,
      "loss": 2.9811,
      "theoretical_loss": 3.6103110687832207,
      "tokens_seen": 1120468992
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003335907723169508,
      "loss": 2.7625,
      "theoretical_loss": 3.6102913937891605,
      "tokens_seen": 1120534528
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033358074222668006,
      "loss": 2.8196,
      "theoretical_loss": 3.6102717202679653,
      "tokens_seen": 1120600064
    },
    {
      "epoch": 0.34,
      "objective/train/advantage_avg": 0.49404722452163696,
      "objective/train/docs_used": 638756,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.641533851623535,
      "objective/train/original_loss": 2.6415340900421143,
      "objective/train/theoretical_loss": 3.6102520482194387,
      "objective/train/tokens_used": 1141125600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2441440373659134,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506457090377808,
      "objective/train/weighted_lm_loss": 2.775209665298462,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0462169647216797,
      "theoretical_loss": 3.6102520482194387,
      "tokens_seen": 1120665600
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033357071213640924,
      "loss": 2.9195,
      "theoretical_loss": 3.6102520482194387,
      "tokens_seen": 1120665600
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003335606820461384,
      "loss": 3.0759,
      "theoretical_loss": 3.610232377643385,
      "tokens_seen": 1120731136
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003335506519558676,
      "loss": 3.002,
      "theoretical_loss": 3.610212708539607,
      "tokens_seen": 1120796672
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003335406218655968,
      "loss": 2.8822,
      "theoretical_loss": 3.6101930409079097,
      "tokens_seen": 1120862208
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033353059177532597,
      "loss": 2.926,
      "theoretical_loss": 3.6101733747480957,
      "tokens_seen": 1120927744
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003335205616850552,
      "loss": 2.9737,
      "theoretical_loss": 3.6101537100599694,
      "tokens_seen": 1120993280
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033351053159478433,
      "loss": 2.6955,
      "theoretical_loss": 3.610134046843334,
      "tokens_seen": 1121058816
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033350050150451356,
      "loss": 2.653,
      "theoretical_loss": 3.6101143850979938,
      "tokens_seen": 1121124352
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003334904714142427,
      "loss": 2.8731,
      "theoretical_loss": 3.610094724823753,
      "tokens_seen": 1121189888
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003334804413239719,
      "loss": 2.8521,
      "theoretical_loss": 3.610075066020415,
      "tokens_seen": 1121255424
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003334704112337011,
      "loss": 2.8172,
      "theoretical_loss": 3.610055408687784,
      "tokens_seen": 1121320960
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003334603811434303,
      "loss": 3.0704,
      "theoretical_loss": 3.610035752825665,
      "tokens_seen": 1121386496
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033345035105315947,
      "loss": 3.0156,
      "theoretical_loss": 3.61001609843386,
      "tokens_seen": 1121452032
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033344032096288865,
      "loss": 2.7099,
      "theoretical_loss": 3.609996445512175,
      "tokens_seen": 1121517568
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033343029087261783,
      "loss": 3.0622,
      "theoretical_loss": 3.609976794060413,
      "tokens_seen": 1121583104
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033342026078234707,
      "loss": 2.8981,
      "theoretical_loss": 3.6099571440783786,
      "tokens_seen": 1121648640
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003334102306920762,
      "loss": 2.6615,
      "theoretical_loss": 3.609937495565876,
      "tokens_seen": 1121714176
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033340020060180543,
      "loss": 3.0177,
      "theoretical_loss": 3.6099178485227092,
      "tokens_seen": 1121779712
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003333901705115346,
      "loss": 2.785,
      "theoretical_loss": 3.6098982029486835,
      "tokens_seen": 1121845248
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003333801404212638,
      "loss": 2.8038,
      "theoretical_loss": 3.6098785588436018,
      "tokens_seen": 1121910784
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.000333370110330993,
      "loss": 2.9443,
      "theoretical_loss": 3.609858916207269,
      "tokens_seen": 1121976320
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033336008024072215,
      "loss": 3.0844,
      "theoretical_loss": 3.6098392750394903,
      "tokens_seen": 1122041856
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033335005015045134,
      "loss": 2.9052,
      "theoretical_loss": 3.6098196353400693,
      "tokens_seen": 1122107392
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033334002006018057,
      "loss": 3.0481,
      "theoretical_loss": 3.6097999971088104,
      "tokens_seen": 1122172928
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033332998996990975,
      "loss": 2.6404,
      "theoretical_loss": 3.6097803603455185,
      "tokens_seen": 1122238464
    },
    {
      "epoch": 0.34,
      "objective/train/advantage_avg": 0.49538326263427734,
      "objective/train/docs_used": 639983,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.1620912551879883,
      "objective/train/original_loss": 3.16209077835083,
      "objective/train/theoretical_loss": 3.6097607250499975,
      "objective/train/tokens_used": 1142764000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24544395506381989,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050786018371582,
      "objective/train/weighted_lm_loss": 3.322768449783325,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0442860126495361,
      "theoretical_loss": 3.6097607250499975,
      "tokens_seen": 1122304000
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033331995987963893,
      "loss": 3.0383,
      "theoretical_loss": 3.6097607250499975,
      "tokens_seen": 1122304000
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003333099297893681,
      "loss": 2.8289,
      "theoretical_loss": 3.609741091222053,
      "tokens_seen": 1122369536
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003332998996990973,
      "loss": 2.8607,
      "theoretical_loss": 3.609721458861489,
      "tokens_seen": 1122435072
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033328986960882653,
      "loss": 3.0609,
      "theoretical_loss": 3.6097018279681103,
      "tokens_seen": 1122500608
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033327983951855566,
      "loss": 2.9022,
      "theoretical_loss": 3.609682198541722,
      "tokens_seen": 1122566144
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003332698094282849,
      "loss": 2.9072,
      "theoretical_loss": 3.609662570582128,
      "tokens_seen": 1122631680
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.000333259779338014,
      "loss": 2.8021,
      "theoretical_loss": 3.6096429440891336,
      "tokens_seen": 1122697216
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033324974924774326,
      "loss": 2.9138,
      "theoretical_loss": 3.6096233190625435,
      "tokens_seen": 1122762752
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033323971915747244,
      "loss": 2.6773,
      "theoretical_loss": 3.6096036955021624,
      "tokens_seen": 1122828288
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003332296890672016,
      "loss": 2.9764,
      "theoretical_loss": 3.6095840734077957,
      "tokens_seen": 1122893824
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003332196589769308,
      "loss": 2.919,
      "theoretical_loss": 3.6095644527792476,
      "tokens_seen": 1122959360
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033320962888666003,
      "loss": 2.6614,
      "theoretical_loss": 3.609544833616324,
      "tokens_seen": 1123024896
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033319959879638916,
      "loss": 2.8271,
      "theoretical_loss": 3.609525215918829,
      "tokens_seen": 1123090432
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003331895687061184,
      "loss": 2.9205,
      "theoretical_loss": 3.6095055996865684,
      "tokens_seen": 1123155968
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003331795386158475,
      "loss": 2.9219,
      "theoretical_loss": 3.6094859849193464,
      "tokens_seen": 1123221504
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033316950852557676,
      "loss": 2.9399,
      "theoretical_loss": 3.6094663716169686,
      "tokens_seen": 1123287040
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033315947843530594,
      "loss": 2.9144,
      "theoretical_loss": 3.6094467597792406,
      "tokens_seen": 1123352576
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003331494483450351,
      "loss": 3.047,
      "theoretical_loss": 3.6094271494059664,
      "tokens_seen": 1123418112
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003331394182547643,
      "loss": 2.9406,
      "theoretical_loss": 3.6094075404969526,
      "tokens_seen": 1123483648
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003331293881644935,
      "loss": 2.8076,
      "theoretical_loss": 3.6093879330520036,
      "tokens_seen": 1123549184
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033311935807422266,
      "loss": 2.4898,
      "theoretical_loss": 3.609368327070925,
      "tokens_seen": 1123614720
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003331093279839519,
      "loss": 2.6361,
      "theoretical_loss": 3.6093487225535217,
      "tokens_seen": 1123680256
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.000333099297893681,
      "loss": 2.7574,
      "theoretical_loss": 3.6093291194995993,
      "tokens_seen": 1123745792
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033308926780341026,
      "loss": 2.9287,
      "theoretical_loss": 3.6093095179089643,
      "tokens_seen": 1123811328
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033307923771313944,
      "loss": 2.8091,
      "theoretical_loss": 3.60928991778142,
      "tokens_seen": 1123876864
    },
    {
      "epoch": 0.34,
      "objective/train/advantage_avg": 0.4958108961582184,
      "objective/train/docs_used": 640598,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.4168853759765625,
      "objective/train/original_loss": 2.4168851375579834,
      "objective/train/theoretical_loss": 3.6092703191167743,
      "objective/train/tokens_used": 1144402400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2458677738904953,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0508309602737427,
      "objective/train/weighted_lm_loss": 2.5396595001220703,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0437887907028198,
      "theoretical_loss": 3.6092703191167743,
      "tokens_seen": 1123942400
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003330692076228686,
      "loss": 2.8181,
      "theoretical_loss": 3.6092703191167743,
      "tokens_seen": 1123942400
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003330591775325978,
      "loss": 2.8051,
      "theoretical_loss": 3.6092507219148304,
      "tokens_seen": 1124007936
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.000333049147442327,
      "loss": 3.0681,
      "theoretical_loss": 3.6092311261753958,
      "tokens_seen": 1124073472
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033303911735205617,
      "loss": 2.8349,
      "theoretical_loss": 3.609211531898275,
      "tokens_seen": 1124139008
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003330290872617854,
      "loss": 3.1234,
      "theoretical_loss": 3.6091919390832734,
      "tokens_seen": 1124204544
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033301905717151453,
      "loss": 2.8517,
      "theoretical_loss": 3.6091723477301976,
      "tokens_seen": 1124270080
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033300902708124376,
      "loss": 2.7463,
      "theoretical_loss": 3.609152757838853,
      "tokens_seen": 1124335616
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003329989969909729,
      "loss": 2.6998,
      "theoretical_loss": 3.6091331694090445,
      "tokens_seen": 1124401152
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003329889669007021,
      "loss": 2.5616,
      "theoretical_loss": 3.609113582440579,
      "tokens_seen": 1124466688
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003329789368104313,
      "loss": 2.9825,
      "theoretical_loss": 3.6090939969332623,
      "tokens_seen": 1124532224
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003329689067201605,
      "loss": 2.8136,
      "theoretical_loss": 3.6090744128868995,
      "tokens_seen": 1124597760
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033295887662988967,
      "loss": 2.8926,
      "theoretical_loss": 3.6090548303012966,
      "tokens_seen": 1124663296
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033294884653961885,
      "loss": 3.081,
      "theoretical_loss": 3.6090352491762605,
      "tokens_seen": 1124728832
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033293881644934803,
      "loss": 3.0123,
      "theoretical_loss": 3.609015669511596,
      "tokens_seen": 1124794368
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033292878635907727,
      "loss": 2.841,
      "theoretical_loss": 3.6089960913071097,
      "tokens_seen": 1124859904
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003329187562688064,
      "loss": 2.6379,
      "theoretical_loss": 3.6089765145626074,
      "tokens_seen": 1124925440
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033290872617853563,
      "loss": 2.999,
      "theoretical_loss": 3.6089569392778955,
      "tokens_seen": 1124990976
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003328986960882648,
      "loss": 2.7889,
      "theoretical_loss": 3.6089373654527797,
      "tokens_seen": 1125056512
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.000332888665997994,
      "loss": 3.0183,
      "theoretical_loss": 3.608917793087066,
      "tokens_seen": 1125122048
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003328786359077232,
      "loss": 2.9659,
      "theoretical_loss": 3.608898222180562,
      "tokens_seen": 1125187584
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033286860581745235,
      "loss": 2.9703,
      "theoretical_loss": 3.6088786527330723,
      "tokens_seen": 1125253120
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033285857572718154,
      "loss": 3.0976,
      "theoretical_loss": 3.6088590847444038,
      "tokens_seen": 1125318656
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033284854563691077,
      "loss": 2.8774,
      "theoretical_loss": 3.6088395182143627,
      "tokens_seen": 1125384192
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003328385155466399,
      "loss": 2.9525,
      "theoretical_loss": 3.6088199531427554,
      "tokens_seen": 1125449728
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033282848545636913,
      "loss": 2.7547,
      "theoretical_loss": 3.608800389529388,
      "tokens_seen": 1125515264
    },
    {
      "epoch": 0.34,
      "objective/train/advantage_avg": 0.49155381321907043,
      "objective/train/docs_used": 641290,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.7015886306762695,
      "objective/train/original_loss": 2.7015886306762695,
      "objective/train/theoretical_loss": 3.6087808273740674,
      "objective/train/tokens_used": 1146040800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24173253774642944,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0503841638565063,
      "objective/train/weighted_lm_loss": 2.837705612182617,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0417574644088745,
      "theoretical_loss": 3.6087808273740674,
      "tokens_seen": 1125580800
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033281845536609826,
      "loss": 2.923,
      "theoretical_loss": 3.6087808273740674,
      "tokens_seen": 1125580800
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003328084252758275,
      "loss": 2.9712,
      "theoretical_loss": 3.6087612666766002,
      "tokens_seen": 1125646336
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003327983951855567,
      "loss": 2.9598,
      "theoretical_loss": 3.608741707436792,
      "tokens_seen": 1125711872
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033278836509528586,
      "loss": 2.7796,
      "theoretical_loss": 3.60872214965445,
      "tokens_seen": 1125777408
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033277833500501504,
      "loss": 3.0172,
      "theoretical_loss": 3.608702593329381,
      "tokens_seen": 1125842944
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003327683049147442,
      "loss": 2.9924,
      "theoretical_loss": 3.6086830384613906,
      "tokens_seen": 1125908480
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003327582748244734,
      "loss": 2.9209,
      "theoretical_loss": 3.608663485050286,
      "tokens_seen": 1125974016
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033274824473420264,
      "loss": 2.827,
      "theoretical_loss": 3.608643933095874,
      "tokens_seen": 1126039552
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033273821464393176,
      "loss": 2.9741,
      "theoretical_loss": 3.608624382597962,
      "tokens_seen": 1126105088
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.000332728184553661,
      "loss": 3.0318,
      "theoretical_loss": 3.608604833556355,
      "tokens_seen": 1126170624
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003327181544633902,
      "loss": 2.775,
      "theoretical_loss": 3.6085852859708614,
      "tokens_seen": 1126236160
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033270812437311936,
      "loss": 2.9482,
      "theoretical_loss": 3.608565739841287,
      "tokens_seen": 1126301696
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033269809428284854,
      "loss": 3.1013,
      "theoretical_loss": 3.608546195167439,
      "tokens_seen": 1126367232
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003326880641925777,
      "loss": 2.9262,
      "theoretical_loss": 3.6085266519491244,
      "tokens_seen": 1126432768
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003326780341023069,
      "loss": 2.9595,
      "theoretical_loss": 3.60850711018615,
      "tokens_seen": 1126498304
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033266800401203614,
      "loss": 2.8066,
      "theoretical_loss": 3.6084875698783225,
      "tokens_seen": 1126563840
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033265797392176527,
      "loss": 2.9026,
      "theoretical_loss": 3.60846803102545,
      "tokens_seen": 1126629376
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003326479438314945,
      "loss": 3.183,
      "theoretical_loss": 3.608448493627338,
      "tokens_seen": 1126694912
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033263791374122363,
      "loss": 2.8795,
      "theoretical_loss": 3.608428957683795,
      "tokens_seen": 1126760448
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033262788365095286,
      "loss": 3.0141,
      "theoretical_loss": 3.608409423194627,
      "tokens_seen": 1126825984
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033261785356068205,
      "loss": 3.0777,
      "theoretical_loss": 3.6083898901596414,
      "tokens_seen": 1126891520
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003326078234704112,
      "loss": 2.9009,
      "theoretical_loss": 3.6083703585786457,
      "tokens_seen": 1126957056
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003325977933801404,
      "loss": 2.5266,
      "theoretical_loss": 3.608350828451447,
      "tokens_seen": 1127022592
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033258776328986964,
      "loss": 3.0571,
      "theoretical_loss": 3.608331299777853,
      "tokens_seen": 1127088128
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003325777331995988,
      "loss": 2.9265,
      "theoretical_loss": 3.60831177255767,
      "tokens_seen": 1127153664
    },
    {
      "epoch": 0.34,
      "objective/train/advantage_avg": 0.4951574504375458,
      "objective/train/docs_used": 642585,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.8781347274780273,
      "objective/train/original_loss": 2.8781347274780273,
      "objective/train/theoretical_loss": 3.6082922467907066,
      "objective/train/tokens_used": 1147679200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2452554553747177,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507625341415405,
      "objective/train/weighted_lm_loss": 3.0242226123809814,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 3.6082922467907066,
      "tokens_seen": 1127219200
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.000332567703109328,
      "loss": 2.8622,
      "theoretical_loss": 3.6082922467907066,
      "tokens_seen": 1127219200
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003325576730190572,
      "loss": 2.8818,
      "theoretical_loss": 3.6082727224767686,
      "tokens_seen": 1127284736
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033254764292878637,
      "loss": 2.9869,
      "theoretical_loss": 3.608253199615665,
      "tokens_seen": 1127350272
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003325376128385156,
      "loss": 2.7244,
      "theoretical_loss": 3.608233678207202,
      "tokens_seen": 1127415808
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033252758274824473,
      "loss": 2.8265,
      "theoretical_loss": 3.608214158251188,
      "tokens_seen": 1127481344
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033251755265797396,
      "loss": 2.7177,
      "theoretical_loss": 3.60819463974743,
      "tokens_seen": 1127546880
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003325075225677031,
      "loss": 3.0186,
      "theoretical_loss": 3.6081751226957355,
      "tokens_seen": 1127612416
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033249749247743233,
      "loss": 2.7211,
      "theoretical_loss": 3.608155607095912,
      "tokens_seen": 1127677952
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003324874623871615,
      "loss": 2.7608,
      "theoretical_loss": 3.6081360929477677,
      "tokens_seen": 1127743488
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003324774322968907,
      "loss": 2.9635,
      "theoretical_loss": 3.60811658025111,
      "tokens_seen": 1127809024
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033246740220661987,
      "loss": 2.9371,
      "theoretical_loss": 3.608097069005747,
      "tokens_seen": 1127874560
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033245737211634905,
      "loss": 2.9268,
      "theoretical_loss": 3.6080775592114858,
      "tokens_seen": 1127940096
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033244734202607823,
      "loss": 2.7979,
      "theoretical_loss": 3.608058050868134,
      "tokens_seen": 1128005632
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033243731193580747,
      "loss": 2.7477,
      "theoretical_loss": 3.6080385439755,
      "tokens_seen": 1128071168
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003324272818455366,
      "loss": 2.9615,
      "theoretical_loss": 3.6080190385333912,
      "tokens_seen": 1128136704
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033241725175526583,
      "loss": 2.6055,
      "theoretical_loss": 3.607999534541616,
      "tokens_seen": 1128202240
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.000332407221664995,
      "loss": 3.149,
      "theoretical_loss": 3.6079800319999817,
      "tokens_seen": 1128267776
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003323971915747242,
      "loss": 2.6785,
      "theoretical_loss": 3.607960530908297,
      "tokens_seen": 1128333312
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003323871614844534,
      "loss": 3.0405,
      "theoretical_loss": 3.6079410312663693,
      "tokens_seen": 1128398848
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033237713139418255,
      "loss": 3.1079,
      "theoretical_loss": 3.6079215330740064,
      "tokens_seen": 1128464384
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033236710130391174,
      "loss": 2.889,
      "theoretical_loss": 3.6079020363310175,
      "tokens_seen": 1128529920
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033235707121364097,
      "loss": 2.9726,
      "theoretical_loss": 3.6078825410372097,
      "tokens_seen": 1128595456
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003323470411233701,
      "loss": 2.6968,
      "theoretical_loss": 3.6078630471923914,
      "tokens_seen": 1128660992
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033233701103309933,
      "loss": 3.0929,
      "theoretical_loss": 3.6078435547963705,
      "tokens_seen": 1128726528
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033232698094282846,
      "loss": 2.8577,
      "theoretical_loss": 3.6078240638489563,
      "tokens_seen": 1128792064
    },
    {
      "epoch": 0.34,
      "objective/train/advantage_avg": 0.4944891631603241,
      "objective/train/docs_used": 643131,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.655165910720825,
      "objective/train/original_loss": 2.655165672302246,
      "objective/train/theoretical_loss": 3.607804574349956,
      "objective/train/tokens_used": 1149317600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24457933008670807,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506922006607056,
      "objective/train/weighted_lm_loss": 2.7898707389831543,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0253151655197144,
      "theoretical_loss": 3.607804574349956,
      "tokens_seen": 1128857600
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003323169508525577,
      "loss": 2.9528,
      "theoretical_loss": 3.607804574349956,
      "tokens_seen": 1128857600
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003323069207622869,
      "loss": 3.2251,
      "theoretical_loss": 3.6077850862991774,
      "tokens_seen": 1128923136
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033229689067201606,
      "loss": 2.8643,
      "theoretical_loss": 3.607765599696431,
      "tokens_seen": 1128988672
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033228686058174524,
      "loss": 2.8438,
      "theoretical_loss": 3.6077461145415226,
      "tokens_seen": 1129054208
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003322768304914744,
      "loss": 2.9113,
      "theoretical_loss": 3.6077266308342617,
      "tokens_seen": 1129119744
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003322668004012036,
      "loss": 3.014,
      "theoretical_loss": 3.607707148574457,
      "tokens_seen": 1129185280
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033225677031093284,
      "loss": 2.7732,
      "theoretical_loss": 3.607687667761917,
      "tokens_seen": 1129250816
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033224674022066196,
      "loss": 3.1155,
      "theoretical_loss": 3.60766818839645,
      "tokens_seen": 1129316352
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003322367101303912,
      "loss": 3.1007,
      "theoretical_loss": 3.6076487104778643,
      "tokens_seen": 1129381888
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003322266800401204,
      "loss": 2.9332,
      "theoretical_loss": 3.6076292340059686,
      "tokens_seen": 1129447424
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033221664994984956,
      "loss": 3.1313,
      "theoretical_loss": 3.607609758980572,
      "tokens_seen": 1129512960
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033220661985957874,
      "loss": 3.0257,
      "theoretical_loss": 3.607590285401482,
      "tokens_seen": 1129578496
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003321965897693079,
      "loss": 2.5952,
      "theoretical_loss": 3.6075708132685085,
      "tokens_seen": 1129644032
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003321865596790371,
      "loss": 2.8384,
      "theoretical_loss": 3.6075513425814596,
      "tokens_seen": 1129709568
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033217652958876634,
      "loss": 2.7836,
      "theoretical_loss": 3.607531873340144,
      "tokens_seen": 1129775104
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033216649949849547,
      "loss": 3.0013,
      "theoretical_loss": 3.6075124055443712,
      "tokens_seen": 1129840640
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003321564694082247,
      "loss": 3.0794,
      "theoretical_loss": 3.607492939193949,
      "tokens_seen": 1129906176
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033214643931795383,
      "loss": 2.8076,
      "theoretical_loss": 3.607473474288687,
      "tokens_seen": 1129971712
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033213640922768306,
      "loss": 3.0339,
      "theoretical_loss": 3.6074540108283943,
      "tokens_seen": 1130037248
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033212637913741225,
      "loss": 2.845,
      "theoretical_loss": 3.607434548812879,
      "tokens_seen": 1130102784
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003321163490471414,
      "loss": 2.9055,
      "theoretical_loss": 3.607415088241951,
      "tokens_seen": 1130168320
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003321063189568706,
      "loss": 2.927,
      "theoretical_loss": 3.607395629115419,
      "tokens_seen": 1130233856
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033209628886659984,
      "loss": 2.7916,
      "theoretical_loss": 3.607376171433091,
      "tokens_seen": 1130299392
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033208625877632897,
      "loss": 2.8193,
      "theoretical_loss": 3.6073567151947774,
      "tokens_seen": 1130364928
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003320762286860582,
      "loss": 2.9517,
      "theoretical_loss": 3.6073372604002865,
      "tokens_seen": 1130430464
    },
    {
      "epoch": 0.34,
      "objective/train/advantage_avg": 0.4944426417350769,
      "objective/train/docs_used": 644333,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.0469279289245605,
      "objective/train/original_loss": 3.0469279289245605,
      "objective/train/theoretical_loss": 3.6073178070494287,
      "objective/train/tokens_used": 1150956000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24454595148563385,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050687313079834,
      "objective/train/weighted_lm_loss": 3.201331853866577,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0396549701690674,
      "theoretical_loss": 3.6073178070494287,
      "tokens_seen": 1130496000
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033206619859578733,
      "loss": 3.0162,
      "theoretical_loss": 3.6073178070494287,
      "tokens_seen": 1130496000
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033205616850551657,
      "loss": 2.9268,
      "theoretical_loss": 3.607298355142012,
      "tokens_seen": 1130561536
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033204613841524575,
      "loss": 3.1017,
      "theoretical_loss": 3.607278904677846,
      "tokens_seen": 1130627072
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033203610832497493,
      "loss": 2.8617,
      "theoretical_loss": 3.60725945565674,
      "tokens_seen": 1130692608
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003320260782347041,
      "loss": 2.7737,
      "theoretical_loss": 3.607240008078503,
      "tokens_seen": 1130758144
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003320160481444333,
      "loss": 3.0997,
      "theoretical_loss": 3.607220561942945,
      "tokens_seen": 1130823680
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003320060180541625,
      "loss": 3.0798,
      "theoretical_loss": 3.6072011172498746,
      "tokens_seen": 1130889216
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003319959879638917,
      "loss": 3.0416,
      "theoretical_loss": 3.607181673999102,
      "tokens_seen": 1130954752
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033198595787362084,
      "loss": 2.9232,
      "theoretical_loss": 3.607162232190436,
      "tokens_seen": 1131020288
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033197592778335007,
      "loss": 2.698,
      "theoretical_loss": 3.6071427918236862,
      "tokens_seen": 1131085824
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003319658976930792,
      "loss": 2.9815,
      "theoretical_loss": 3.607123352898663,
      "tokens_seen": 1131151360
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033195586760280843,
      "loss": 2.8733,
      "theoretical_loss": 3.6071039154151743,
      "tokens_seen": 1131216896
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003319458375125376,
      "loss": 2.8949,
      "theoretical_loss": 3.6070844793730314,
      "tokens_seen": 1131282432
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003319358074222668,
      "loss": 2.8196,
      "theoretical_loss": 3.6070650447720425,
      "tokens_seen": 1131347968
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.000331925777331996,
      "loss": 2.9382,
      "theoretical_loss": 3.607045611612018,
      "tokens_seen": 1131413504
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003319157472417252,
      "loss": 2.7466,
      "theoretical_loss": 3.607026179892768,
      "tokens_seen": 1131479040
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033190571715145434,
      "loss": 3.0122,
      "theoretical_loss": 3.6070067496141016,
      "tokens_seen": 1131544576
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003318956870611836,
      "loss": 2.8461,
      "theoretical_loss": 3.6069873207758287,
      "tokens_seen": 1131610112
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003318856569709127,
      "loss": 2.8326,
      "theoretical_loss": 3.6069678933777594,
      "tokens_seen": 1131675648
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033187562688064194,
      "loss": 3.094,
      "theoretical_loss": 3.606948467419703,
      "tokens_seen": 1131741184
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003318655967903711,
      "loss": 3.1046,
      "theoretical_loss": 3.606929042901469,
      "tokens_seen": 1131806720
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003318555667001003,
      "loss": 2.6427,
      "theoretical_loss": 3.6069096198228694,
      "tokens_seen": 1131872256
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003318455366098295,
      "loss": 3.0802,
      "theoretical_loss": 3.606890198183712,
      "tokens_seen": 1131937792
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033183550651955866,
      "loss": 2.9285,
      "theoretical_loss": 3.606870777983807,
      "tokens_seen": 1132003328
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003318254764292879,
      "loss": 2.9281,
      "theoretical_loss": 3.606851359222966,
      "tokens_seen": 1132068864
    },
    {
      "epoch": 0.34,
      "objective/train/advantage_avg": 0.4939291775226593,
      "objective/train/docs_used": 644979,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.2712085247039795,
      "objective/train/original_loss": 2.2712087631225586,
      "objective/train/theoretical_loss": 3.6068319419009978,
      "objective/train/tokens_used": 1152594400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24407535791397095,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506336688995361,
      "objective/train/weighted_lm_loss": 2.3862221240997314,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 3.6068319419009978,
      "tokens_seen": 1132134400
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003318154463390171,
      "loss": 2.7693,
      "theoretical_loss": 3.6068319419009978,
      "tokens_seen": 1132134400
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033180541624874626,
      "loss": 2.6787,
      "theoretical_loss": 3.606812526017712,
      "tokens_seen": 1132199936
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033179538615847544,
      "loss": 2.7631,
      "theoretical_loss": 3.6067931115729204,
      "tokens_seen": 1132265472
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003317853560682046,
      "loss": 2.7227,
      "theoretical_loss": 3.606773698566432,
      "tokens_seen": 1132331008
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003317753259779338,
      "loss": 2.9314,
      "theoretical_loss": 3.606754286998057,
      "tokens_seen": 1132396544
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033176529588766304,
      "loss": 2.8175,
      "theoretical_loss": 3.6067348768676064,
      "tokens_seen": 1132462080
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033175526579739216,
      "loss": 2.8481,
      "theoretical_loss": 3.6067154681748894,
      "tokens_seen": 1132527616
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003317452357071214,
      "loss": 2.8165,
      "theoretical_loss": 3.6066960609197176,
      "tokens_seen": 1132593152
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003317352056168506,
      "loss": 3.2462,
      "theoretical_loss": 3.6066766551019,
      "tokens_seen": 1132658688
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033172517552657976,
      "loss": 2.7136,
      "theoretical_loss": 3.6066572507212475,
      "tokens_seen": 1132724224
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033171514543630894,
      "loss": 2.5549,
      "theoretical_loss": 3.6066378477775713,
      "tokens_seen": 1132789760
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003317051153460381,
      "loss": 2.9815,
      "theoretical_loss": 3.606618446270681,
      "tokens_seen": 1132855296
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003316950852557673,
      "loss": 2.5531,
      "theoretical_loss": 3.6065990462003876,
      "tokens_seen": 1132920832
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033168505516549654,
      "loss": 2.6821,
      "theoretical_loss": 3.606579647566501,
      "tokens_seen": 1132986368
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033167502507522567,
      "loss": 2.9914,
      "theoretical_loss": 3.6065602503688323,
      "tokens_seen": 1133051904
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003316649949849549,
      "loss": 2.7835,
      "theoretical_loss": 3.606540854607192,
      "tokens_seen": 1133117440
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033165496489468403,
      "loss": 2.8802,
      "theoretical_loss": 3.6065214602813906,
      "tokens_seen": 1133182976
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033164493480441326,
      "loss": 2.8739,
      "theoretical_loss": 3.606502067391239,
      "tokens_seen": 1133248512
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033163490471414245,
      "loss": 2.9013,
      "theoretical_loss": 3.6064826759365474,
      "tokens_seen": 1133314048
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003316248746238716,
      "loss": 2.8992,
      "theoretical_loss": 3.606463285917127,
      "tokens_seen": 1133379584
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003316148445336008,
      "loss": 2.9398,
      "theoretical_loss": 3.6064438973327886,
      "tokens_seen": 1133445120
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033160481444333004,
      "loss": 3.0429,
      "theoretical_loss": 3.606424510183343,
      "tokens_seen": 1133510656
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033159478435305917,
      "loss": 2.8844,
      "theoretical_loss": 3.6064051244686013,
      "tokens_seen": 1133576192
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003315847542627884,
      "loss": 2.8092,
      "theoretical_loss": 3.606385740188373,
      "tokens_seen": 1133641728
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033157472417251753,
      "loss": 2.8964,
      "theoretical_loss": 3.606366357342471,
      "tokens_seen": 1133707264
    },
    {
      "epoch": 0.34,
      "objective/train/advantage_avg": 0.4957732558250427,
      "objective/train/docs_used": 646384,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.4400062561035156,
      "objective/train/original_loss": 2.4400062561035156,
      "objective/train/theoretical_loss": 3.6063469759307054,
      "objective/train/tokens_used": 1154232800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2458311766386032,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0508270263671875,
      "objective/train/weighted_lm_loss": 2.5638866424560547,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0451053380966187,
      "theoretical_loss": 3.6063469759307054,
      "tokens_seen": 1133772800
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033156469408224677,
      "loss": 2.8696,
      "theoretical_loss": 3.6063469759307054,
      "tokens_seen": 1133772800
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033155466399197595,
      "loss": 2.8078,
      "theoretical_loss": 3.606327595952887,
      "tokens_seen": 1133838336
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033154463390170513,
      "loss": 2.7798,
      "theoretical_loss": 3.606308217408827,
      "tokens_seen": 1133903872
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003315346038114343,
      "loss": 2.8635,
      "theoretical_loss": 3.606288840298336,
      "tokens_seen": 1133969408
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003315245737211635,
      "loss": 2.6716,
      "theoretical_loss": 3.606269464621226,
      "tokens_seen": 1134034944
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003315145436308927,
      "loss": 2.9841,
      "theoretical_loss": 3.6062500903773076,
      "tokens_seen": 1134100480
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003315045135406219,
      "loss": 2.8424,
      "theoretical_loss": 3.606230717566392,
      "tokens_seen": 1134166016
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033149448345035104,
      "loss": 2.9492,
      "theoretical_loss": 3.6062113461882914,
      "tokens_seen": 1134231552
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033148445336008027,
      "loss": 2.8128,
      "theoretical_loss": 3.606191976242815,
      "tokens_seen": 1134297088
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003314744232698094,
      "loss": 2.758,
      "theoretical_loss": 3.6061726077297758,
      "tokens_seen": 1134362624
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033146439317953863,
      "loss": 2.9532,
      "theoretical_loss": 3.6061532406489842,
      "tokens_seen": 1134428160
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003314543630892678,
      "loss": 2.8561,
      "theoretical_loss": 3.6061338750002525,
      "tokens_seen": 1134493696
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.000331444332998997,
      "loss": 2.9804,
      "theoretical_loss": 3.606114510783391,
      "tokens_seen": 1134559232
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003314343029087262,
      "loss": 2.925,
      "theoretical_loss": 3.6060951479982117,
      "tokens_seen": 1134624768
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003314242728184554,
      "loss": 2.9312,
      "theoretical_loss": 3.6060757866445265,
      "tokens_seen": 1134690304
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033141424272818454,
      "loss": 2.8627,
      "theoretical_loss": 3.606056426722146,
      "tokens_seen": 1134755840
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003314042126379138,
      "loss": 2.9057,
      "theoretical_loss": 3.606037068230882,
      "tokens_seen": 1134821376
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003313941825476429,
      "loss": 2.7519,
      "theoretical_loss": 3.606017711170547,
      "tokens_seen": 1134886912
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033138415245737214,
      "loss": 2.7276,
      "theoretical_loss": 3.605998355540951,
      "tokens_seen": 1134952448
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003313741223671013,
      "loss": 2.8354,
      "theoretical_loss": 3.6059790013419066,
      "tokens_seen": 1135017984
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003313640922768305,
      "loss": 2.9614,
      "theoretical_loss": 3.6059596485732257,
      "tokens_seen": 1135083520
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003313540621865597,
      "loss": 2.8505,
      "theoretical_loss": 3.605940297234719,
      "tokens_seen": 1135149056
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033134403209628886,
      "loss": 2.6938,
      "theoretical_loss": 3.6059209473261995,
      "tokens_seen": 1135214592
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033133400200601804,
      "loss": 3.0686,
      "theoretical_loss": 3.6059015988474785,
      "tokens_seen": 1135280128
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003313239719157473,
      "loss": 3.211,
      "theoretical_loss": 3.6058822517983673,
      "tokens_seen": 1135345664
    },
    {
      "epoch": 0.34,
      "objective/train/advantage_avg": 0.4933038055896759,
      "objective/train/docs_used": 647116,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.9581212997436523,
      "objective/train/original_loss": 2.9581212997436523,
      "objective/train/theoretical_loss": 3.6058629061786784,
      "objective/train/tokens_used": 1155871200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24344830214977264,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050567865371704,
      "objective/train/weighted_lm_loss": 3.1075186729431152,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0338950157165527,
      "theoretical_loss": 3.6058629061786784,
      "tokens_seen": 1135411200
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003313139418254764,
      "loss": 2.861,
      "theoretical_loss": 3.6058629061786784,
      "tokens_seen": 1135411200
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033130391173520564,
      "loss": 3.0142,
      "theoretical_loss": 3.605843561988223,
      "tokens_seen": 1135476736
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033129388164493477,
      "loss": 2.946,
      "theoretical_loss": 3.605824219226814,
      "tokens_seen": 1135542272
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.000331283851554664,
      "loss": 2.8896,
      "theoretical_loss": 3.605804877894263,
      "tokens_seen": 1135607808
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003312738214643932,
      "loss": 2.8401,
      "theoretical_loss": 3.6057855379903816,
      "tokens_seen": 1135673344
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033126379137412236,
      "loss": 2.7952,
      "theoretical_loss": 3.6057661995149823,
      "tokens_seen": 1135738880
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033125376128385155,
      "loss": 2.8961,
      "theoretical_loss": 3.6057468624678766,
      "tokens_seen": 1135804416
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003312437311935808,
      "loss": 2.8015,
      "theoretical_loss": 3.6057275268488773,
      "tokens_seen": 1135869952
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003312337011033099,
      "loss": 3.0624,
      "theoretical_loss": 3.605708192657796,
      "tokens_seen": 1135935488
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033122367101303914,
      "loss": 2.9859,
      "theoretical_loss": 3.605688859894445,
      "tokens_seen": 1136001024
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033121364092276827,
      "loss": 2.9158,
      "theoretical_loss": 3.6056695285586366,
      "tokens_seen": 1136066560
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003312036108324975,
      "loss": 2.9385,
      "theoretical_loss": 3.605650198650183,
      "tokens_seen": 1136132096
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003311935807422267,
      "loss": 2.792,
      "theoretical_loss": 3.6056308701688975,
      "tokens_seen": 1136197632
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033118355065195587,
      "loss": 2.8281,
      "theoretical_loss": 3.6056115431145903,
      "tokens_seen": 1136263168
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033117352056168505,
      "loss": 3.0135,
      "theoretical_loss": 3.605592217487075,
      "tokens_seen": 1136328704
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033116349047141423,
      "loss": 2.8992,
      "theoretical_loss": 3.6055728932861646,
      "tokens_seen": 1136394240
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003311534603811434,
      "loss": 3.0411,
      "theoretical_loss": 3.6055535705116704,
      "tokens_seen": 1136459776
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033114343029087265,
      "loss": 3.1255,
      "theoretical_loss": 3.605534249163405,
      "tokens_seen": 1136525312
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033113340020060177,
      "loss": 2.6231,
      "theoretical_loss": 3.605514929241182,
      "tokens_seen": 1136590848
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.000331123370110331,
      "loss": 3.0704,
      "theoretical_loss": 3.6054956107448124,
      "tokens_seen": 1136656384
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033111334002006013,
      "loss": 2.7427,
      "theoretical_loss": 3.60547629367411,
      "tokens_seen": 1136721920
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033110330992978937,
      "loss": 2.9486,
      "theoretical_loss": 3.6054569780288865,
      "tokens_seen": 1136787456
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033109327983951855,
      "loss": 2.8825,
      "theoretical_loss": 3.6054376638089547,
      "tokens_seen": 1136852992
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033108324974924773,
      "loss": 2.7999,
      "theoretical_loss": 3.605418351014128,
      "tokens_seen": 1136918528
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033107321965897697,
      "loss": 3.0317,
      "theoretical_loss": 3.6053990396442184,
      "tokens_seen": 1136984064
    },
    {
      "epoch": 0.34,
      "objective/train/advantage_avg": 0.49545836448669434,
      "objective/train/docs_used": 648555,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.0606226921081543,
      "objective/train/original_loss": 2.0606226921081543,
      "objective/train/theoretical_loss": 3.605379729699039,
      "objective/train/tokens_used": 1157509600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24554617702960968,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507941246032715,
      "objective/train/weighted_lm_loss": 2.1652987003326416,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 3.605379729699039,
      "tokens_seen": 1137049600
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033106318956870615,
      "loss": 2.676,
      "theoretical_loss": 3.605379729699039,
      "tokens_seen": 1137049600
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033105315947843533,
      "loss": 2.9802,
      "theoretical_loss": 3.6053604211784025,
      "tokens_seen": 1137115136
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003310431293881645,
      "loss": 2.9972,
      "theoretical_loss": 3.6053411140821208,
      "tokens_seen": 1137180672
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003310330992978937,
      "loss": 2.8973,
      "theoretical_loss": 3.6053218084100083,
      "tokens_seen": 1137246208
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003310230692076229,
      "loss": 2.7979,
      "theoretical_loss": 3.605302504161877,
      "tokens_seen": 1137311744
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003310130391173521,
      "loss": 2.5663,
      "theoretical_loss": 3.6052832013375404,
      "tokens_seen": 1137377280
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033100300902708124,
      "loss": 3.1995,
      "theoretical_loss": 3.605263899936811,
      "tokens_seen": 1137442816
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033099297893681047,
      "loss": 2.8861,
      "theoretical_loss": 3.6052445999595015,
      "tokens_seen": 1137508352
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003309829488465396,
      "loss": 2.6822,
      "theoretical_loss": 3.6052253014054254,
      "tokens_seen": 1137573888
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033097291875626883,
      "loss": 2.912,
      "theoretical_loss": 3.605206004274396,
      "tokens_seen": 1137639424
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.000330962888665998,
      "loss": 2.84,
      "theoretical_loss": 3.605186708566225,
      "tokens_seen": 1137704960
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003309528585757272,
      "loss": 2.841,
      "theoretical_loss": 3.6051674142807277,
      "tokens_seen": 1137770496
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003309428284854564,
      "loss": 2.8187,
      "theoretical_loss": 3.6051481214177157,
      "tokens_seen": 1137836032
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003309327983951856,
      "loss": 2.7196,
      "theoretical_loss": 3.605128829977003,
      "tokens_seen": 1137901568
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033092276830491474,
      "loss": 2.8147,
      "theoretical_loss": 3.605109539958402,
      "tokens_seen": 1137967104
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.000330912738214644,
      "loss": 2.7863,
      "theoretical_loss": 3.6050902513617267,
      "tokens_seen": 1138032640
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003309027081243731,
      "loss": 2.9059,
      "theoretical_loss": 3.60507096418679,
      "tokens_seen": 1138098176
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033089267803410234,
      "loss": 2.812,
      "theoretical_loss": 3.605051678433406,
      "tokens_seen": 1138163712
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003308826479438315,
      "loss": 2.7967,
      "theoretical_loss": 3.6050323941013867,
      "tokens_seen": 1138229248
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003308726178535607,
      "loss": 2.7695,
      "theoretical_loss": 3.605013111190547,
      "tokens_seen": 1138294784
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.0003308625877632899,
      "loss": 2.7157,
      "theoretical_loss": 3.6049938297006987,
      "tokens_seen": 1138360320
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033085255767301906,
      "loss": 3.059,
      "theoretical_loss": 3.604974549631657,
      "tokens_seen": 1138425856
    },
    {
      "epoch": 0.34,
      "learning_rate": 0.00033084252758274824,
      "loss": 2.7534,
      "theoretical_loss": 3.6049552709832344,
      "tokens_seen": 1138491392
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003308324974924775,
      "loss": 2.8542,
      "theoretical_loss": 3.6049359937552454,
      "tokens_seen": 1138556928
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003308224674022066,
      "loss": 2.9959,
      "theoretical_loss": 3.6049167179475026,
      "tokens_seen": 1138622464
    },
    {
      "epoch": 0.35,
      "objective/train/advantage_avg": 0.4941384792327881,
      "objective/train/docs_used": 649005,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.891140937805176,
      "objective/train/original_loss": 2.891141176223755,
      "objective/train/theoretical_loss": 3.6048974435598193,
      "objective/train/tokens_used": 1159148000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2442622184753418,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050655484199524,
      "objective/train/weighted_lm_loss": 3.0376694202423096,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0363596677780151,
      "theoretical_loss": 3.6048974435598193,
      "tokens_seen": 1138688000
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00033081243731193584,
      "loss": 2.9563,
      "theoretical_loss": 3.6048974435598193,
      "tokens_seen": 1138688000
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00033080240722166497,
      "loss": 2.856,
      "theoretical_loss": 3.6048781705920105,
      "tokens_seen": 1138753536
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003307923771313942,
      "loss": 2.9807,
      "theoretical_loss": 3.6048588990438892,
      "tokens_seen": 1138819072
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003307823470411234,
      "loss": 2.8722,
      "theoretical_loss": 3.6048396289152684,
      "tokens_seen": 1138884608
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00033077231695085256,
      "loss": 2.7469,
      "theoretical_loss": 3.604820360205964,
      "tokens_seen": 1138950144
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00033076228686058175,
      "loss": 2.8284,
      "theoretical_loss": 3.604801092915787,
      "tokens_seen": 1139015680
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.000330752256770311,
      "loss": 2.8072,
      "theoretical_loss": 3.6047818270445533,
      "tokens_seen": 1139081216
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003307422266800401,
      "loss": 3.0602,
      "theoretical_loss": 3.604762562592076,
      "tokens_seen": 1139146752
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00033073219658976934,
      "loss": 2.8319,
      "theoretical_loss": 3.6047432995581694,
      "tokens_seen": 1139212288
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00033072216649949847,
      "loss": 2.9984,
      "theoretical_loss": 3.604724037942647,
      "tokens_seen": 1139277824
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003307121364092277,
      "loss": 2.6201,
      "theoretical_loss": 3.604704777745323,
      "tokens_seen": 1139343360
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003307021063189569,
      "loss": 2.7269,
      "theoretical_loss": 3.604685518966011,
      "tokens_seen": 1139408896
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00033069207622868607,
      "loss": 2.8749,
      "theoretical_loss": 3.604666261604526,
      "tokens_seen": 1139474432
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00033068204613841525,
      "loss": 2.8842,
      "theoretical_loss": 3.604647005660681,
      "tokens_seen": 1139539968
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00033067201604814443,
      "loss": 2.7906,
      "theoretical_loss": 3.604627751134291,
      "tokens_seen": 1139605504
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003306619859578736,
      "loss": 2.7835,
      "theoretical_loss": 3.6046084980251694,
      "tokens_seen": 1139671040
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00033065195586760285,
      "loss": 2.968,
      "theoretical_loss": 3.604589246333131,
      "tokens_seen": 1139736576
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00033064192577733197,
      "loss": 2.8688,
      "theoretical_loss": 3.6045699960579896,
      "tokens_seen": 1139802112
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003306318956870612,
      "loss": 2.6978,
      "theoretical_loss": 3.60455074719956,
      "tokens_seen": 1139867648
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00033062186559679034,
      "loss": 3.0966,
      "theoretical_loss": 3.6045314997576563,
      "tokens_seen": 1139933184
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00033061183550651957,
      "loss": 2.8243,
      "theoretical_loss": 3.604512253732092,
      "tokens_seen": 1139998720
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00033060180541624875,
      "loss": 2.7575,
      "theoretical_loss": 3.604493009122683,
      "tokens_seen": 1140064256
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00033059177532597793,
      "loss": 2.6091,
      "theoretical_loss": 3.604473765929242,
      "tokens_seen": 1140129792
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003305817452357071,
      "loss": 2.7487,
      "theoretical_loss": 3.6044545241515844,
      "tokens_seen": 1140195328
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00033057171514543635,
      "loss": 3.0302,
      "theoretical_loss": 3.604435283789525,
      "tokens_seen": 1140260864
    },
    {
      "epoch": 0.35,
      "objective/train/advantage_avg": 0.4940442442893982,
      "objective/train/docs_used": 649993,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.8935441970825195,
      "objective/train/original_loss": 2.8935439586639404,
      "objective/train/theoretical_loss": 3.6044160448428775,
      "objective/train/tokens_used": 1160786400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24413363635540009,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506454706192017,
      "objective/train/weighted_lm_loss": 3.04008412361145,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.044721245765686,
      "theoretical_loss": 3.6044160448428775,
      "tokens_seen": 1140326400
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003305616850551655,
      "loss": 2.9712,
      "theoretical_loss": 3.6044160448428775,
      "tokens_seen": 1140326400
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003305516549648947,
      "loss": 2.8345,
      "theoretical_loss": 3.6043968073114563,
      "tokens_seen": 1140391936
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00033054162487462384,
      "loss": 3.0514,
      "theoretical_loss": 3.6043775711950774,
      "tokens_seen": 1140457472
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003305315947843531,
      "loss": 2.9652,
      "theoretical_loss": 3.6043583364935543,
      "tokens_seen": 1140523008
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00033052156469408225,
      "loss": 3.0085,
      "theoretical_loss": 3.6043391032067014,
      "tokens_seen": 1140588544
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00033051153460381144,
      "loss": 3.0567,
      "theoretical_loss": 3.604319871334334,
      "tokens_seen": 1140654080
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003305015045135406,
      "loss": 2.8581,
      "theoretical_loss": 3.604300640876266,
      "tokens_seen": 1140719616
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003304914744232698,
      "loss": 2.9407,
      "theoretical_loss": 3.6042814118323134,
      "tokens_seen": 1140785152
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.000330481444332999,
      "loss": 2.8596,
      "theoretical_loss": 3.60426218420229,
      "tokens_seen": 1140850688
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003304714142427282,
      "loss": 3.0508,
      "theoretical_loss": 3.6042429579860116,
      "tokens_seen": 1140916224
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00033046138415245734,
      "loss": 2.7652,
      "theoretical_loss": 3.604223733183292,
      "tokens_seen": 1140981760
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003304513540621866,
      "loss": 2.8942,
      "theoretical_loss": 3.6042045097939464,
      "tokens_seen": 1141047296
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003304413239719157,
      "loss": 2.8815,
      "theoretical_loss": 3.6041852878177902,
      "tokens_seen": 1141112832
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00033043129388164494,
      "loss": 2.8552,
      "theoretical_loss": 3.604166067254637,
      "tokens_seen": 1141178368
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003304212637913741,
      "loss": 3.0883,
      "theoretical_loss": 3.6041468481043037,
      "tokens_seen": 1141243904
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003304112337011033,
      "loss": 2.7326,
      "theoretical_loss": 3.6041276303666043,
      "tokens_seen": 1141309440
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003304012036108325,
      "loss": 2.7849,
      "theoretical_loss": 3.6041084140413533,
      "tokens_seen": 1141374976
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003303911735205617,
      "loss": 3.081,
      "theoretical_loss": 3.6040891991283672,
      "tokens_seen": 1141440512
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00033038114343029084,
      "loss": 2.8855,
      "theoretical_loss": 3.60406998562746,
      "tokens_seen": 1141506048
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003303711133400201,
      "loss": 2.9601,
      "theoretical_loss": 3.6040507735384475,
      "tokens_seen": 1141571584
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003303610832497492,
      "loss": 3.0497,
      "theoretical_loss": 3.6040315628611443,
      "tokens_seen": 1141637120
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00033035105315947844,
      "loss": 2.6201,
      "theoretical_loss": 3.6040123535953663,
      "tokens_seen": 1141702656
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003303410230692076,
      "loss": 2.9207,
      "theoretical_loss": 3.6039931457409287,
      "tokens_seen": 1141768192
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003303309929789368,
      "loss": 3.0965,
      "theoretical_loss": 3.6039739392976458,
      "tokens_seen": 1141833728
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00033032096288866604,
      "loss": 2.8224,
      "theoretical_loss": 3.603954734265334,
      "tokens_seen": 1141899264
    },
    {
      "epoch": 0.35,
      "objective/train/advantage_avg": 0.4966695308685303,
      "objective/train/docs_used": 650682,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.2465829849243164,
      "objective/train/original_loss": 2.2465829849243164,
      "objective/train/theoretical_loss": 3.603935530643809,
      "objective/train/tokens_used": 1162424800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24670898914337158,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050921082496643,
      "objective/train/weighted_lm_loss": 2.3609776496887207,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0437887907028198,
      "theoretical_loss": 3.603935530643809,
      "tokens_seen": 1141964800
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00033031093279839517,
      "loss": 2.7561,
      "theoretical_loss": 3.603935530643809,
      "tokens_seen": 1141964800
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003303009027081244,
      "loss": 3.0646,
      "theoretical_loss": 3.603916328432885,
      "tokens_seen": 1142030336
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003302908726178536,
      "loss": 2.9299,
      "theoretical_loss": 3.603897127632378,
      "tokens_seen": 1142095872
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00033028084252758276,
      "loss": 3.2936,
      "theoretical_loss": 3.603877928242104,
      "tokens_seen": 1142161408
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00033027081243731195,
      "loss": 2.6017,
      "theoretical_loss": 3.6038587302618774,
      "tokens_seen": 1142226944
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003302607823470412,
      "loss": 2.6472,
      "theoretical_loss": 3.6038395336915148,
      "tokens_seen": 1142292480
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003302507522567703,
      "loss": 2.7874,
      "theoretical_loss": 3.603820338530831,
      "tokens_seen": 1142358016
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00033024072216649954,
      "loss": 3.2792,
      "theoretical_loss": 3.6038011447796423,
      "tokens_seen": 1142423552
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00033023069207622867,
      "loss": 2.8717,
      "theoretical_loss": 3.603781952437764,
      "tokens_seen": 1142489088
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003302206619859579,
      "loss": 2.9483,
      "theoretical_loss": 3.603762761505012,
      "tokens_seen": 1142554624
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003302106318956871,
      "loss": 2.9258,
      "theoretical_loss": 3.6037435719812017,
      "tokens_seen": 1142620160
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00033020060180541627,
      "loss": 2.9274,
      "theoretical_loss": 3.603724383866149,
      "tokens_seen": 1142685696
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00033019057171514545,
      "loss": 2.909,
      "theoretical_loss": 3.6037051971596696,
      "tokens_seen": 1142751232
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00033018054162487463,
      "loss": 2.551,
      "theoretical_loss": 3.6036860118615794,
      "tokens_seen": 1142816768
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003301705115346038,
      "loss": 3.0,
      "theoretical_loss": 3.6036668279716944,
      "tokens_seen": 1142882304
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00033016048144433305,
      "loss": 2.9583,
      "theoretical_loss": 3.60364764548983,
      "tokens_seen": 1142947840
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003301504513540622,
      "loss": 2.9128,
      "theoretical_loss": 3.603628464415803,
      "tokens_seen": 1143013376
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003301404212637914,
      "loss": 2.9471,
      "theoretical_loss": 3.603609284749429,
      "tokens_seen": 1143078912
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00033013039117352054,
      "loss": 2.8373,
      "theoretical_loss": 3.6035901064905236,
      "tokens_seen": 1143144448
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00033012036108324977,
      "loss": 2.8431,
      "theoretical_loss": 3.603570929638903,
      "tokens_seen": 1143209984
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00033011033099297895,
      "loss": 2.9028,
      "theoretical_loss": 3.603551754194384,
      "tokens_seen": 1143275520
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00033010030090270813,
      "loss": 2.9905,
      "theoretical_loss": 3.6035325801567812,
      "tokens_seen": 1143341056
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003300902708124373,
      "loss": 2.84,
      "theoretical_loss": 3.603513407525912,
      "tokens_seen": 1143406592
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00033008024072216655,
      "loss": 2.7352,
      "theoretical_loss": 3.603494236301592,
      "tokens_seen": 1143472128
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003300702106318957,
      "loss": 3.0544,
      "theoretical_loss": 3.603475066483638,
      "tokens_seen": 1143537664
    },
    {
      "epoch": 0.35,
      "objective/train/advantage_avg": 0.4950396716594696,
      "objective/train/docs_used": 652177,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.7208878993988037,
      "objective/train/original_loss": 2.7208878993988037,
      "objective/train/theoretical_loss": 3.603455898071866,
      "objective/train/tokens_used": 1164063200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24511826038360596,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507500171661377,
      "objective/train/weighted_lm_loss": 2.8590657711029053,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0425469875335693,
      "theoretical_loss": 3.603455898071866,
      "tokens_seen": 1143603200
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003300601805416249,
      "loss": 2.7063,
      "theoretical_loss": 3.603455898071866,
      "tokens_seen": 1143603200
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00033005015045135404,
      "loss": 2.7607,
      "theoretical_loss": 3.6034367310660915,
      "tokens_seen": 1143668736
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003300401203610833,
      "loss": 2.7517,
      "theoretical_loss": 3.603417565466132,
      "tokens_seen": 1143734272
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00033003009027081245,
      "loss": 2.9355,
      "theoretical_loss": 3.6033984012718028,
      "tokens_seen": 1143799808
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00033002006018054164,
      "loss": 2.8552,
      "theoretical_loss": 3.603379238482921,
      "tokens_seen": 1143865344
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003300100300902708,
      "loss": 3.0163,
      "theoretical_loss": 3.603360077099303,
      "tokens_seen": 1143930880
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00033,
      "loss": 2.5997,
      "theoretical_loss": 3.6033409171207644,
      "tokens_seen": 1143996416
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003299899699097292,
      "loss": 3.0457,
      "theoretical_loss": 3.603321758547123,
      "tokens_seen": 1144061952
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003299799398194584,
      "loss": 3.041,
      "theoretical_loss": 3.603302601378194,
      "tokens_seen": 1144127488
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032996990972918754,
      "loss": 2.7973,
      "theoretical_loss": 3.603283445613795,
      "tokens_seen": 1144193024
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003299598796389168,
      "loss": 2.8409,
      "theoretical_loss": 3.6032642912537423,
      "tokens_seen": 1144258560
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003299498495486459,
      "loss": 2.9353,
      "theoretical_loss": 3.6032451382978516,
      "tokens_seen": 1144324096
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032993981945837514,
      "loss": 2.8523,
      "theoretical_loss": 3.603225986745941,
      "tokens_seen": 1144389632
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003299297893681043,
      "loss": 2.6294,
      "theoretical_loss": 3.6032068365978267,
      "tokens_seen": 1144455168
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003299197592778335,
      "loss": 2.8304,
      "theoretical_loss": 3.6031876878533247,
      "tokens_seen": 1144520704
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003299097291875627,
      "loss": 2.9226,
      "theoretical_loss": 3.603168540512253,
      "tokens_seen": 1144586240
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003298996990972919,
      "loss": 2.9258,
      "theoretical_loss": 3.603149394574427,
      "tokens_seen": 1144651776
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032988966900702104,
      "loss": 2.9331,
      "theoretical_loss": 3.6031302500396647,
      "tokens_seen": 1144717312
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003298796389167503,
      "loss": 2.7435,
      "theoretical_loss": 3.6031111069077824,
      "tokens_seen": 1144782848
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003298696088264794,
      "loss": 2.7645,
      "theoretical_loss": 3.6030919651785966,
      "tokens_seen": 1144848384
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032985957873620864,
      "loss": 2.8144,
      "theoretical_loss": 3.6030728248519255,
      "tokens_seen": 1144913920
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003298495486459378,
      "loss": 2.8812,
      "theoretical_loss": 3.603053685927585,
      "tokens_seen": 1144979456
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.000329839518555667,
      "loss": 2.6941,
      "theoretical_loss": 3.6030345484053923,
      "tokens_seen": 1145044992
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003298294884653962,
      "loss": 2.9309,
      "theoretical_loss": 3.6030154122851643,
      "tokens_seen": 1145110528
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032981945837512537,
      "loss": 2.7523,
      "theoretical_loss": 3.6029962775667186,
      "tokens_seen": 1145176064
    },
    {
      "epoch": 0.35,
      "objective/train/advantage_avg": 0.4958942234516144,
      "objective/train/docs_used": 652765,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.8058180809020996,
      "objective/train/original_loss": 2.8058176040649414,
      "objective/train/theoretical_loss": 3.6029771442498717,
      "objective/train/tokens_used": 1165701600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24595928192138672,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0508397817611694,
      "objective/train/weighted_lm_loss": 2.948303699493408,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0363596677780151,
      "theoretical_loss": 3.6029771442498717,
      "tokens_seen": 1145241600
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032980942828485455,
      "loss": 2.8532,
      "theoretical_loss": 3.6029771442498717,
      "tokens_seen": 1145241600
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003297993981945838,
      "loss": 2.8371,
      "theoretical_loss": 3.602958012334441,
      "tokens_seen": 1145307136
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003297893681043129,
      "loss": 2.961,
      "theoretical_loss": 3.6029388818202444,
      "tokens_seen": 1145372672
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032977933801404215,
      "loss": 2.9746,
      "theoretical_loss": 3.6029197527070975,
      "tokens_seen": 1145438208
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003297693079237713,
      "loss": 2.8803,
      "theoretical_loss": 3.6029006249948194,
      "tokens_seen": 1145503744
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003297592778335005,
      "loss": 2.9449,
      "theoretical_loss": 3.602881498683226,
      "tokens_seen": 1145569280
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003297492477432297,
      "loss": 2.5592,
      "theoretical_loss": 3.602862373772135,
      "tokens_seen": 1145634816
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032973921765295887,
      "loss": 2.8782,
      "theoretical_loss": 3.6028432502613637,
      "tokens_seen": 1145700352
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032972918756268805,
      "loss": 3.0902,
      "theoretical_loss": 3.602824128150729,
      "tokens_seen": 1145765888
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003297191574724173,
      "loss": 2.9146,
      "theoretical_loss": 3.60280500744005,
      "tokens_seen": 1145831424
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003297091273821464,
      "loss": 2.7844,
      "theoretical_loss": 3.6027858881291426,
      "tokens_seen": 1145896960
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032969909729187565,
      "loss": 2.9437,
      "theoretical_loss": 3.6027667702178245,
      "tokens_seen": 1145962496
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003296890672016048,
      "loss": 2.6841,
      "theoretical_loss": 3.6027476537059133,
      "tokens_seen": 1146028032
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.000329679037111334,
      "loss": 2.866,
      "theoretical_loss": 3.602728538593227,
      "tokens_seen": 1146093568
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003296690070210632,
      "loss": 2.9602,
      "theoretical_loss": 3.6027094248795826,
      "tokens_seen": 1146159104
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003296589769307924,
      "loss": 3.02,
      "theoretical_loss": 3.602690312564798,
      "tokens_seen": 1146224640
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032964894684052155,
      "loss": 2.5448,
      "theoretical_loss": 3.6026712016486906,
      "tokens_seen": 1146290176
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032963891675025074,
      "loss": 2.8966,
      "theoretical_loss": 3.6026520921310787,
      "tokens_seen": 1146355712
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003296288866599799,
      "loss": 2.7417,
      "theoretical_loss": 3.6026329840117794,
      "tokens_seen": 1146421248
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032961885656970915,
      "loss": 2.7626,
      "theoretical_loss": 3.6026138772906107,
      "tokens_seen": 1146486784
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003296088264794383,
      "loss": 2.8846,
      "theoretical_loss": 3.6025947719673903,
      "tokens_seen": 1146552320
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003295987963891675,
      "loss": 2.762,
      "theoretical_loss": 3.6025756680419363,
      "tokens_seen": 1146617856
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003295887662988967,
      "loss": 2.9088,
      "theoretical_loss": 3.6025565655140657,
      "tokens_seen": 1146683392
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003295787362086259,
      "loss": 2.928,
      "theoretical_loss": 3.6025374643835977,
      "tokens_seen": 1146748928
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003295687061183551,
      "loss": 3.1767,
      "theoretical_loss": 3.602518364650349,
      "tokens_seen": 1146814464
    },
    {
      "debugging/Self-BLEU-5": 0.30559870208625683,
      "debugging/distinct-1-grams": 0.7412044636353954,
      "debugging/distinct-2-grams": 0.895262206148282,
      "debugging/entropy-1-grams": 5.238226014708625,
      "debugging/entropy-2-grams": 5.800866939604313,
      "debugging/length": 478.8,
      "debugging/num_segments": 5,
      "debugging/raw_token_scores_avg": 0.0040895272977650166,
      "debugging/raw_token_scores_std": 0.007617930416017771,
      "debugging/score": 0.006858498138102529,
      "debugging/score_std": 0.007628355764891668,
      "epoch": 0.35,
      "objective/train/advantage_avg": 0.49591249227523804,
      "objective/train/docs_used": 652965,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.657623767852783,
      "objective/train/original_loss": 2.657623767852783,
      "objective/train/theoretical_loss": 3.6024992663141386,
      "objective/train/tokens_used": 1167340000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2459871917963028,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0508415699005127,
      "objective/train/weighted_lm_loss": 2.792459487915039,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0437887907028198,
      "theoretical_loss": 3.6024992663141386,
      "tokens_seen": 1146880000
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032955867602808424,
      "loss": 2.9049,
      "theoretical_loss": 3.6024992663141386,
      "tokens_seen": 1146880000
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003295486459378135,
      "loss": 2.7405,
      "theoretical_loss": 3.602480169374784,
      "tokens_seen": 1146945536
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032953861584754265,
      "loss": 3.0526,
      "theoretical_loss": 3.602461073832103,
      "tokens_seen": 1147011072
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032952858575727184,
      "loss": 2.9796,
      "theoretical_loss": 3.6024419796859135,
      "tokens_seen": 1147076608
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.000329518555667001,
      "loss": 2.9202,
      "theoretical_loss": 3.6024228869360346,
      "tokens_seen": 1147142144
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003295085255767302,
      "loss": 2.9189,
      "theoretical_loss": 3.6024037955822834,
      "tokens_seen": 1147207680
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003294984954864594,
      "loss": 2.7842,
      "theoretical_loss": 3.6023847056244787,
      "tokens_seen": 1147273216
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003294884653961886,
      "loss": 2.8371,
      "theoretical_loss": 3.6023656170624383,
      "tokens_seen": 1147338752
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032947843530591774,
      "loss": 2.888,
      "theoretical_loss": 3.6023465298959807,
      "tokens_seen": 1147404288
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.000329468405215647,
      "loss": 3.0059,
      "theoretical_loss": 3.6023274441249242,
      "tokens_seen": 1147469824
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003294583751253761,
      "loss": 3.0053,
      "theoretical_loss": 3.6023083597490864,
      "tokens_seen": 1147535360
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032944834503510534,
      "loss": 2.7909,
      "theoretical_loss": 3.6022892767682873,
      "tokens_seen": 1147600896
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003294383149448345,
      "loss": 3.0081,
      "theoretical_loss": 3.602270195182343,
      "tokens_seen": 1147666432
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003294282848545637,
      "loss": 2.7,
      "theoretical_loss": 3.6022511149910734,
      "tokens_seen": 1147731968
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003294182547642929,
      "loss": 2.9326,
      "theoretical_loss": 3.602232036194297,
      "tokens_seen": 1147797504
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003294082246740221,
      "loss": 2.771,
      "theoretical_loss": 3.6022129587918315,
      "tokens_seen": 1147863040
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032939819458375124,
      "loss": 2.886,
      "theoretical_loss": 3.602193882783496,
      "tokens_seen": 1147928576
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003293881644934805,
      "loss": 2.6986,
      "theoretical_loss": 3.6021748081691083,
      "tokens_seen": 1147994112
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003293781344032096,
      "loss": 2.9266,
      "theoretical_loss": 3.602155734948488,
      "tokens_seen": 1148059648
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032936810431293884,
      "loss": 2.8844,
      "theoretical_loss": 3.602136663121453,
      "tokens_seen": 1148125184
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.000329358074222668,
      "loss": 3.0639,
      "theoretical_loss": 3.602117592687822,
      "tokens_seen": 1148190720
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003293480441323972,
      "loss": 2.9995,
      "theoretical_loss": 3.602098523647414,
      "tokens_seen": 1148256256
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003293380140421264,
      "loss": 3.0239,
      "theoretical_loss": 3.6020794560000473,
      "tokens_seen": 1148321792
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032932798395185557,
      "loss": 3.0413,
      "theoretical_loss": 3.6020603897455414,
      "tokens_seen": 1148387328
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032931795386158475,
      "loss": 2.9567,
      "theoretical_loss": 3.602041324883714,
      "tokens_seen": 1148452864
    },
    {
      "epoch": 0.35,
      "objective/train/advantage_avg": 0.49469423294067383,
      "objective/train/docs_used": 654364,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.058694362640381,
      "objective/train/original_loss": 3.058694362640381,
      "objective/train/theoretical_loss": 3.6020222614143846,
      "objective/train/tokens_used": 1168978400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2447659969329834,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507136583328247,
      "objective/train/weighted_lm_loss": 3.213829755783081,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0461554527282715,
      "theoretical_loss": 3.6020222614143846,
      "tokens_seen": 1148518400
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.000329307923771314,
      "loss": 2.9724,
      "theoretical_loss": 3.6020222614143846,
      "tokens_seen": 1148518400
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003292978936810431,
      "loss": 2.8922,
      "theoretical_loss": 3.602003199337372,
      "tokens_seen": 1148583936
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032928786359077235,
      "loss": 3.0294,
      "theoretical_loss": 3.6019841386524947,
      "tokens_seen": 1148649472
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003292778335005015,
      "loss": 3.0337,
      "theoretical_loss": 3.601965079359572,
      "tokens_seen": 1148715008
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003292678034102307,
      "loss": 2.8526,
      "theoretical_loss": 3.6019460214584225,
      "tokens_seen": 1148780544
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003292577733199599,
      "loss": 3.0099,
      "theoretical_loss": 3.6019269649488654,
      "tokens_seen": 1148846080
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032924774322968907,
      "loss": 2.687,
      "theoretical_loss": 3.60190790983072,
      "tokens_seen": 1148911616
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032923771313941825,
      "loss": 2.8747,
      "theoretical_loss": 3.6018888561038045,
      "tokens_seen": 1148977152
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003292276830491475,
      "loss": 2.4407,
      "theoretical_loss": 3.601869803767939,
      "tokens_seen": 1149042688
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003292176529588766,
      "loss": 2.9407,
      "theoretical_loss": 3.601850752822942,
      "tokens_seen": 1149108224
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032920762286860585,
      "loss": 2.894,
      "theoretical_loss": 3.601831703268633,
      "tokens_seen": 1149173760
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.000329197592778335,
      "loss": 2.8075,
      "theoretical_loss": 3.6018126551048306,
      "tokens_seen": 1149239296
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003291875626880642,
      "loss": 2.8382,
      "theoretical_loss": 3.601793608331354,
      "tokens_seen": 1149304832
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003291775325977934,
      "loss": 3.0142,
      "theoretical_loss": 3.601774562948024,
      "tokens_seen": 1149370368
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003291675025075226,
      "loss": 2.9381,
      "theoretical_loss": 3.6017555189546577,
      "tokens_seen": 1149435904
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032915747241725175,
      "loss": 2.8582,
      "theoretical_loss": 3.6017364763510757,
      "tokens_seen": 1149501440
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032914744232698094,
      "loss": 2.7562,
      "theoretical_loss": 3.6017174351370973,
      "tokens_seen": 1149566976
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003291374122367101,
      "loss": 2.8231,
      "theoretical_loss": 3.6016983953125417,
      "tokens_seen": 1149632512
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032912738214643935,
      "loss": 3.0828,
      "theoretical_loss": 3.601679356877228,
      "tokens_seen": 1149698048
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003291173520561685,
      "loss": 2.8569,
      "theoretical_loss": 3.601660319830976,
      "tokens_seen": 1149763584
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003291073219658977,
      "loss": 2.9217,
      "theoretical_loss": 3.6016412841736054,
      "tokens_seen": 1149829120
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003290972918756269,
      "loss": 2.9543,
      "theoretical_loss": 3.601622249904935,
      "tokens_seen": 1149894656
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003290872617853561,
      "loss": 2.988,
      "theoretical_loss": 3.6016032170247847,
      "tokens_seen": 1149960192
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032907723169508526,
      "loss": 2.5583,
      "theoretical_loss": 3.6015841855329747,
      "tokens_seen": 1150025728
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032906720160481444,
      "loss": 2.8341,
      "theoretical_loss": 3.6015651554293235,
      "tokens_seen": 1150091264
    },
    {
      "epoch": 0.35,
      "objective/train/advantage_avg": 0.4958597421646118,
      "objective/train/docs_used": 654951,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.919132947921753,
      "objective/train/original_loss": 2.919133424758911,
      "objective/train/theoretical_loss": 3.601546126713652,
      "objective/train/tokens_used": 1170616800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24592910706996918,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0508360862731934,
      "objective/train/weighted_lm_loss": 3.0674471855163574,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0437887907028198,
      "theoretical_loss": 3.601546126713652,
      "tokens_seen": 1150156800
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003290571715145436,
      "loss": 2.8321,
      "theoretical_loss": 3.601546126713652,
      "tokens_seen": 1150156800
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032904714142427286,
      "loss": 2.8075,
      "theoretical_loss": 3.601527099385779,
      "tokens_seen": 1150222336
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.000329037111334002,
      "loss": 2.8891,
      "theoretical_loss": 3.6015080734455243,
      "tokens_seen": 1150287872
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003290270812437312,
      "loss": 2.6713,
      "theoretical_loss": 3.601489048892708,
      "tokens_seen": 1150353408
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032901705115346034,
      "loss": 3.1221,
      "theoretical_loss": 3.6014700257271492,
      "tokens_seen": 1150418944
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003290070210631896,
      "loss": 2.9539,
      "theoretical_loss": 3.601451003948669,
      "tokens_seen": 1150484480
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032899699097291876,
      "loss": 3.0598,
      "theoretical_loss": 3.6014319835570863,
      "tokens_seen": 1150550016
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032898696088264794,
      "loss": 2.7145,
      "theoretical_loss": 3.601412964552221,
      "tokens_seen": 1150615552
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003289769307923771,
      "loss": 2.8423,
      "theoretical_loss": 3.601393946933893,
      "tokens_seen": 1150681088
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003289669007021063,
      "loss": 2.8472,
      "theoretical_loss": 3.601374930701923,
      "tokens_seen": 1150746624
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003289568706118355,
      "loss": 2.8762,
      "theoretical_loss": 3.60135591585613,
      "tokens_seen": 1150812160
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003289468405215647,
      "loss": 2.9916,
      "theoretical_loss": 3.6013369023963344,
      "tokens_seen": 1150877696
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032893681043129385,
      "loss": 3.0374,
      "theoretical_loss": 3.6013178903223566,
      "tokens_seen": 1150943232
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003289267803410231,
      "loss": 2.8051,
      "theoretical_loss": 3.601298879634016,
      "tokens_seen": 1151008768
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032891675025075226,
      "loss": 2.8643,
      "theoretical_loss": 3.6012798703311337,
      "tokens_seen": 1151074304
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032890672016048145,
      "loss": 2.9017,
      "theoretical_loss": 3.6012608624135294,
      "tokens_seen": 1151139840
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003288966900702106,
      "loss": 2.7859,
      "theoretical_loss": 3.601241855881023,
      "tokens_seen": 1151205376
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003288866599799398,
      "loss": 2.8812,
      "theoretical_loss": 3.6012228507334347,
      "tokens_seen": 1151270912
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.000328876629889669,
      "loss": 2.9192,
      "theoretical_loss": 3.601203846970585,
      "tokens_seen": 1151336448
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003288665997993982,
      "loss": 2.7098,
      "theoretical_loss": 3.6011848445922943,
      "tokens_seen": 1151401984
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032885656970912735,
      "loss": 2.7382,
      "theoretical_loss": 3.6011658435983827,
      "tokens_seen": 1151467520
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003288465396188566,
      "loss": 3.0284,
      "theoretical_loss": 3.601146843988671,
      "tokens_seen": 1151533056
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003288365095285857,
      "loss": 2.8275,
      "theoretical_loss": 3.6011278457629787,
      "tokens_seen": 1151598592
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032882647943831495,
      "loss": 2.642,
      "theoretical_loss": 3.601108848921127,
      "tokens_seen": 1151664128
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003288164493480442,
      "loss": 2.8829,
      "theoretical_loss": 3.6010898534629363,
      "tokens_seen": 1151729664
    },
    {
      "epoch": 0.35,
      "objective/train/advantage_avg": 0.492430716753006,
      "objective/train/docs_used": 656079,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.4873809814453125,
      "objective/train/original_loss": 2.4873814582824707,
      "objective/train/theoretical_loss": 3.601070859388227,
      "objective/train/tokens_used": 1172255200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24256907403469086,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0504759550094604,
      "objective/train/weighted_lm_loss": 2.613170862197876,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0396549701690674,
      "theoretical_loss": 3.601070859388227,
      "tokens_seen": 1151795200
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003288064192577733,
      "loss": 2.8852,
      "theoretical_loss": 3.601070859388227,
      "tokens_seen": 1151795200
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032879638916750255,
      "loss": 2.8757,
      "theoretical_loss": 3.601051866696819,
      "tokens_seen": 1151860736
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003287863590772317,
      "loss": 2.8964,
      "theoretical_loss": 3.601032875388534,
      "tokens_seen": 1151926272
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003287763289869609,
      "loss": 2.9721,
      "theoretical_loss": 3.6010138854631917,
      "tokens_seen": 1151991808
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003287662988966901,
      "loss": 2.7914,
      "theoretical_loss": 3.6009948969206134,
      "tokens_seen": 1152057344
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032875626880641927,
      "loss": 2.6574,
      "theoretical_loss": 3.6009759097606193,
      "tokens_seen": 1152122880
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032874623871614845,
      "loss": 2.779,
      "theoretical_loss": 3.6009569239830306,
      "tokens_seen": 1152188416
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003287362086258777,
      "loss": 3.0419,
      "theoretical_loss": 3.600937939587667,
      "tokens_seen": 1152253952
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003287261785356068,
      "loss": 3.0339,
      "theoretical_loss": 3.60091895657435,
      "tokens_seen": 1152319488
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032871614844533605,
      "loss": 3.0737,
      "theoretical_loss": 3.6008999749429007,
      "tokens_seen": 1152385024
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003287061183550652,
      "loss": 2.988,
      "theoretical_loss": 3.600880994693139,
      "tokens_seen": 1152450560
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003286960882647944,
      "loss": 2.7938,
      "theoretical_loss": 3.6008620158248865,
      "tokens_seen": 1152516096
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003286860581745236,
      "loss": 3.0142,
      "theoretical_loss": 3.6008430383379637,
      "tokens_seen": 1152581632
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003286760280842528,
      "loss": 2.7938,
      "theoretical_loss": 3.600824062232192,
      "tokens_seen": 1152647168
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032866599799398195,
      "loss": 2.7851,
      "theoretical_loss": 3.600805087507392,
      "tokens_seen": 1152712704
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032865596790371114,
      "loss": 2.9273,
      "theoretical_loss": 3.600786114163385,
      "tokens_seen": 1152778240
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003286459378134403,
      "loss": 2.7735,
      "theoretical_loss": 3.600767142199991,
      "tokens_seen": 1152843776
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032863590772316955,
      "loss": 2.9401,
      "theoretical_loss": 3.600748171617033,
      "tokens_seen": 1152909312
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003286258776328987,
      "loss": 2.9033,
      "theoretical_loss": 3.60072920241433,
      "tokens_seen": 1152974848
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003286158475426279,
      "loss": 2.897,
      "theoretical_loss": 3.6007102345917046,
      "tokens_seen": 1153040384
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003286058174523571,
      "loss": 2.9427,
      "theoretical_loss": 3.600691268148977,
      "tokens_seen": 1153105920
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003285957873620863,
      "loss": 2.9239,
      "theoretical_loss": 3.600672303085969,
      "tokens_seen": 1153171456
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032858575727181546,
      "loss": 2.7671,
      "theoretical_loss": 3.600653339402502,
      "tokens_seen": 1153236992
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032857572718154464,
      "loss": 2.8273,
      "theoretical_loss": 3.600634377098397,
      "tokens_seen": 1153302528
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003285656970912738,
      "loss": 3.0716,
      "theoretical_loss": 3.6006154161734747,
      "tokens_seen": 1153368064
    },
    {
      "epoch": 0.35,
      "objective/train/advantage_avg": 0.4940102994441986,
      "objective/train/docs_used": 656588,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.126746892929077,
      "objective/train/original_loss": 2.126746654510498,
      "objective/train/theoretical_loss": 3.6005964566275575,
      "objective/train/tokens_used": 1173893600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24409544467926025,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506417751312256,
      "objective/train/weighted_lm_loss": 2.2344069480895996,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0462769269943237,
      "theoretical_loss": 3.6005964566275575,
      "tokens_seen": 1153433600
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032855566700100306,
      "loss": 2.5882,
      "theoretical_loss": 3.6005964566275575,
      "tokens_seen": 1153433600
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003285456369107322,
      "loss": 2.9156,
      "theoretical_loss": 3.600577498460466,
      "tokens_seen": 1153499136
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003285356068204614,
      "loss": 2.7626,
      "theoretical_loss": 3.600558541672022,
      "tokens_seen": 1153564672
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032852557673019054,
      "loss": 2.8846,
      "theoretical_loss": 3.6005395862620464,
      "tokens_seen": 1153630208
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003285155466399198,
      "loss": 2.997,
      "theoretical_loss": 3.600520632230361,
      "tokens_seen": 1153695744
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032850551654964896,
      "loss": 3.2447,
      "theoretical_loss": 3.6005016795767877,
      "tokens_seen": 1153761280
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032849548645937814,
      "loss": 2.7742,
      "theoretical_loss": 3.600482728301148,
      "tokens_seen": 1153826816
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003284854563691073,
      "loss": 2.9401,
      "theoretical_loss": 3.6004637784032623,
      "tokens_seen": 1153892352
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003284754262788365,
      "loss": 3.0247,
      "theoretical_loss": 3.600444829882953,
      "tokens_seen": 1153957888
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003284653961885657,
      "loss": 2.827,
      "theoretical_loss": 3.600425882740042,
      "tokens_seen": 1154023424
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003284553660982949,
      "loss": 2.9554,
      "theoretical_loss": 3.6004069369743505,
      "tokens_seen": 1154088960
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032844533600802405,
      "loss": 2.8913,
      "theoretical_loss": 3.6003879925857003,
      "tokens_seen": 1154154496
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003284353059177533,
      "loss": 2.7404,
      "theoretical_loss": 3.6003690495739136,
      "tokens_seen": 1154220032
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032842527582748246,
      "loss": 2.9344,
      "theoretical_loss": 3.600350107938812,
      "tokens_seen": 1154285568
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032841524573721165,
      "loss": 2.9195,
      "theoretical_loss": 3.6003311676802165,
      "tokens_seen": 1154351104
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003284052156469408,
      "loss": 2.8993,
      "theoretical_loss": 3.6003122287979497,
      "tokens_seen": 1154416640
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032839518555667,
      "loss": 2.6731,
      "theoretical_loss": 3.600293291291833,
      "tokens_seen": 1154482176
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003283851554663992,
      "loss": 3.1147,
      "theoretical_loss": 3.600274355161689,
      "tokens_seen": 1154547712
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003283751253761284,
      "loss": 3.005,
      "theoretical_loss": 3.600255420407339,
      "tokens_seen": 1154613248
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032836509528585755,
      "loss": 2.7958,
      "theoretical_loss": 3.6002364870286057,
      "tokens_seen": 1154678784
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003283550651955868,
      "loss": 2.9454,
      "theoretical_loss": 3.60021755502531,
      "tokens_seen": 1154744320
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003283450351053159,
      "loss": 2.9544,
      "theoretical_loss": 3.600198624397274,
      "tokens_seen": 1154809856
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032833500501504515,
      "loss": 2.8893,
      "theoretical_loss": 3.6001796951443206,
      "tokens_seen": 1154875392
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032832497492477433,
      "loss": 2.8027,
      "theoretical_loss": 3.6001607672662717,
      "tokens_seen": 1154940928
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003283149448345035,
      "loss": 3.033,
      "theoretical_loss": 3.600141840762949,
      "tokens_seen": 1155006464
    },
    {
      "epoch": 0.35,
      "objective/train/advantage_avg": 0.4931885898113251,
      "objective/train/docs_used": 657761,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.5654172897338867,
      "objective/train/original_loss": 2.5654170513153076,
      "objective/train/theoretical_loss": 3.600122915634175,
      "objective/train/tokens_used": 1175532000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2433491051197052,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505558252334595,
      "objective/train/weighted_lm_loss": 2.6949381828308105,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0396549701690674,
      "theoretical_loss": 3.600122915634175,
      "tokens_seen": 1155072000
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003283049147442327,
      "loss": 2.9414,
      "theoretical_loss": 3.600122915634175,
      "tokens_seen": 1155072000
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003282948846539619,
      "loss": 2.9454,
      "theoretical_loss": 3.600103991879772,
      "tokens_seen": 1155137536
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032828485456369105,
      "loss": 2.947,
      "theoretical_loss": 3.6000850694995616,
      "tokens_seen": 1155203072
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003282748244734203,
      "loss": 2.8781,
      "theoretical_loss": 3.600066148493367,
      "tokens_seen": 1155268608
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003282647943831494,
      "loss": 2.5888,
      "theoretical_loss": 3.6000472288610097,
      "tokens_seen": 1155334144
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032825476429287865,
      "loss": 2.8675,
      "theoretical_loss": 3.600028310602312,
      "tokens_seen": 1155399680
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032824473420260783,
      "loss": 3.0554,
      "theoretical_loss": 3.600009393717097,
      "tokens_seen": 1155465216
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.000328234704112337,
      "loss": 2.9345,
      "theoretical_loss": 3.5999904782051866,
      "tokens_seen": 1155530752
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003282246740220662,
      "loss": 2.7929,
      "theoretical_loss": 3.599971564066403,
      "tokens_seen": 1155596288
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003282146439317954,
      "loss": 2.8442,
      "theoretical_loss": 3.599952651300569,
      "tokens_seen": 1155661824
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032820461384152456,
      "loss": 2.8033,
      "theoretical_loss": 3.599933739907507,
      "tokens_seen": 1155727360
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003281945837512538,
      "loss": 2.754,
      "theoretical_loss": 3.59991482988704,
      "tokens_seen": 1155792896
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003281845536609829,
      "loss": 2.6579,
      "theoretical_loss": 3.5998959212389896,
      "tokens_seen": 1155858432
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032817452357071215,
      "loss": 2.8525,
      "theoretical_loss": 3.5998770139631793,
      "tokens_seen": 1155923968
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003281644934804413,
      "loss": 3.0683,
      "theoretical_loss": 3.5998581080594305,
      "tokens_seen": 1155989504
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003281544633901705,
      "loss": 2.9011,
      "theoretical_loss": 3.5998392035275675,
      "tokens_seen": 1156055040
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003281444332998997,
      "loss": 2.868,
      "theoretical_loss": 3.5998203003674116,
      "tokens_seen": 1156120576
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003281344032096289,
      "loss": 2.9364,
      "theoretical_loss": 3.5998013985787862,
      "tokens_seen": 1156186112
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032812437311935806,
      "loss": 3.0453,
      "theoretical_loss": 3.599782498161514,
      "tokens_seen": 1156251648
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003281143430290873,
      "loss": 2.7941,
      "theoretical_loss": 3.599763599115418,
      "tokens_seen": 1156317184
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003281043129388164,
      "loss": 3.1228,
      "theoretical_loss": 3.5997447014403203,
      "tokens_seen": 1156382720
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032809428284854566,
      "loss": 2.774,
      "theoretical_loss": 3.5997258051360443,
      "tokens_seen": 1156448256
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003280842527582748,
      "loss": 2.813,
      "theoretical_loss": 3.5997069102024124,
      "tokens_seen": 1156513792
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.000328074222668004,
      "loss": 2.8104,
      "theoretical_loss": 3.5996880166392486,
      "tokens_seen": 1156579328
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032806419257773326,
      "loss": 2.929,
      "theoretical_loss": 3.5996691244463745,
      "tokens_seen": 1156644864
    },
    {
      "epoch": 0.35,
      "objective/train/advantage_avg": 0.4947555661201477,
      "objective/train/docs_used": 658495,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.2309415340423584,
      "objective/train/original_loss": 3.2309412956237793,
      "objective/train/theoretical_loss": 3.5996502336236142,
      "objective/train/tokens_used": 1177170400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24485407769680023,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.05072021484375,
      "objective/train/weighted_lm_loss": 3.3947880268096924,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0417574644088745,
      "theoretical_loss": 3.5996502336236142,
      "tokens_seen": 1156710400
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003280541624874624,
      "loss": 2.8828,
      "theoretical_loss": 3.5996502336236142,
      "tokens_seen": 1156710400
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003280441323971916,
      "loss": 2.9309,
      "theoretical_loss": 3.59963134417079,
      "tokens_seen": 1156775936
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032803410230692074,
      "loss": 3.0309,
      "theoretical_loss": 3.5996124560877254,
      "tokens_seen": 1156841472
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032802407221665,
      "loss": 3.1143,
      "theoretical_loss": 3.5995935693742434,
      "tokens_seen": 1156907008
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032801404212637916,
      "loss": 2.539,
      "theoretical_loss": 3.5995746840301663,
      "tokens_seen": 1156972544
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032800401203610834,
      "loss": 2.8157,
      "theoretical_loss": 3.5995558000553185,
      "tokens_seen": 1157038080
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003279939819458375,
      "loss": 2.7844,
      "theoretical_loss": 3.5995369174495226,
      "tokens_seen": 1157103616
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003279839518555667,
      "loss": 2.9399,
      "theoretical_loss": 3.5995180362126016,
      "tokens_seen": 1157169152
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003279739217652959,
      "loss": 2.9038,
      "theoretical_loss": 3.599499156344379,
      "tokens_seen": 1157234688
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003279638916750251,
      "loss": 2.6326,
      "theoretical_loss": 3.5994802778446786,
      "tokens_seen": 1157300224
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032795386158475425,
      "loss": 2.8728,
      "theoretical_loss": 3.5994614007133228,
      "tokens_seen": 1157365760
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003279438314944835,
      "loss": 2.7632,
      "theoretical_loss": 3.5994425249501347,
      "tokens_seen": 1157431296
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032793380140421266,
      "loss": 2.6131,
      "theoretical_loss": 3.599423650554939,
      "tokens_seen": 1157496832
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032792377131394185,
      "loss": 2.7877,
      "theoretical_loss": 3.599404777527558,
      "tokens_seen": 1157562368
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.000327913741223671,
      "loss": 2.8645,
      "theoretical_loss": 3.599385905867816,
      "tokens_seen": 1157627904
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003279037111334002,
      "loss": 3.0612,
      "theoretical_loss": 3.5993670355755363,
      "tokens_seen": 1157693440
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003278936810431294,
      "loss": 2.8465,
      "theoretical_loss": 3.5993481666505414,
      "tokens_seen": 1157758976
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003278836509528586,
      "loss": 3.0832,
      "theoretical_loss": 3.5993292990926555,
      "tokens_seen": 1157824512
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032787362086258775,
      "loss": 2.8486,
      "theoretical_loss": 3.599310432901703,
      "tokens_seen": 1157890048
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.000327863590772317,
      "loss": 2.8013,
      "theoretical_loss": 3.599291568077506,
      "tokens_seen": 1157955584
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003278535606820461,
      "loss": 2.8286,
      "theoretical_loss": 3.5992727046198896,
      "tokens_seen": 1158021120
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032784353059177535,
      "loss": 2.7082,
      "theoretical_loss": 3.599253842528676,
      "tokens_seen": 1158086656
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032783350050150453,
      "loss": 2.9315,
      "theoretical_loss": 3.599234981803691,
      "tokens_seen": 1158152192
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003278234704112337,
      "loss": 3.0083,
      "theoretical_loss": 3.599216122444756,
      "tokens_seen": 1158217728
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003278134403209629,
      "loss": 2.7939,
      "theoretical_loss": 3.5991972644516954,
      "tokens_seen": 1158283264
    },
    {
      "epoch": 0.35,
      "objective/train/advantage_avg": 0.49521058797836304,
      "objective/train/docs_used": 659118,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.734828233718872,
      "objective/train/original_loss": 2.734827995300293,
      "objective/train/theoretical_loss": 3.5991784078243345,
      "objective/train/tokens_used": 1178808800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24528232216835022,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507678985595703,
      "objective/train/weighted_lm_loss": 2.873640537261963,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0417574644088745,
      "theoretical_loss": 3.5991784078243345,
      "tokens_seen": 1158348800
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032780341023069213,
      "loss": 2.7742,
      "theoretical_loss": 3.5991784078243345,
      "tokens_seen": 1158348800
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032779338014042125,
      "loss": 3.1435,
      "theoretical_loss": 3.5991595525624955,
      "tokens_seen": 1158414336
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003277833500501505,
      "loss": 2.7749,
      "theoretical_loss": 3.5991406986660026,
      "tokens_seen": 1158479872
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003277733199598796,
      "loss": 2.7018,
      "theoretical_loss": 3.5991218461346803,
      "tokens_seen": 1158545408
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032776328986960885,
      "loss": 2.9718,
      "theoretical_loss": 3.5991029949683515,
      "tokens_seen": 1158610944
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032775325977933803,
      "loss": 2.8078,
      "theoretical_loss": 3.5990841451668416,
      "tokens_seen": 1158676480
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003277432296890672,
      "loss": 2.8239,
      "theoretical_loss": 3.5990652967299734,
      "tokens_seen": 1158742016
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003277331995987964,
      "loss": 2.877,
      "theoretical_loss": 3.5990464496575716,
      "tokens_seen": 1158807552
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003277231695085256,
      "loss": 2.8859,
      "theoretical_loss": 3.5990276039494598,
      "tokens_seen": 1158873088
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032771313941825476,
      "loss": 2.9173,
      "theoretical_loss": 3.5990087596054625,
      "tokens_seen": 1158938624
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.000327703109327984,
      "loss": 3.03,
      "theoretical_loss": 3.5989899166254036,
      "tokens_seen": 1159004160
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003276930792377131,
      "loss": 2.946,
      "theoretical_loss": 3.598971075009107,
      "tokens_seen": 1159069696
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032768304914744235,
      "loss": 2.8643,
      "theoretical_loss": 3.5989522347563976,
      "tokens_seen": 1159135232
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003276730190571715,
      "loss": 3.1447,
      "theoretical_loss": 3.598933395867099,
      "tokens_seen": 1159200768
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003276629889669007,
      "loss": 3.1015,
      "theoretical_loss": 3.598914558341036,
      "tokens_seen": 1159266304
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003276529588766299,
      "loss": 2.8751,
      "theoretical_loss": 3.5988957221780318,
      "tokens_seen": 1159331840
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003276429287863591,
      "loss": 3.1111,
      "theoretical_loss": 3.5988768873779122,
      "tokens_seen": 1159397376
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032763289869608826,
      "loss": 2.914,
      "theoretical_loss": 3.598858053940501,
      "tokens_seen": 1159462912
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003276228686058175,
      "loss": 2.8826,
      "theoretical_loss": 3.598839221865622,
      "tokens_seen": 1159528448
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003276128385155466,
      "loss": 2.9658,
      "theoretical_loss": 3.5988203911531005,
      "tokens_seen": 1159593984
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032760280842527586,
      "loss": 2.7103,
      "theoretical_loss": 3.5988015618027602,
      "tokens_seen": 1159659520
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.000327592778335005,
      "loss": 2.6761,
      "theoretical_loss": 3.598782733814426,
      "tokens_seen": 1159725056
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003275827482447342,
      "loss": 2.848,
      "theoretical_loss": 3.5987639071879225,
      "tokens_seen": 1159790592
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003275727181544634,
      "loss": 2.8598,
      "theoretical_loss": 3.5987450819230737,
      "tokens_seen": 1159856128
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003275626880641926,
      "loss": 2.9525,
      "theoretical_loss": 3.5987262580197052,
      "tokens_seen": 1159921664
    },
    {
      "epoch": 0.35,
      "objective/train/advantage_avg": 0.4953756332397461,
      "objective/train/docs_used": 660623,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.576237201690674,
      "objective/train/original_loss": 2.5762369632720947,
      "objective/train/theoretical_loss": 3.5987074354776407,
      "objective/train/tokens_used": 1180447200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24545502662658691,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507853031158447,
      "objective/train/weighted_lm_loss": 2.7069735527038574,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.045280933380127,
      "theoretical_loss": 3.5987074354776407,
      "tokens_seen": 1159987200
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032755265797392176,
      "loss": 2.8205,
      "theoretical_loss": 3.5987074354776407,
      "tokens_seen": 1159987200
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032754262788365094,
      "loss": 3.045,
      "theoretical_loss": 3.598688614296705,
      "tokens_seen": 1160052736
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003275325977933801,
      "loss": 2.7518,
      "theoretical_loss": 3.598669794476723,
      "tokens_seen": 1160118272
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032752256770310936,
      "loss": 2.9508,
      "theoretical_loss": 3.59865097601752,
      "tokens_seen": 1160183808
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003275125376128385,
      "loss": 2.7842,
      "theoretical_loss": 3.598632158918919,
      "tokens_seen": 1160249344
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003275025075225677,
      "loss": 2.7329,
      "theoretical_loss": 3.598613343180747,
      "tokens_seen": 1160314880
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032749247743229685,
      "loss": 2.9075,
      "theoretical_loss": 3.598594528802827,
      "tokens_seen": 1160380416
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003274824473420261,
      "loss": 2.7688,
      "theoretical_loss": 3.5985757157849854,
      "tokens_seen": 1160445952
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032747241725175527,
      "loss": 2.7686,
      "theoretical_loss": 3.5985569041270455,
      "tokens_seen": 1160511488
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032746238716148445,
      "loss": 2.8857,
      "theoretical_loss": 3.598538093828833,
      "tokens_seen": 1160577024
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032745235707121363,
      "loss": 2.9668,
      "theoretical_loss": 3.598519284890173,
      "tokens_seen": 1160642560
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032744232698094286,
      "loss": 2.7681,
      "theoretical_loss": 3.59850047731089,
      "tokens_seen": 1160708096
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.000327432296890672,
      "loss": 2.886,
      "theoretical_loss": 3.598481671090809,
      "tokens_seen": 1160773632
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003274222668004012,
      "loss": 2.7953,
      "theoretical_loss": 3.598462866229756,
      "tokens_seen": 1160839168
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032741223671013035,
      "loss": 3.0347,
      "theoretical_loss": 3.5984440627275553,
      "tokens_seen": 1160904704
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003274022066198596,
      "loss": 2.8008,
      "theoretical_loss": 3.5984252605840323,
      "tokens_seen": 1160970240
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032739217652958877,
      "loss": 2.8241,
      "theoretical_loss": 3.598406459799011,
      "tokens_seen": 1161035776
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032738214643931795,
      "loss": 2.9277,
      "theoretical_loss": 3.5983876603723184,
      "tokens_seen": 1161101312
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032737211634904713,
      "loss": 2.8271,
      "theoretical_loss": 3.5983688623037784,
      "tokens_seen": 1161166848
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003273620862587763,
      "loss": 2.9703,
      "theoretical_loss": 3.598350065593217,
      "tokens_seen": 1161232384
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003273520561685055,
      "loss": 2.8439,
      "theoretical_loss": 3.598331270240459,
      "tokens_seen": 1161297920
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032734202607823473,
      "loss": 2.8492,
      "theoretical_loss": 3.5983124762453293,
      "tokens_seen": 1161363456
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032733199598796386,
      "loss": 2.7999,
      "theoretical_loss": 3.5982936836076544,
      "tokens_seen": 1161428992
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003273219658976931,
      "loss": 2.8928,
      "theoretical_loss": 3.598274892327259,
      "tokens_seen": 1161494528
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032731193580742233,
      "loss": 2.839,
      "theoretical_loss": 3.598256102403968,
      "tokens_seen": 1161560064
    },
    {
      "epoch": 0.35,
      "objective/train/advantage_avg": 0.49305057525634766,
      "objective/train/docs_used": 661405,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.471402883529663,
      "objective/train/original_loss": 2.471402883529663,
      "objective/train/theoretical_loss": 3.5982373138376076,
      "objective/train/tokens_used": 1182085600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24318672716617584,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505411624908447,
      "objective/train/weighted_lm_loss": 2.5961039066314697,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 3.5982373138376076,
      "tokens_seen": 1161625600
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032730190571715145,
      "loss": 2.8419,
      "theoretical_loss": 3.5982373138376076,
      "tokens_seen": 1161625600
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003272918756268807,
      "loss": 3.0815,
      "theoretical_loss": 3.5982185266280036,
      "tokens_seen": 1161691136
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003272818455366098,
      "loss": 2.8085,
      "theoretical_loss": 3.59819974077498,
      "tokens_seen": 1161756672
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032727181544633905,
      "loss": 2.955,
      "theoretical_loss": 3.5981809562783633,
      "tokens_seen": 1161822208
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032726178535606823,
      "loss": 2.9146,
      "theoretical_loss": 3.5981621731379794,
      "tokens_seen": 1161887744
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003272517552657974,
      "loss": 2.9146,
      "theoretical_loss": 3.5981433913536534,
      "tokens_seen": 1161953280
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003272417251755266,
      "loss": 2.9295,
      "theoretical_loss": 3.598124610925211,
      "tokens_seen": 1162018816
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003272316950852558,
      "loss": 3.0132,
      "theoretical_loss": 3.598105831852478,
      "tokens_seen": 1162084352
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032722166499498496,
      "loss": 2.8967,
      "theoretical_loss": 3.59808705413528,
      "tokens_seen": 1162149888
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003272116349047142,
      "loss": 3.0337,
      "theoretical_loss": 3.598068277773442,
      "tokens_seen": 1162215424
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003272016048144433,
      "loss": 2.8175,
      "theoretical_loss": 3.5980495027667905,
      "tokens_seen": 1162280960
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032719157472417256,
      "loss": 3.3339,
      "theoretical_loss": 3.598030729115152,
      "tokens_seen": 1162346496
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003271815446339017,
      "loss": 3.0155,
      "theoretical_loss": 3.5980119568183504,
      "tokens_seen": 1162412032
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003271715145436309,
      "loss": 3.0181,
      "theoretical_loss": 3.5979931858762138,
      "tokens_seen": 1162477568
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003271614844533601,
      "loss": 2.7135,
      "theoretical_loss": 3.597974416288566,
      "tokens_seen": 1162543104
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003271514543630893,
      "loss": 2.8082,
      "theoretical_loss": 3.597955648055234,
      "tokens_seen": 1162608640
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032714142427281846,
      "loss": 2.7132,
      "theoretical_loss": 3.5979368811760444,
      "tokens_seen": 1162674176
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003271313941825477,
      "loss": 2.963,
      "theoretical_loss": 3.5979181156508218,
      "tokens_seen": 1162739712
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003271213640922768,
      "loss": 2.8334,
      "theoretical_loss": 3.5978993514793927,
      "tokens_seen": 1162805248
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032711133400200606,
      "loss": 2.8579,
      "theoretical_loss": 3.5978805886615834,
      "tokens_seen": 1162870784
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003271013039117352,
      "loss": 3.0803,
      "theoretical_loss": 3.59786182719722,
      "tokens_seen": 1162936320
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003270912738214644,
      "loss": 3.0927,
      "theoretical_loss": 3.597843067086128,
      "tokens_seen": 1163001856
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003270812437311936,
      "loss": 2.7269,
      "theoretical_loss": 3.597824308328134,
      "tokens_seen": 1163067392
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003270712136409228,
      "loss": 2.7231,
      "theoretical_loss": 3.5978055509230646,
      "tokens_seen": 1163132928
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032706118355065196,
      "loss": 3.084,
      "theoretical_loss": 3.597786794870745,
      "tokens_seen": 1163198464
    },
    {
      "epoch": 0.35,
      "objective/train/advantage_avg": 0.49491509795188904,
      "objective/train/docs_used": 662532,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.7950148582458496,
      "objective/train/original_loss": 2.7950146198272705,
      "objective/train/theoretical_loss": 3.597768040171002,
      "objective/train/tokens_used": 1183724000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24500247836112976,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507370233535767,
      "objective/train/weighted_lm_loss": 2.936765193939209,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0396549701690674,
      "theoretical_loss": 3.597768040171002,
      "tokens_seen": 1163264000
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032705115346038114,
      "loss": 2.9613,
      "theoretical_loss": 3.597768040171002,
      "tokens_seen": 1163264000
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003270411233701103,
      "loss": 2.7243,
      "theoretical_loss": 3.597749286823662,
      "tokens_seen": 1163329536
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032703109327983956,
      "loss": 3.0316,
      "theoretical_loss": 3.5977305348285515,
      "tokens_seen": 1163395072
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003270210631895687,
      "loss": 2.8197,
      "theoretical_loss": 3.5977117841854955,
      "tokens_seen": 1163460608
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003270110330992979,
      "loss": 2.9212,
      "theoretical_loss": 3.597693034894322,
      "tokens_seen": 1163526144
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032700100300902705,
      "loss": 2.8564,
      "theoretical_loss": 3.597674286954857,
      "tokens_seen": 1163591680
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003269909729187563,
      "loss": 2.7702,
      "theoretical_loss": 3.597655540366926,
      "tokens_seen": 1163657216
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032698094282848547,
      "loss": 2.7888,
      "theoretical_loss": 3.5976367951303563,
      "tokens_seen": 1163722752
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032697091273821465,
      "loss": 2.7537,
      "theoretical_loss": 3.5976180512449742,
      "tokens_seen": 1163788288
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032696088264794383,
      "loss": 2.9082,
      "theoretical_loss": 3.597599308710606,
      "tokens_seen": 1163853824
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032695085255767306,
      "loss": 2.7397,
      "theoretical_loss": 3.5975805675270784,
      "tokens_seen": 1163919360
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003269408224674022,
      "loss": 2.9092,
      "theoretical_loss": 3.5975618276942187,
      "tokens_seen": 1163984896
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003269307923771314,
      "loss": 2.8955,
      "theoretical_loss": 3.597543089211853,
      "tokens_seen": 1164050432
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032692076228686055,
      "loss": 3.0518,
      "theoretical_loss": 3.597524352079807,
      "tokens_seen": 1164115968
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003269107321965898,
      "loss": 3.0636,
      "theoretical_loss": 3.5975056162979087,
      "tokens_seen": 1164181504
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032690070210631897,
      "loss": 3.1221,
      "theoretical_loss": 3.5974868818659838,
      "tokens_seen": 1164247040
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032689067201604815,
      "loss": 2.8969,
      "theoretical_loss": 3.59746814878386,
      "tokens_seen": 1164312576
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032688064192577733,
      "loss": 2.7382,
      "theoretical_loss": 3.5974494170513642,
      "tokens_seen": 1164378112
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003268706118355065,
      "loss": 3.0365,
      "theoretical_loss": 3.597430686668322,
      "tokens_seen": 1164443648
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003268605817452357,
      "loss": 2.6606,
      "theoretical_loss": 3.5974119576345607,
      "tokens_seen": 1164509184
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032685055165496493,
      "loss": 2.9452,
      "theoretical_loss": 3.597393229949908,
      "tokens_seen": 1164574720
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032684052156469406,
      "loss": 2.6907,
      "theoretical_loss": 3.5973745036141898,
      "tokens_seen": 1164640256
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003268304914744233,
      "loss": 3.0827,
      "theoretical_loss": 3.597355778627233,
      "tokens_seen": 1164705792
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003268204613841524,
      "loss": 2.8949,
      "theoretical_loss": 3.597337054988866,
      "tokens_seen": 1164771328
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032681043129388165,
      "loss": 2.817,
      "theoretical_loss": 3.597318332698914,
      "tokens_seen": 1164836864
    },
    {
      "epoch": 0.35,
      "objective/train/advantage_avg": 0.4948095977306366,
      "objective/train/docs_used": 663185,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.7626781463623047,
      "objective/train/original_loss": 2.762678384780884,
      "objective/train/theoretical_loss": 3.5972996117572054,
      "objective/train/tokens_used": 1185362400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2449006289243698,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507259368896484,
      "objective/train/weighted_lm_loss": 2.902876853942871,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 3.5972996117572054,
      "tokens_seen": 1164902400
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032680040120361084,
      "loss": 2.9154,
      "theoretical_loss": 3.5972996117572054,
      "tokens_seen": 1164902400
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032679037111334,
      "loss": 2.8716,
      "theoretical_loss": 3.5972808921635666,
      "tokens_seen": 1164967936
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003267803410230692,
      "loss": 2.9228,
      "theoretical_loss": 3.5972621739178248,
      "tokens_seen": 1165033472
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032677031093279843,
      "loss": 3.0011,
      "theoretical_loss": 3.5972434570198066,
      "tokens_seen": 1165099008
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032676028084252756,
      "loss": 2.8875,
      "theoretical_loss": 3.5972247414693403,
      "tokens_seen": 1165164544
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003267502507522568,
      "loss": 2.9574,
      "theoretical_loss": 3.5972060272662523,
      "tokens_seen": 1165230080
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003267402206619859,
      "loss": 2.7856,
      "theoretical_loss": 3.59718731441037,
      "tokens_seen": 1165295616
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032673019057171516,
      "loss": 2.956,
      "theoretical_loss": 3.5971686029015206,
      "tokens_seen": 1165361152
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032672016048144434,
      "loss": 2.9326,
      "theoretical_loss": 3.597149892739532,
      "tokens_seen": 1165426688
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003267101303911735,
      "loss": 2.804,
      "theoretical_loss": 3.5971311839242306,
      "tokens_seen": 1165492224
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003267001003009027,
      "loss": 3.0556,
      "theoretical_loss": 3.597112476455444,
      "tokens_seen": 1165557760
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003266900702106319,
      "loss": 2.8035,
      "theoretical_loss": 3.5970937703329997,
      "tokens_seen": 1165623296
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032668004012036106,
      "loss": 2.773,
      "theoretical_loss": 3.5970750655567256,
      "tokens_seen": 1165688832
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003266700100300903,
      "loss": 2.9927,
      "theoretical_loss": 3.597056362126448,
      "tokens_seen": 1165754368
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003266599799398194,
      "loss": 3.107,
      "theoretical_loss": 3.5970376600419955,
      "tokens_seen": 1165819904
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032664994984954866,
      "loss": 2.8262,
      "theoretical_loss": 3.5970189593031954,
      "tokens_seen": 1165885440
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032663991975927784,
      "loss": 2.7892,
      "theoretical_loss": 3.597000259909875,
      "tokens_seen": 1165950976
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.000326629889669007,
      "loss": 2.6868,
      "theoretical_loss": 3.5969815618618615,
      "tokens_seen": 1166016512
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003266198595787362,
      "loss": 2.8763,
      "theoretical_loss": 3.596962865158983,
      "tokens_seen": 1166082048
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003266098294884654,
      "loss": 2.8099,
      "theoretical_loss": 3.5969441698010676,
      "tokens_seen": 1166147584
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032659979939819457,
      "loss": 2.8537,
      "theoretical_loss": 3.596925475787942,
      "tokens_seen": 1166213120
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003265897693079238,
      "loss": 3.099,
      "theoretical_loss": 3.596906783119435,
      "tokens_seen": 1166278656
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032657973921765293,
      "loss": 3.0048,
      "theoretical_loss": 3.5968880917953725,
      "tokens_seen": 1166344192
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032656970912738216,
      "loss": 2.661,
      "theoretical_loss": 3.5968694018155842,
      "tokens_seen": 1166409728
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032655967903711135,
      "loss": 2.7861,
      "theoretical_loss": 3.5968507131798972,
      "tokens_seen": 1166475264
    },
    {
      "epoch": 0.35,
      "objective/train/advantage_avg": 0.4915192723274231,
      "objective/train/docs_used": 664585,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.618856191635132,
      "objective/train/original_loss": 2.618856430053711,
      "objective/train/theoretical_loss": 3.5968320258881388,
      "objective/train/tokens_used": 1187000800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24168150126934052,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0503803491592407,
      "objective/train/weighted_lm_loss": 2.7508347034454346,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0408108234405518,
      "theoretical_loss": 3.5968320258881388,
      "tokens_seen": 1166540800
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003265496489468405,
      "loss": 2.7606,
      "theoretical_loss": 3.5968320258881388,
      "tokens_seen": 1166540800
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032653961885656976,
      "loss": 2.8881,
      "theoretical_loss": 3.5968133399401374,
      "tokens_seen": 1166606336
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003265295887662989,
      "loss": 2.8512,
      "theoretical_loss": 3.5967946553357213,
      "tokens_seen": 1166671872
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003265195586760281,
      "loss": 2.796,
      "theoretical_loss": 3.5967759720747177,
      "tokens_seen": 1166737408
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032650952858575725,
      "loss": 2.7292,
      "theoretical_loss": 3.5967572901569547,
      "tokens_seen": 1166802944
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003264994984954865,
      "loss": 2.9709,
      "theoretical_loss": 3.5967386095822604,
      "tokens_seen": 1166868480
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032648946840521567,
      "loss": 2.77,
      "theoretical_loss": 3.5967199303504627,
      "tokens_seen": 1166934016
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032647943831494485,
      "loss": 2.7306,
      "theoretical_loss": 3.59670125246139,
      "tokens_seen": 1166999552
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032646940822467403,
      "loss": 3.0962,
      "theoretical_loss": 3.5966825759148704,
      "tokens_seen": 1167065088
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032645937813440326,
      "loss": 2.7582,
      "theoretical_loss": 3.5966639007107313,
      "tokens_seen": 1167130624
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003264493480441324,
      "loss": 2.6194,
      "theoretical_loss": 3.5966452268488016,
      "tokens_seen": 1167196160
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003264393179538616,
      "loss": 2.8416,
      "theoretical_loss": 3.5966265543289095,
      "tokens_seen": 1167261696
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032642928786359075,
      "loss": 2.7304,
      "theoretical_loss": 3.5966078831508828,
      "tokens_seen": 1167327232
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032641925777332,
      "loss": 2.8386,
      "theoretical_loss": 3.5965892133145494,
      "tokens_seen": 1167392768
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032640922768304917,
      "loss": 2.7517,
      "theoretical_loss": 3.5965705448197385,
      "tokens_seen": 1167458304
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032639919759277835,
      "loss": 2.7883,
      "theoretical_loss": 3.5965518776662773,
      "tokens_seen": 1167523840
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032638916750250753,
      "loss": 2.8793,
      "theoretical_loss": 3.596533211853995,
      "tokens_seen": 1167589376
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003263791374122367,
      "loss": 2.7731,
      "theoretical_loss": 3.5965145473827205,
      "tokens_seen": 1167654912
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003263691073219659,
      "loss": 2.8481,
      "theoretical_loss": 3.5964958842522803,
      "tokens_seen": 1167720448
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032635907723169513,
      "loss": 2.8686,
      "theoretical_loss": 3.5964772224625046,
      "tokens_seen": 1167785984
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032634904714142426,
      "loss": 2.543,
      "theoretical_loss": 3.5964585620132214,
      "tokens_seen": 1167851520
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003263390170511535,
      "loss": 2.9241,
      "theoretical_loss": 3.596439902904259,
      "tokens_seen": 1167917056
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003263289869608826,
      "loss": 2.8784,
      "theoretical_loss": 3.5964212451354456,
      "tokens_seen": 1167982592
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032631895687061185,
      "loss": 2.8998,
      "theoretical_loss": 3.59640258870661,
      "tokens_seen": 1168048128
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032630892678034104,
      "loss": 2.8064,
      "theoretical_loss": 3.5963839336175814,
      "tokens_seen": 1168113664
    },
    {
      "epoch": 0.35,
      "objective/train/advantage_avg": 0.494951993227005,
      "objective/train/docs_used": 665171,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.9782660007476807,
      "objective/train/original_loss": 2.9782660007476807,
      "objective/train/theoretical_loss": 3.5963652798681878,
      "objective/train/tokens_used": 1188639200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24502962827682495,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507409572601318,
      "objective/train/weighted_lm_loss": 3.1293485164642334,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 3.5963652798681878,
      "tokens_seen": 1168179200
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003262988966900702,
      "loss": 2.7999,
      "theoretical_loss": 3.5963652798681878,
      "tokens_seen": 1168179200
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003262888665997994,
      "loss": 3.0042,
      "theoretical_loss": 3.596346627458258,
      "tokens_seen": 1168244736
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032627883650952863,
      "loss": 2.8448,
      "theoretical_loss": 3.5963279763876206,
      "tokens_seen": 1168310272
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032626880641925776,
      "loss": 2.881,
      "theoretical_loss": 3.596309326656104,
      "tokens_seen": 1168375808
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.000326258776328987,
      "loss": 2.9066,
      "theoretical_loss": 3.596290678263538,
      "tokens_seen": 1168441344
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003262487462387161,
      "loss": 2.9432,
      "theoretical_loss": 3.5962720312097503,
      "tokens_seen": 1168506880
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032623871614844536,
      "loss": 2.9339,
      "theoretical_loss": 3.5962533854945704,
      "tokens_seen": 1168572416
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032622868605817454,
      "loss": 2.9755,
      "theoretical_loss": 3.5962347411178266,
      "tokens_seen": 1168637952
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003262186559679037,
      "loss": 2.9731,
      "theoretical_loss": 3.5962160980793483,
      "tokens_seen": 1168703488
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003262086258776329,
      "loss": 2.7131,
      "theoretical_loss": 3.596197456378964,
      "tokens_seen": 1168769024
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003261985957873621,
      "loss": 2.8499,
      "theoretical_loss": 3.5961788160165025,
      "tokens_seen": 1168834560
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032618856569709126,
      "loss": 2.8755,
      "theoretical_loss": 3.596160176991794,
      "tokens_seen": 1168900096
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003261785356068205,
      "loss": 2.859,
      "theoretical_loss": 3.596141539304665,
      "tokens_seen": 1168965632
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003261685055165496,
      "loss": 3.0543,
      "theoretical_loss": 3.5961229029549475,
      "tokens_seen": 1169031168
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032615847542627886,
      "loss": 3.0449,
      "theoretical_loss": 3.5961042679424686,
      "tokens_seen": 1169096704
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032614844533600804,
      "loss": 2.8344,
      "theoretical_loss": 3.596085634267058,
      "tokens_seen": 1169162240
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003261384152457372,
      "loss": 3.2041,
      "theoretical_loss": 3.596067001928545,
      "tokens_seen": 1169227776
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003261283851554664,
      "loss": 2.9428,
      "theoretical_loss": 3.596048370926758,
      "tokens_seen": 1169293312
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003261183550651956,
      "loss": 2.9236,
      "theoretical_loss": 3.5960297412615265,
      "tokens_seen": 1169358848
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032610832497492477,
      "loss": 2.6909,
      "theoretical_loss": 3.596011112932681,
      "tokens_seen": 1169424384
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.000326098294884654,
      "loss": 2.7432,
      "theoretical_loss": 3.5959924859400485,
      "tokens_seen": 1169489920
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032608826479438313,
      "loss": 2.8599,
      "theoretical_loss": 3.5959738602834603,
      "tokens_seen": 1169555456
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032607823470411236,
      "loss": 2.8985,
      "theoretical_loss": 3.5959552359627445,
      "tokens_seen": 1169620992
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003260682046138415,
      "loss": 2.76,
      "theoretical_loss": 3.595936612977731,
      "tokens_seen": 1169686528
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003260581745235707,
      "loss": 3.054,
      "theoretical_loss": 3.595917991328249,
      "tokens_seen": 1169752064
    },
    {
      "epoch": 0.35,
      "objective/train/advantage_avg": 0.4958721399307251,
      "objective/train/docs_used": 666471,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.893946886062622,
      "objective/train/original_loss": 2.893946409225464,
      "objective/train/theoretical_loss": 3.595899371014127,
      "objective/train/tokens_used": 1190277600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24593736231327057,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0508372783660889,
      "objective/train/weighted_lm_loss": 3.041038990020752,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0432155132293701,
      "theoretical_loss": 3.595899371014127,
      "tokens_seen": 1169817600
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003260481444332999,
      "loss": 2.9151,
      "theoretical_loss": 3.595899371014127,
      "tokens_seen": 1169817600
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003260381143430291,
      "loss": 2.7026,
      "theoretical_loss": 3.5958807520351965,
      "tokens_seen": 1169883136
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032602808425275827,
      "loss": 3.2527,
      "theoretical_loss": 3.595862134391285,
      "tokens_seen": 1169948672
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032601805416248745,
      "loss": 2.9173,
      "theoretical_loss": 3.5958435180822237,
      "tokens_seen": 1170014208
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032600802407221663,
      "loss": 2.9107,
      "theoretical_loss": 3.595824903107841,
      "tokens_seen": 1170079744
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032599799398194587,
      "loss": 2.9346,
      "theoretical_loss": 3.5958062894679665,
      "tokens_seen": 1170145280
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.000325987963891675,
      "loss": 3.0631,
      "theoretical_loss": 3.5957876771624298,
      "tokens_seen": 1170210816
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032597793380140423,
      "loss": 3.1189,
      "theoretical_loss": 3.5957690661910613,
      "tokens_seen": 1170276352
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003259679037111334,
      "loss": 2.7205,
      "theoretical_loss": 3.5957504565536893,
      "tokens_seen": 1170341888
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003259578736208626,
      "loss": 2.5598,
      "theoretical_loss": 3.5957318482501446,
      "tokens_seen": 1170407424
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003259478435305918,
      "loss": 2.9026,
      "theoretical_loss": 3.595713241280257,
      "tokens_seen": 1170472960
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032593781344032095,
      "loss": 3.0266,
      "theoretical_loss": 3.5956946356438553,
      "tokens_seen": 1170538496
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032592778335005014,
      "loss": 2.9706,
      "theoretical_loss": 3.59567603134077,
      "tokens_seen": 1170604032
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032591775325977937,
      "loss": 2.8718,
      "theoretical_loss": 3.595657428370831,
      "tokens_seen": 1170669568
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003259077231695085,
      "loss": 2.9046,
      "theoretical_loss": 3.5956388267338673,
      "tokens_seen": 1170735104
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032589769307923773,
      "loss": 2.812,
      "theoretical_loss": 3.59562022642971,
      "tokens_seen": 1170800640
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032588766298896686,
      "loss": 2.7396,
      "theoretical_loss": 3.5956016274581883,
      "tokens_seen": 1170866176
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003258776328986961,
      "loss": 2.8424,
      "theoretical_loss": 3.595583029819132,
      "tokens_seen": 1170931712
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003258676028084253,
      "loss": 3.0359,
      "theoretical_loss": 3.5955644335123713,
      "tokens_seen": 1170997248
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032585757271815446,
      "loss": 2.7554,
      "theoretical_loss": 3.5955458385377366,
      "tokens_seen": 1171062784
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032584754262788364,
      "loss": 3.0112,
      "theoretical_loss": 3.595527244895057,
      "tokens_seen": 1171128320
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003258375125376128,
      "loss": 2.8483,
      "theoretical_loss": 3.5955086525841633,
      "tokens_seen": 1171193856
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.000325827482447342,
      "loss": 2.6669,
      "theoretical_loss": 3.5954900616048855,
      "tokens_seen": 1171259392
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.00032581745235707124,
      "loss": 2.792,
      "theoretical_loss": 3.5954714719570537,
      "tokens_seen": 1171324928
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003258074222668004,
      "loss": 3.1317,
      "theoretical_loss": 3.5954528836404975,
      "tokens_seen": 1171390464
    },
    {
      "epoch": 0.35,
      "objective/train/advantage_avg": 0.4940839409828186,
      "objective/train/docs_used": 666886,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.0478076934814453,
      "objective/train/original_loss": 3.047807216644287,
      "objective/train/theoretical_loss": 3.595434296655048,
      "objective/train/tokens_used": 1191916000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.244205504655838,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506497621536255,
      "objective/train/weighted_lm_loss": 3.202256679534912,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0363596677780151,
      "theoretical_loss": 3.595434296655048,
      "tokens_seen": 1171456000
    },
    {
      "epoch": 0.35,
      "learning_rate": 0.0003257973921765296,
      "loss": 2.9154,
      "theoretical_loss": 3.595434296655048,
      "tokens_seen": 1171456000
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032578736208625883,
      "loss": 2.7502,
      "theoretical_loss": 3.595415711000535,
      "tokens_seen": 1171521536
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032577733199598796,
      "loss": 3.0564,
      "theoretical_loss": 3.5953971266767883,
      "tokens_seen": 1171587072
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003257673019057172,
      "loss": 2.6937,
      "theoretical_loss": 3.5953785436836387,
      "tokens_seen": 1171652608
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003257572718154463,
      "loss": 2.7523,
      "theoretical_loss": 3.5953599620209165,
      "tokens_seen": 1171718144
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032574724172517556,
      "loss": 3.0916,
      "theoretical_loss": 3.5953413816884523,
      "tokens_seen": 1171783680
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032573721163490474,
      "loss": 2.9794,
      "theoretical_loss": 3.5953228026860757,
      "tokens_seen": 1171849216
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003257271815446339,
      "loss": 3.0836,
      "theoretical_loss": 3.595304225013618,
      "tokens_seen": 1171914752
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003257171514543631,
      "loss": 2.8659,
      "theoretical_loss": 3.5952856486709086,
      "tokens_seen": 1171980288
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003257071213640923,
      "loss": 2.8522,
      "theoretical_loss": 3.595267073657779,
      "tokens_seen": 1172045824
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032569709127382146,
      "loss": 2.9312,
      "theoretical_loss": 3.5952484999740593,
      "tokens_seen": 1172111360
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003256870611835507,
      "loss": 3.0689,
      "theoretical_loss": 3.59522992761958,
      "tokens_seen": 1172176896
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003256770310932798,
      "loss": 2.9538,
      "theoretical_loss": 3.5952113565941715,
      "tokens_seen": 1172242432
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032566700100300906,
      "loss": 2.7398,
      "theoretical_loss": 3.5951927868976643,
      "tokens_seen": 1172307968
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032565697091273824,
      "loss": 2.7934,
      "theoretical_loss": 3.59517421852989,
      "tokens_seen": 1172373504
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003256469408224674,
      "loss": 2.8896,
      "theoretical_loss": 3.595155651490678,
      "tokens_seen": 1172439040
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003256369107321966,
      "loss": 3.0066,
      "theoretical_loss": 3.5951370857798595,
      "tokens_seen": 1172504576
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003256268806419258,
      "loss": 3.1,
      "theoretical_loss": 3.595118521397265,
      "tokens_seen": 1172570112
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032561685055165497,
      "loss": 2.9337,
      "theoretical_loss": 3.5950999583427263,
      "tokens_seen": 1172635648
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003256068204613842,
      "loss": 2.9907,
      "theoretical_loss": 3.595081396616073,
      "tokens_seen": 1172701184
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032559679037111333,
      "loss": 2.9037,
      "theoretical_loss": 3.595062836217136,
      "tokens_seen": 1172766720
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032558676028084256,
      "loss": 2.7379,
      "theoretical_loss": 3.5950442771457465,
      "tokens_seen": 1172832256
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003255767301905717,
      "loss": 2.9713,
      "theoretical_loss": 3.595025719401735,
      "tokens_seen": 1172897792
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003255667001003009,
      "loss": 2.9855,
      "theoretical_loss": 3.595007162984934,
      "tokens_seen": 1172963328
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003255566700100301,
      "loss": 2.783,
      "theoretical_loss": 3.5949886078951714,
      "tokens_seen": 1173028864
    },
    {
      "epoch": 0.36,
      "objective/train/advantage_avg": 0.49514082074165344,
      "objective/train/docs_used": 667905,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.5253190994262695,
      "objective/train/original_loss": 2.5253186225891113,
      "objective/train/theoretical_loss": 3.594970054132281,
      "objective/train/tokens_used": 1193554400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24521245062351227,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507606267929077,
      "objective/train/weighted_lm_loss": 2.6533918380737305,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0417574644088745,
      "theoretical_loss": 3.594970054132281,
      "tokens_seen": 1173094400
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003255466399197593,
      "loss": 2.8836,
      "theoretical_loss": 3.594970054132281,
      "tokens_seen": 1173094400
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032553660982948847,
      "loss": 3.111,
      "theoretical_loss": 3.5949515016960922,
      "tokens_seen": 1173159936
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032552657973921765,
      "loss": 2.9571,
      "theoretical_loss": 3.594932950586437,
      "tokens_seen": 1173225472
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032551654964894683,
      "loss": 2.8276,
      "theoretical_loss": 3.5949144008031455,
      "tokens_seen": 1173291008
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032550651955867607,
      "loss": 2.8463,
      "theoretical_loss": 3.5948958523460495,
      "tokens_seen": 1173356544
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003254964894684052,
      "loss": 3.1208,
      "theoretical_loss": 3.59487730521498,
      "tokens_seen": 1173422080
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032548645937813443,
      "loss": 2.8357,
      "theoretical_loss": 3.5948587594097674,
      "tokens_seen": 1173487616
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003254764292878636,
      "loss": 2.9326,
      "theoretical_loss": 3.594840214930244,
      "tokens_seen": 1173553152
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003254663991975928,
      "loss": 2.6981,
      "theoretical_loss": 3.5948216717762405,
      "tokens_seen": 1173618688
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.000325456369107322,
      "loss": 2.7153,
      "theoretical_loss": 3.5948031299475884,
      "tokens_seen": 1173684224
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032544633901705115,
      "loss": 2.9323,
      "theoretical_loss": 3.5947845894441186,
      "tokens_seen": 1173749760
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032543630892678034,
      "loss": 2.7989,
      "theoretical_loss": 3.5947660502656626,
      "tokens_seen": 1173815296
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032542627883650957,
      "loss": 2.8779,
      "theoretical_loss": 3.5947475124120514,
      "tokens_seen": 1173880832
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003254162487462387,
      "loss": 3.0145,
      "theoretical_loss": 3.594728975883117,
      "tokens_seen": 1173946368
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032540621865596793,
      "loss": 2.7851,
      "theoretical_loss": 3.5947104406786905,
      "tokens_seen": 1174011904
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032539618856569706,
      "loss": 2.8397,
      "theoretical_loss": 3.594691906798603,
      "tokens_seen": 1174077440
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003253861584754263,
      "loss": 2.8494,
      "theoretical_loss": 3.5946733742426864,
      "tokens_seen": 1174142976
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003253761283851555,
      "loss": 2.5804,
      "theoretical_loss": 3.5946548430107716,
      "tokens_seen": 1174208512
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032536609829488466,
      "loss": 2.732,
      "theoretical_loss": 3.5946363131026917,
      "tokens_seen": 1174274048
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032535606820461384,
      "loss": 2.9611,
      "theoretical_loss": 3.594617784518276,
      "tokens_seen": 1174339584
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.000325346038114343,
      "loss": 2.8593,
      "theoretical_loss": 3.5945992572573577,
      "tokens_seen": 1174405120
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003253360080240722,
      "loss": 2.711,
      "theoretical_loss": 3.594580731319768,
      "tokens_seen": 1174470656
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032532597793380144,
      "loss": 2.9637,
      "theoretical_loss": 3.594562206705338,
      "tokens_seen": 1174536192
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032531594784353056,
      "loss": 2.8753,
      "theoretical_loss": 3.5945436834139004,
      "tokens_seen": 1174601728
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003253059177532598,
      "loss": 2.8191,
      "theoretical_loss": 3.594525161445286,
      "tokens_seen": 1174667264
    },
    {
      "epoch": 0.36,
      "objective/train/advantage_avg": 0.49575820565223694,
      "objective/train/docs_used": 668590,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.414501667022705,
      "objective/train/original_loss": 2.414501905441284,
      "objective/train/theoretical_loss": 3.594506640799327,
      "objective/train/tokens_used": 1195192800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24581876397132874,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0508253574371338,
      "objective/train/weighted_lm_loss": 2.5372376441955566,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0408108234405518,
      "theoretical_loss": 3.594506640799327,
      "tokens_seen": 1174732800
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.000325295887662989,
      "loss": 2.8104,
      "theoretical_loss": 3.594506640799327,
      "tokens_seen": 1174732800
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032528585757271816,
      "loss": 2.9968,
      "theoretical_loss": 3.5944881214758553,
      "tokens_seen": 1174798336
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032527582748244734,
      "loss": 2.6427,
      "theoretical_loss": 3.594469603474702,
      "tokens_seen": 1174863872
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003252657973921765,
      "loss": 2.9796,
      "theoretical_loss": 3.5944510867957,
      "tokens_seen": 1174929408
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003252557673019057,
      "loss": 3.0401,
      "theoretical_loss": 3.59443257143868,
      "tokens_seen": 1174994944
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032524573721163494,
      "loss": 3.0106,
      "theoretical_loss": 3.5944140574034744,
      "tokens_seen": 1175060480
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032523570712136407,
      "loss": 2.7207,
      "theoretical_loss": 3.594395544689915,
      "tokens_seen": 1175126016
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003252256770310933,
      "loss": 3.0378,
      "theoretical_loss": 3.5943770332978344,
      "tokens_seen": 1175191552
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032521564694082243,
      "loss": 3.1263,
      "theoretical_loss": 3.594358523227064,
      "tokens_seen": 1175257088
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032520561685055166,
      "loss": 2.9581,
      "theoretical_loss": 3.5943400144774356,
      "tokens_seen": 1175322624
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032519558676028084,
      "loss": 2.9763,
      "theoretical_loss": 3.5943215070487815,
      "tokens_seen": 1175388160
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032518555667001,
      "loss": 2.9715,
      "theoretical_loss": 3.5943030009409345,
      "tokens_seen": 1175453696
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003251755265797392,
      "loss": 2.7858,
      "theoretical_loss": 3.5942844961537257,
      "tokens_seen": 1175519232
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032516549648946844,
      "loss": 2.715,
      "theoretical_loss": 3.5942659926869873,
      "tokens_seen": 1175584768
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032515546639919757,
      "loss": 2.9315,
      "theoretical_loss": 3.5942474905405524,
      "tokens_seen": 1175650304
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003251454363089268,
      "loss": 2.9729,
      "theoretical_loss": 3.594228989714252,
      "tokens_seen": 1175715840
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032513540621865593,
      "loss": 2.8497,
      "theoretical_loss": 3.594210490207919,
      "tokens_seen": 1175781376
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032512537612838517,
      "loss": 3.0651,
      "theoretical_loss": 3.5941919920213854,
      "tokens_seen": 1175846912
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032511534603811435,
      "loss": 2.7655,
      "theoretical_loss": 3.5941734951544837,
      "tokens_seen": 1175912448
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032510531594784353,
      "loss": 2.8843,
      "theoretical_loss": 3.5941549996070465,
      "tokens_seen": 1175977984
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003250952858575727,
      "loss": 3.127,
      "theoretical_loss": 3.594136505378905,
      "tokens_seen": 1176043520
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003250852557673019,
      "loss": 2.8956,
      "theoretical_loss": 3.594118012469893,
      "tokens_seen": 1176109056
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032507522567703107,
      "loss": 3.1121,
      "theoretical_loss": 3.5940995208798423,
      "tokens_seen": 1176174592
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003250651955867603,
      "loss": 2.9684,
      "theoretical_loss": 3.594081030608585,
      "tokens_seen": 1176240128
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003250551654964895,
      "loss": 3.0358,
      "theoretical_loss": 3.5940625416559544,
      "tokens_seen": 1176305664
    },
    {
      "epoch": 0.36,
      "objective/train/advantage_avg": 0.4960206151008606,
      "objective/train/docs_used": 669752,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.9082226753234863,
      "objective/train/original_loss": 2.9082226753234863,
      "objective/train/theoretical_loss": 3.594044054021782,
      "objective/train/tokens_used": 1196831200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2460746318101883,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0508531332015991,
      "objective/train/weighted_lm_loss": 3.056011915206909,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0428944826126099,
      "theoretical_loss": 3.594044054021782,
      "tokens_seen": 1176371200
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032504513540621867,
      "loss": 3.0063,
      "theoretical_loss": 3.594044054021782,
      "tokens_seen": 1176371200
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032503510531594785,
      "loss": 2.9551,
      "theoretical_loss": 3.594025567705901,
      "tokens_seen": 1176436736
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032502507522567703,
      "loss": 2.8636,
      "theoretical_loss": 3.5940070827081443,
      "tokens_seen": 1176502272
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032501504513540627,
      "loss": 2.8222,
      "theoretical_loss": 3.5939885990283438,
      "tokens_seen": 1176567808
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003250050150451354,
      "loss": 2.8374,
      "theoretical_loss": 3.5939701166663323,
      "tokens_seen": 1176633344
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032499498495486463,
      "loss": 2.8361,
      "theoretical_loss": 3.5939516356219423,
      "tokens_seen": 1176698880
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003249849548645938,
      "loss": 2.7237,
      "theoretical_loss": 3.593933155895007,
      "tokens_seen": 1176764416
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.000324974924774323,
      "loss": 2.8954,
      "theoretical_loss": 3.5939146774853583,
      "tokens_seen": 1176829952
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003249648946840522,
      "loss": 2.7578,
      "theoretical_loss": 3.59389620039283,
      "tokens_seen": 1176895488
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032495486459378135,
      "loss": 2.8751,
      "theoretical_loss": 3.593877724617254,
      "tokens_seen": 1176961024
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032494483450351054,
      "loss": 2.9233,
      "theoretical_loss": 3.5938592501584634,
      "tokens_seen": 1177026560
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032493480441323977,
      "loss": 2.6285,
      "theoretical_loss": 3.5938407770162915,
      "tokens_seen": 1177092096
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003249247743229689,
      "loss": 3.0085,
      "theoretical_loss": 3.5938223051905704,
      "tokens_seen": 1177157632
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032491474423269813,
      "loss": 2.8171,
      "theoretical_loss": 3.5938038346811334,
      "tokens_seen": 1177223168
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032490471414242726,
      "loss": 2.9261,
      "theoretical_loss": 3.593785365487814,
      "tokens_seen": 1177288704
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003248946840521565,
      "loss": 3.0764,
      "theoretical_loss": 3.5937668976104438,
      "tokens_seen": 1177354240
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003248846539618857,
      "loss": 2.8647,
      "theoretical_loss": 3.5937484310488568,
      "tokens_seen": 1177419776
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032487462387161486,
      "loss": 2.9039,
      "theoretical_loss": 3.5937299658028863,
      "tokens_seen": 1177485312
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032486459378134404,
      "loss": 2.8821,
      "theoretical_loss": 3.593711501872364,
      "tokens_seen": 1177550848
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003248545636910732,
      "loss": 3.0758,
      "theoretical_loss": 3.5936930392571242,
      "tokens_seen": 1177616384
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003248445336008024,
      "loss": 2.7327,
      "theoretical_loss": 3.5936745779569996,
      "tokens_seen": 1177681920
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032483450351053164,
      "loss": 2.8406,
      "theoretical_loss": 3.5936561179718236,
      "tokens_seen": 1177747456
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032482447342026076,
      "loss": 2.9997,
      "theoretical_loss": 3.5936376593014288,
      "tokens_seen": 1177812992
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032481444332999,
      "loss": 2.8099,
      "theoretical_loss": 3.593619201945649,
      "tokens_seen": 1177878528
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003248044132397192,
      "loss": 2.8673,
      "theoretical_loss": 3.5936007459043173,
      "tokens_seen": 1177944064
    },
    {
      "epoch": 0.36,
      "objective/train/advantage_avg": 0.49515965580940247,
      "objective/train/docs_used": 670392,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.114396810531616,
      "objective/train/original_loss": 3.114396572113037,
      "objective/train/theoretical_loss": 3.593582291177267,
      "objective/train/tokens_used": 1198469600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2452389895915985,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.05076265335083,
      "objective/train/weighted_lm_loss": 3.272519111633301,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0417574644088745,
      "theoretical_loss": 3.593582291177267,
      "tokens_seen": 1178009600
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032479438314944836,
      "loss": 3.1547,
      "theoretical_loss": 3.593582291177267,
      "tokens_seen": 1178009600
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032478435305917754,
      "loss": 2.7225,
      "theoretical_loss": 3.5935638377643313,
      "tokens_seen": 1178075136
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003247743229689067,
      "loss": 2.898,
      "theoretical_loss": 3.593545385665343,
      "tokens_seen": 1178140672
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003247642928786359,
      "loss": 2.6989,
      "theoretical_loss": 3.593526934880136,
      "tokens_seen": 1178206208
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032475426278836514,
      "loss": 2.9021,
      "theoretical_loss": 3.5935084854085444,
      "tokens_seen": 1178271744
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032474423269809427,
      "loss": 2.787,
      "theoretical_loss": 3.5934900372504,
      "tokens_seen": 1178337280
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003247342026078235,
      "loss": 3.1855,
      "theoretical_loss": 3.593471590405538,
      "tokens_seen": 1178402816
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032472417251755263,
      "loss": 2.834,
      "theoretical_loss": 3.5934531448737905,
      "tokens_seen": 1178468352
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032471414242728186,
      "loss": 2.7697,
      "theoretical_loss": 3.593434700654991,
      "tokens_seen": 1178533888
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032470411233701104,
      "loss": 2.799,
      "theoretical_loss": 3.5934162577489746,
      "tokens_seen": 1178599424
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003246940822467402,
      "loss": 2.8619,
      "theoretical_loss": 3.5933978161555733,
      "tokens_seen": 1178664960
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003246840521564694,
      "loss": 2.7531,
      "theoretical_loss": 3.5933793758746213,
      "tokens_seen": 1178730496
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032467402206619864,
      "loss": 2.8648,
      "theoretical_loss": 3.5933609369059525,
      "tokens_seen": 1178796032
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032466399197592777,
      "loss": 2.9594,
      "theoretical_loss": 3.5933424992494,
      "tokens_seen": 1178861568
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.000324653961885657,
      "loss": 2.905,
      "theoretical_loss": 3.5933240629047973,
      "tokens_seen": 1178927104
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032464393179538613,
      "loss": 3.0878,
      "theoretical_loss": 3.5933056278719793,
      "tokens_seen": 1178992640
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032463390170511537,
      "loss": 2.847,
      "theoretical_loss": 3.5932871941507782,
      "tokens_seen": 1179058176
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032462387161484455,
      "loss": 2.7057,
      "theoretical_loss": 3.593268761741029,
      "tokens_seen": 1179123712
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032461384152457373,
      "loss": 2.8581,
      "theoretical_loss": 3.593250330642565,
      "tokens_seen": 1179189248
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003246038114343029,
      "loss": 2.8883,
      "theoretical_loss": 3.5932319008552205,
      "tokens_seen": 1179254784
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003245937813440321,
      "loss": 3.0249,
      "theoretical_loss": 3.593213472378829,
      "tokens_seen": 1179320320
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032458375125376127,
      "loss": 2.8007,
      "theoretical_loss": 3.593195045213224,
      "tokens_seen": 1179385856
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003245737211634905,
      "loss": 2.9593,
      "theoretical_loss": 3.59317661935824,
      "tokens_seen": 1179451392
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032456369107321963,
      "loss": 2.67,
      "theoretical_loss": 3.593158194813711,
      "tokens_seen": 1179516928
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032455366098294887,
      "loss": 3.005,
      "theoretical_loss": 3.593139771579471,
      "tokens_seen": 1179582464
    },
    {
      "epoch": 0.36,
      "objective/train/advantage_avg": 0.49207794666290283,
      "objective/train/docs_used": 671618,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.018869161605835,
      "objective/train/original_loss": 3.018869400024414,
      "objective/train/theoretical_loss": 3.5931213496553536,
      "objective/train/tokens_used": 1200108000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24228186905384064,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0504392385482788,
      "objective/train/weighted_lm_loss": 3.171093225479126,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 3.5931213496553536,
      "tokens_seen": 1179648000
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.000324543630892678,
      "loss": 2.9659,
      "theoretical_loss": 3.5931213496553536,
      "tokens_seen": 1179648000
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032453360080240723,
      "loss": 2.832,
      "theoretical_loss": 3.5931029290411933,
      "tokens_seen": 1179713536
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003245235707121364,
      "loss": 2.7953,
      "theoretical_loss": 3.5930845097368236,
      "tokens_seen": 1179779072
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003245135406218656,
      "loss": 2.6699,
      "theoretical_loss": 3.593066091742079,
      "tokens_seen": 1179844608
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003245035105315948,
      "loss": 2.8844,
      "theoretical_loss": 3.593047675056794,
      "tokens_seen": 1179910144
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.000324493480441324,
      "loss": 3.0492,
      "theoretical_loss": 3.5930292596808027,
      "tokens_seen": 1179975680
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032448345035105314,
      "loss": 2.9498,
      "theoretical_loss": 3.5930108456139385,
      "tokens_seen": 1180041216
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003244734202607824,
      "loss": 3.1039,
      "theoretical_loss": 3.592992432856037,
      "tokens_seen": 1180106752
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003244633901705115,
      "loss": 2.9722,
      "theoretical_loss": 3.5929740214069312,
      "tokens_seen": 1180172288
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032445336008024074,
      "loss": 2.8372,
      "theoretical_loss": 3.5929556112664556,
      "tokens_seen": 1180237824
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003244433299899699,
      "loss": 3.02,
      "theoretical_loss": 3.592937202434445,
      "tokens_seen": 1180303360
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003244332998996991,
      "loss": 2.7732,
      "theoretical_loss": 3.5929187949107333,
      "tokens_seen": 1180368896
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003244232698094283,
      "loss": 2.9735,
      "theoretical_loss": 3.5929003886951554,
      "tokens_seen": 1180434432
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032441323971915746,
      "loss": 2.8111,
      "theoretical_loss": 3.5928819837875454,
      "tokens_seen": 1180499968
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032440320962888664,
      "loss": 2.9192,
      "theoretical_loss": 3.5928635801877378,
      "tokens_seen": 1180565504
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003243931795386159,
      "loss": 2.9333,
      "theoretical_loss": 3.5928451778955672,
      "tokens_seen": 1180631040
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.000324383149448345,
      "loss": 2.8515,
      "theoretical_loss": 3.5928267769108677,
      "tokens_seen": 1180696576
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032437311935807424,
      "loss": 2.9192,
      "theoretical_loss": 3.5928083772334745,
      "tokens_seen": 1180762112
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032436308926780337,
      "loss": 2.9242,
      "theoretical_loss": 3.5927899788632214,
      "tokens_seen": 1180827648
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003243530591775326,
      "loss": 2.8693,
      "theoretical_loss": 3.592771581799944,
      "tokens_seen": 1180893184
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003243430290872618,
      "loss": 2.8162,
      "theoretical_loss": 3.592753186043476,
      "tokens_seen": 1180958720
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032433299899699096,
      "loss": 2.7437,
      "theoretical_loss": 3.592734791593652,
      "tokens_seen": 1181024256
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003243229689067202,
      "loss": 2.8231,
      "theoretical_loss": 3.5927163984503077,
      "tokens_seen": 1181089792
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003243129388164494,
      "loss": 2.9478,
      "theoretical_loss": 3.592698006613277,
      "tokens_seen": 1181155328
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032430290872617856,
      "loss": 3.0174,
      "theoretical_loss": 3.592679616082395,
      "tokens_seen": 1181220864
    },
    {
      "epoch": 0.36,
      "objective/train/advantage_avg": 0.4940136969089508,
      "objective/train/docs_used": 672229,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.8143558502197266,
      "objective/train/original_loss": 2.8143553733825684,
      "objective/train/theoretical_loss": 3.592661226857496,
      "objective/train/tokens_used": 1201746400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.244103342294693,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506422519683838,
      "objective/train/weighted_lm_loss": 2.9568512439727783,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0454468727111816,
      "theoretical_loss": 3.592661226857496,
      "tokens_seen": 1181286400
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032429287863590774,
      "loss": 2.8921,
      "theoretical_loss": 3.592661226857496,
      "tokens_seen": 1181286400
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003242828485456369,
      "loss": 2.6724,
      "theoretical_loss": 3.5926428389384153,
      "tokens_seen": 1181351936
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003242728184553661,
      "loss": 2.8728,
      "theoretical_loss": 3.5926244523249875,
      "tokens_seen": 1181417472
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032426278836509534,
      "loss": 2.8192,
      "theoretical_loss": 3.592606067017048,
      "tokens_seen": 1181483008
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032425275827482447,
      "loss": 2.8934,
      "theoretical_loss": 3.5925876830144308,
      "tokens_seen": 1181548544
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003242427281845537,
      "loss": 3.0534,
      "theoretical_loss": 3.5925693003169714,
      "tokens_seen": 1181614080
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032423269809428283,
      "loss": 2.8036,
      "theoretical_loss": 3.5925509189245046,
      "tokens_seen": 1181679616
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032422266800401206,
      "loss": 2.7481,
      "theoretical_loss": 3.5925325388368656,
      "tokens_seen": 1181745152
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032421263791374125,
      "loss": 2.885,
      "theoretical_loss": 3.5925141600538892,
      "tokens_seen": 1181810688
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003242026078234704,
      "loss": 2.8686,
      "theoretical_loss": 3.5924957825754102,
      "tokens_seen": 1181876224
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003241925777331996,
      "loss": 2.8037,
      "theoretical_loss": 3.592477406401265,
      "tokens_seen": 1181941760
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032418254764292884,
      "loss": 2.7917,
      "theoretical_loss": 3.5924590315312868,
      "tokens_seen": 1182007296
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032417251755265797,
      "loss": 2.9785,
      "theoretical_loss": 3.5924406579653114,
      "tokens_seen": 1182072832
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003241624874623872,
      "loss": 2.7805,
      "theoretical_loss": 3.592422285703175,
      "tokens_seen": 1182138368
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032415245737211633,
      "loss": 2.9041,
      "theoretical_loss": 3.592403914744712,
      "tokens_seen": 1182203904
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032414242728184557,
      "loss": 2.7011,
      "theoretical_loss": 3.592385545089757,
      "tokens_seen": 1182269440
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032413239719157475,
      "loss": 2.6634,
      "theoretical_loss": 3.592367176738146,
      "tokens_seen": 1182334976
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032412236710130393,
      "loss": 2.7571,
      "theoretical_loss": 3.592348809689715,
      "tokens_seen": 1182400512
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003241123370110331,
      "loss": 3.1472,
      "theoretical_loss": 3.592330443944298,
      "tokens_seen": 1182466048
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003241023069207623,
      "loss": 2.6195,
      "theoretical_loss": 3.5923120795017307,
      "tokens_seen": 1182531584
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032409227683049147,
      "loss": 2.9052,
      "theoretical_loss": 3.592293716361849,
      "tokens_seen": 1182597120
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003240822467402207,
      "loss": 3.0112,
      "theoretical_loss": 3.592275354524488,
      "tokens_seen": 1182662656
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032407221664994984,
      "loss": 2.9191,
      "theoretical_loss": 3.5922569939894826,
      "tokens_seen": 1182728192
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032406218655967907,
      "loss": 2.6356,
      "theoretical_loss": 3.5922386347566695,
      "tokens_seen": 1182793728
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003240521564694082,
      "loss": 2.9223,
      "theoretical_loss": 3.592220276825883,
      "tokens_seen": 1182859264
    },
    {
      "epoch": 0.36,
      "objective/train/advantage_avg": 0.49580124020576477,
      "objective/train/docs_used": 673021,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.9240059852600098,
      "objective/train/original_loss": 2.924006462097168,
      "objective/train/theoretical_loss": 3.592201920196959,
      "objective/train/tokens_used": 1203384800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24586646258831024,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0508300065994263,
      "objective/train/weighted_lm_loss": 3.072517156600952,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0408108234405518,
      "theoretical_loss": 3.592201920196959,
      "tokens_seen": 1182924800
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032404212637913743,
      "loss": 2.8912,
      "theoretical_loss": 3.592201920196959,
      "tokens_seen": 1182924800
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003240320962888666,
      "loss": 3.023,
      "theoretical_loss": 3.5921835648697336,
      "tokens_seen": 1182990336
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003240220661985958,
      "loss": 2.7273,
      "theoretical_loss": 3.5921652108440414,
      "tokens_seen": 1183055872
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.000324012036108325,
      "loss": 2.8395,
      "theoretical_loss": 3.592146858119719,
      "tokens_seen": 1183121408
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003240020060180542,
      "loss": 2.6968,
      "theoretical_loss": 3.5921285066966018,
      "tokens_seen": 1183186944
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032399197592778334,
      "loss": 3.0166,
      "theoretical_loss": 3.592110156574525,
      "tokens_seen": 1183252480
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003239819458375126,
      "loss": 2.9789,
      "theoretical_loss": 3.5920918077533246,
      "tokens_seen": 1183318016
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003239719157472417,
      "loss": 2.8923,
      "theoretical_loss": 3.5920734602328364,
      "tokens_seen": 1183383552
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032396188565697094,
      "loss": 2.8262,
      "theoretical_loss": 3.5920551140128962,
      "tokens_seen": 1183449088
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003239518555667001,
      "loss": 2.9397,
      "theoretical_loss": 3.5920367690933395,
      "tokens_seen": 1183514624
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003239418254764293,
      "loss": 2.7184,
      "theoretical_loss": 3.5920184254740026,
      "tokens_seen": 1183580160
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003239317953861585,
      "loss": 3.0342,
      "theoretical_loss": 3.592000083154721,
      "tokens_seen": 1183645696
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032392176529588766,
      "loss": 2.9876,
      "theoretical_loss": 3.5919817421353306,
      "tokens_seen": 1183711232
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032391173520561684,
      "loss": 3.0286,
      "theoretical_loss": 3.5919634024156677,
      "tokens_seen": 1183776768
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003239017051153461,
      "loss": 3.0371,
      "theoretical_loss": 3.591945063995568,
      "tokens_seen": 1183842304
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003238916750250752,
      "loss": 2.9177,
      "theoretical_loss": 3.591926726874867,
      "tokens_seen": 1183907840
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032388164493480444,
      "loss": 2.7562,
      "theoretical_loss": 3.5919083910534013,
      "tokens_seen": 1183973376
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032387161484453357,
      "loss": 2.8356,
      "theoretical_loss": 3.591890056531007,
      "tokens_seen": 1184038912
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003238615847542628,
      "loss": 2.5876,
      "theoretical_loss": 3.59187172330752,
      "tokens_seen": 1184104448
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.000323851554663992,
      "loss": 2.8068,
      "theoretical_loss": 3.5918533913827764,
      "tokens_seen": 1184169984
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032384152457372116,
      "loss": 2.8374,
      "theoretical_loss": 3.591835060756612,
      "tokens_seen": 1184235520
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032383149448345034,
      "loss": 3.0198,
      "theoretical_loss": 3.591816731428863,
      "tokens_seen": 1184301056
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003238214643931796,
      "loss": 2.6972,
      "theoretical_loss": 3.591798403399366,
      "tokens_seen": 1184366592
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003238114343029087,
      "loss": 2.627,
      "theoretical_loss": 3.5917800766679573,
      "tokens_seen": 1184432128
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032380140421263794,
      "loss": 2.8132,
      "theoretical_loss": 3.591761751234473,
      "tokens_seen": 1184497664
    },
    {
      "epoch": 0.36,
      "objective/train/advantage_avg": 0.49458229541778564,
      "objective/train/docs_used": 674157,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.0906736850738525,
      "objective/train/original_loss": 3.0906734466552734,
      "objective/train/theoretical_loss": 3.5917434270987485,
      "objective/train/tokens_used": 1205023200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24467535316944122,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507019758224487,
      "objective/train/weighted_lm_loss": 3.2473340034484863,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.042169213294983,
      "theoretical_loss": 3.5917434270987485,
      "tokens_seen": 1184563200
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032379137412236707,
      "loss": 2.8947,
      "theoretical_loss": 3.5917434270987485,
      "tokens_seen": 1184563200
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003237813440320963,
      "loss": 2.7723,
      "theoretical_loss": 3.5917251042606217,
      "tokens_seen": 1184628736
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003237713139418255,
      "loss": 3.0311,
      "theoretical_loss": 3.5917067827199274,
      "tokens_seen": 1184694272
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032376128385155467,
      "loss": 2.9365,
      "theoretical_loss": 3.5916884624765033,
      "tokens_seen": 1184759808
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032375125376128385,
      "loss": 2.808,
      "theoretical_loss": 3.5916701435301848,
      "tokens_seen": 1184825344
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032374122367101303,
      "loss": 2.919,
      "theoretical_loss": 3.591651825880809,
      "tokens_seen": 1184890880
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003237311935807422,
      "loss": 3.1995,
      "theoretical_loss": 3.5916335095282115,
      "tokens_seen": 1184956416
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032372116349047145,
      "loss": 2.9158,
      "theoretical_loss": 3.5916151944722294,
      "tokens_seen": 1185021952
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032371113340020057,
      "loss": 2.8454,
      "theoretical_loss": 3.5915968807126997,
      "tokens_seen": 1185087488
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003237011033099298,
      "loss": 2.9809,
      "theoretical_loss": 3.591578568249458,
      "tokens_seen": 1185153024
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032369107321965893,
      "loss": 2.7049,
      "theoretical_loss": 3.5915602570823415,
      "tokens_seen": 1185218560
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032368104312938817,
      "loss": 2.9336,
      "theoretical_loss": 3.5915419472111862,
      "tokens_seen": 1185284096
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032367101303911735,
      "loss": 2.8331,
      "theoretical_loss": 3.591523638635829,
      "tokens_seen": 1185349632
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032366098294884653,
      "loss": 2.662,
      "theoretical_loss": 3.5915053313561067,
      "tokens_seen": 1185415168
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003236509528585757,
      "loss": 2.8663,
      "theoretical_loss": 3.591487025371856,
      "tokens_seen": 1185480704
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032364092276830495,
      "loss": 3.0999,
      "theoretical_loss": 3.5914687206829137,
      "tokens_seen": 1185546240
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003236308926780341,
      "loss": 3.0526,
      "theoretical_loss": 3.591450417289116,
      "tokens_seen": 1185611776
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003236208625877633,
      "loss": 2.9525,
      "theoretical_loss": 3.5914321151903006,
      "tokens_seen": 1185677312
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032361083249749244,
      "loss": 2.9786,
      "theoretical_loss": 3.5914138143863035,
      "tokens_seen": 1185742848
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003236008024072217,
      "loss": 2.8271,
      "theoretical_loss": 3.591395514876962,
      "tokens_seen": 1185808384
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032359077231695085,
      "loss": 2.8665,
      "theoretical_loss": 3.5913772166621123,
      "tokens_seen": 1185873920
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032358074222668004,
      "loss": 3.1085,
      "theoretical_loss": 3.591358919741592,
      "tokens_seen": 1185939456
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032357071213640927,
      "loss": 2.5701,
      "theoretical_loss": 3.5913406241152384,
      "tokens_seen": 1186004992
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003235606820461384,
      "loss": 2.737,
      "theoretical_loss": 3.591322329782887,
      "tokens_seen": 1186070528
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032355065195586763,
      "loss": 2.94,
      "theoretical_loss": 3.5913040367443765,
      "tokens_seen": 1186136064
    },
    {
      "epoch": 0.36,
      "objective/train/advantage_avg": 0.4935271739959717,
      "objective/train/docs_used": 674881,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.7687864303588867,
      "objective/train/original_loss": 2.768787145614624,
      "objective/train/theoretical_loss": 3.591285744999542,
      "objective/train/tokens_used": 1206661600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24365460872650146,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050591230392456,
      "objective/train/weighted_lm_loss": 2.9090044498443604,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0338950157165527,
      "theoretical_loss": 3.591285744999542,
      "tokens_seen": 1186201600
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003235406218655968,
      "loss": 2.6937,
      "theoretical_loss": 3.591285744999542,
      "tokens_seen": 1186201600
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.000323530591775326,
      "loss": 2.7283,
      "theoretical_loss": 3.5912674545482224,
      "tokens_seen": 1186267136
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003235205616850552,
      "loss": 2.7142,
      "theoretical_loss": 3.591249165390254,
      "tokens_seen": 1186332672
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003235105315947844,
      "loss": 2.8564,
      "theoretical_loss": 3.591230877525474,
      "tokens_seen": 1186398208
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032350050150451354,
      "loss": 2.6947,
      "theoretical_loss": 3.591212590953719,
      "tokens_seen": 1186463744
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003234904714142428,
      "loss": 2.6455,
      "theoretical_loss": 3.5911943056748266,
      "tokens_seen": 1186529280
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003234804413239719,
      "loss": 2.8973,
      "theoretical_loss": 3.5911760216886344,
      "tokens_seen": 1186594816
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032347041123370114,
      "loss": 2.7296,
      "theoretical_loss": 3.591157738994979,
      "tokens_seen": 1186660352
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003234603811434303,
      "loss": 2.9495,
      "theoretical_loss": 3.5911394575936977,
      "tokens_seen": 1186725888
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003234503510531595,
      "loss": 2.8086,
      "theoretical_loss": 3.591121177484628,
      "tokens_seen": 1186791424
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003234403209628887,
      "loss": 2.9428,
      "theoretical_loss": 3.5911028986676072,
      "tokens_seen": 1186856960
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032343029087261786,
      "loss": 2.7319,
      "theoretical_loss": 3.591084621142473,
      "tokens_seen": 1186922496
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032342026078234704,
      "loss": 2.7711,
      "theoretical_loss": 3.591066344909062,
      "tokens_seen": 1186988032
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003234102306920763,
      "loss": 2.8984,
      "theoretical_loss": 3.5910480699672114,
      "tokens_seen": 1187053568
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003234002006018054,
      "loss": 2.7906,
      "theoretical_loss": 3.59102979631676,
      "tokens_seen": 1187119104
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032339017051153464,
      "loss": 2.7501,
      "theoretical_loss": 3.591011523957544,
      "tokens_seen": 1187184640
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032338014042126377,
      "loss": 2.7593,
      "theoretical_loss": 3.5909932528894015,
      "tokens_seen": 1187250176
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.000323370110330993,
      "loss": 2.897,
      "theoretical_loss": 3.59097498311217,
      "tokens_seen": 1187315712
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003233600802407222,
      "loss": 2.795,
      "theoretical_loss": 3.590956714625687,
      "tokens_seen": 1187381248
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032335005015045136,
      "loss": 2.9367,
      "theoretical_loss": 3.5909384474297896,
      "tokens_seen": 1187446784
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032334002006018054,
      "loss": 3.1467,
      "theoretical_loss": 3.590920181524316,
      "tokens_seen": 1187512320
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003233299899699098,
      "loss": 2.8754,
      "theoretical_loss": 3.5909019169091034,
      "tokens_seen": 1187577856
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003233199598796389,
      "loss": 2.9394,
      "theoretical_loss": 3.59088365358399,
      "tokens_seen": 1187643392
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032330992978936814,
      "loss": 2.7288,
      "theoretical_loss": 3.590865391548813,
      "tokens_seen": 1187708928
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032329989969909727,
      "loss": 2.887,
      "theoretical_loss": 3.59084713080341,
      "tokens_seen": 1187774464
    },
    {
      "epoch": 0.36,
      "objective/train/advantage_avg": 0.4946969151496887,
      "objective/train/docs_used": 676005,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.794457197189331,
      "objective/train/original_loss": 2.794457197189331,
      "objective/train/theoretical_loss": 3.5908288713476195,
      "objective/train/tokens_used": 1208300000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24478185176849365,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507138967514038,
      "objective/train/weighted_lm_loss": 2.9360580444335938,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 3.5908288713476195,
      "tokens_seen": 1187840000
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003232898696088265,
      "loss": 3.0231,
      "theoretical_loss": 3.5908288713476195,
      "tokens_seen": 1187840000
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003232798395185557,
      "loss": 2.8406,
      "theoretical_loss": 3.5908106131812785,
      "tokens_seen": 1187905536
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032326980942828487,
      "loss": 3.013,
      "theoretical_loss": 3.5907923563042257,
      "tokens_seen": 1187971072
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032325977933801405,
      "loss": 2.9738,
      "theoretical_loss": 3.590774100716298,
      "tokens_seen": 1188036608
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032324974924774323,
      "loss": 2.601,
      "theoretical_loss": 3.5907558464173337,
      "tokens_seen": 1188102144
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003232397191574724,
      "loss": 2.9197,
      "theoretical_loss": 3.5907375934071704,
      "tokens_seen": 1188167680
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032322968906720165,
      "loss": 2.9967,
      "theoretical_loss": 3.590719341685647,
      "tokens_seen": 1188233216
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032321965897693077,
      "loss": 2.8295,
      "theoretical_loss": 3.5907010912526,
      "tokens_seen": 1188298752
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032320962888666,
      "loss": 2.9279,
      "theoretical_loss": 3.590682842107868,
      "tokens_seen": 1188364288
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032319959879638913,
      "loss": 2.8647,
      "theoretical_loss": 3.5906645942512894,
      "tokens_seen": 1188429824
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032318956870611837,
      "loss": 2.7311,
      "theoretical_loss": 3.5906463476827026,
      "tokens_seen": 1188495360
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032317953861584755,
      "loss": 3.0905,
      "theoretical_loss": 3.5906281024019444,
      "tokens_seen": 1188560896
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032316950852557673,
      "loss": 2.8197,
      "theoretical_loss": 3.590609858408854,
      "tokens_seen": 1188626432
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003231594784353059,
      "loss": 2.9471,
      "theoretical_loss": 3.5905916157032682,
      "tokens_seen": 1188691968
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032314944834503515,
      "loss": 2.8606,
      "theoretical_loss": 3.590573374285027,
      "tokens_seen": 1188757504
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003231394182547643,
      "loss": 2.7276,
      "theoretical_loss": 3.590555134153967,
      "tokens_seen": 1188823040
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003231293881644935,
      "loss": 2.717,
      "theoretical_loss": 3.5905368953099277,
      "tokens_seen": 1188888576
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032311935807422264,
      "loss": 2.7629,
      "theoretical_loss": 3.590518657752746,
      "tokens_seen": 1188954112
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003231093279839519,
      "loss": 2.5946,
      "theoretical_loss": 3.590500421482261,
      "tokens_seen": 1189019648
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032309929789368105,
      "loss": 2.8061,
      "theoretical_loss": 3.5904821864983116,
      "tokens_seen": 1189085184
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032308926780341024,
      "loss": 2.8536,
      "theoretical_loss": 3.590463952800735,
      "tokens_seen": 1189150720
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003230792377131394,
      "loss": 2.7329,
      "theoretical_loss": 3.5904457203893703,
      "tokens_seen": 1189216256
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003230692076228686,
      "loss": 3.1824,
      "theoretical_loss": 3.590427489264055,
      "tokens_seen": 1189281792
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003230591775325978,
      "loss": 2.6863,
      "theoretical_loss": 3.5904092594246286,
      "tokens_seen": 1189347328
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.000323049147442327,
      "loss": 2.9824,
      "theoretical_loss": 3.5903910308709293,
      "tokens_seen": 1189412864
    },
    {
      "epoch": 0.36,
      "objective/train/advantage_avg": 0.49257975816726685,
      "objective/train/docs_used": 676669,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.8434104919433594,
      "objective/train/original_loss": 2.8434107303619385,
      "objective/train/theoretical_loss": 3.590372803602795,
      "objective/train/tokens_used": 1209938400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24274925887584686,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0504919290542603,
      "objective/train/weighted_lm_loss": 2.986964702606201,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0338950157165527,
      "theoretical_loss": 3.590372803602795,
      "tokens_seen": 1189478400
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032303911735205614,
      "loss": 2.8803,
      "theoretical_loss": 3.590372803602795,
      "tokens_seen": 1189478400
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003230290872617854,
      "loss": 2.9275,
      "theoretical_loss": 3.5903545776200643,
      "tokens_seen": 1189543936
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003230190571715145,
      "loss": 2.9132,
      "theoretical_loss": 3.5903363529225762,
      "tokens_seen": 1189609472
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032300902708124374,
      "loss": 2.7332,
      "theoretical_loss": 3.5903181295101696,
      "tokens_seen": 1189675008
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003229989969909729,
      "loss": 2.843,
      "theoretical_loss": 3.5902999073826827,
      "tokens_seen": 1189740544
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003229889669007021,
      "loss": 2.9953,
      "theoretical_loss": 3.590281686539953,
      "tokens_seen": 1189806080
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003229789368104313,
      "loss": 3.183,
      "theoretical_loss": 3.5902634669818214,
      "tokens_seen": 1189871616
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003229689067201605,
      "loss": 2.6819,
      "theoretical_loss": 3.5902452487081247,
      "tokens_seen": 1189937152
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032295887662988964,
      "loss": 2.6797,
      "theoretical_loss": 3.590227031718702,
      "tokens_seen": 1190002688
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003229488465396189,
      "loss": 2.7041,
      "theoretical_loss": 3.590208816013393,
      "tokens_seen": 1190068224
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.000322938816449348,
      "loss": 2.9415,
      "theoretical_loss": 3.5901906015920355,
      "tokens_seen": 1190133760
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032292878635907724,
      "loss": 2.7858,
      "theoretical_loss": 3.5901723884544685,
      "tokens_seen": 1190199296
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003229187562688064,
      "loss": 2.7954,
      "theoretical_loss": 3.5901541766005316,
      "tokens_seen": 1190264832
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003229087261785356,
      "loss": 2.9495,
      "theoretical_loss": 3.5901359660300622,
      "tokens_seen": 1190330368
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003228986960882648,
      "loss": 2.6711,
      "theoretical_loss": 3.5901177567429006,
      "tokens_seen": 1190395904
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032288866599799397,
      "loss": 2.6509,
      "theoretical_loss": 3.5900995487388845,
      "tokens_seen": 1190461440
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032287863590772315,
      "loss": 2.8005,
      "theoretical_loss": 3.590081342017854,
      "tokens_seen": 1190526976
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003228686058174524,
      "loss": 2.9485,
      "theoretical_loss": 3.5900631365796474,
      "tokens_seen": 1190592512
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003228585757271815,
      "loss": 2.6249,
      "theoretical_loss": 3.590044932424104,
      "tokens_seen": 1190658048
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032284854563691074,
      "loss": 3.1037,
      "theoretical_loss": 3.5900267295510626,
      "tokens_seen": 1190723584
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003228385155466399,
      "loss": 2.7543,
      "theoretical_loss": 3.5900085279603626,
      "tokens_seen": 1190789120
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003228284854563691,
      "loss": 2.894,
      "theoretical_loss": 3.5899903276518423,
      "tokens_seen": 1190854656
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032281845536609834,
      "loss": 2.9113,
      "theoretical_loss": 3.589972128625342,
      "tokens_seen": 1190920192
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032280842527582747,
      "loss": 2.9225,
      "theoretical_loss": 3.5899539308807,
      "tokens_seen": 1190985728
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003227983951855567,
      "loss": 2.9478,
      "theoretical_loss": 3.5899357344177556,
      "tokens_seen": 1191051264
    },
    {
      "epoch": 0.36,
      "objective/train/advantage_avg": 0.4943808317184448,
      "objective/train/docs_used": 677774,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.9239799976348877,
      "objective/train/original_loss": 2.923980474472046,
      "objective/train/theoretical_loss": 3.5899175392363483,
      "objective/train/tokens_used": 1211576800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24447819590568542,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506808757781982,
      "objective/train/weighted_lm_loss": 3.072112798690796,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 3.5899175392363483,
      "tokens_seen": 1191116800
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003227883650952859,
      "loss": 2.9022,
      "theoretical_loss": 3.5899175392363483,
      "tokens_seen": 1191116800
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032277833500501507,
      "loss": 2.6572,
      "theoretical_loss": 3.5898993453363173,
      "tokens_seen": 1191182336
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032276830491474425,
      "loss": 2.9185,
      "theoretical_loss": 3.5898811527175014,
      "tokens_seen": 1191247872
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032275827482447343,
      "loss": 3.0508,
      "theoretical_loss": 3.5898629613797404,
      "tokens_seen": 1191313408
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003227482447342026,
      "loss": 2.7461,
      "theoretical_loss": 3.589844771322874,
      "tokens_seen": 1191378944
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032273821464393185,
      "loss": 2.7624,
      "theoretical_loss": 3.58982658254674,
      "tokens_seen": 1191444480
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032272818455366097,
      "loss": 2.7839,
      "theoretical_loss": 3.58980839505118,
      "tokens_seen": 1191510016
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003227181544633902,
      "loss": 2.6166,
      "theoretical_loss": 3.5897902088360314,
      "tokens_seen": 1191575552
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032270812437311933,
      "loss": 2.818,
      "theoretical_loss": 3.589772023901135,
      "tokens_seen": 1191641088
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032269809428284857,
      "loss": 3.0844,
      "theoretical_loss": 3.5897538402463294,
      "tokens_seen": 1191706624
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032268806419257775,
      "loss": 2.899,
      "theoretical_loss": 3.5897356578714543,
      "tokens_seen": 1191772160
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032267803410230693,
      "loss": 2.7363,
      "theoretical_loss": 3.5897174767763493,
      "tokens_seen": 1191837696
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003226680040120361,
      "loss": 2.9087,
      "theoretical_loss": 3.5896992969608545,
      "tokens_seen": 1191903232
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032265797392176535,
      "loss": 2.7159,
      "theoretical_loss": 3.589681118424809,
      "tokens_seen": 1191968768
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003226479438314945,
      "loss": 2.9022,
      "theoretical_loss": 3.5896629411680525,
      "tokens_seen": 1192034304
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003226379137412237,
      "loss": 2.8913,
      "theoretical_loss": 3.589644765190424,
      "tokens_seen": 1192099840
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032262788365095284,
      "loss": 2.9361,
      "theoretical_loss": 3.5896265904917644,
      "tokens_seen": 1192165376
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003226178535606821,
      "loss": 2.7845,
      "theoretical_loss": 3.5896084170719127,
      "tokens_seen": 1192230912
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032260782347041125,
      "loss": 2.882,
      "theoretical_loss": 3.589590244930709,
      "tokens_seen": 1192296448
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032259779338014044,
      "loss": 2.6455,
      "theoretical_loss": 3.589572074067992,
      "tokens_seen": 1192361984
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003225877632898696,
      "loss": 3.0553,
      "theoretical_loss": 3.5895539044836027,
      "tokens_seen": 1192427520
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003225777331995988,
      "loss": 2.8418,
      "theoretical_loss": 3.58953573617738,
      "tokens_seen": 1192493056
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.000322567703109328,
      "loss": 2.6196,
      "theoretical_loss": 3.589517569149165,
      "tokens_seen": 1192558592
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003225576730190572,
      "loss": 3.0974,
      "theoretical_loss": 3.589499403398796,
      "tokens_seen": 1192624128
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032254764292878634,
      "loss": 2.8725,
      "theoretical_loss": 3.5894812389261146,
      "tokens_seen": 1192689664
    },
    {
      "epoch": 0.36,
      "objective/train/advantage_avg": 0.49461430311203003,
      "objective/train/docs_used": 678420,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.7961504459381104,
      "objective/train/original_loss": 2.7961506843566895,
      "objective/train/theoretical_loss": 3.589463075730959,
      "objective/train/tokens_used": 1213215200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24469050765037537,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507053136825562,
      "objective/train/weighted_lm_loss": 2.9379916191101074,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0442860126495361,
      "theoretical_loss": 3.589463075730959,
      "tokens_seen": 1192755200
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003225376128385156,
      "loss": 2.8336,
      "theoretical_loss": 3.589463075730959,
      "tokens_seen": 1192755200
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003225275827482447,
      "loss": 2.8032,
      "theoretical_loss": 3.5894449138131703,
      "tokens_seen": 1192820736
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032251755265797394,
      "loss": 3.0116,
      "theoretical_loss": 3.5894267531725887,
      "tokens_seen": 1192886272
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003225075225677031,
      "loss": 3.1578,
      "theoretical_loss": 3.589408593809053,
      "tokens_seen": 1192951808
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003224974924774323,
      "loss": 3.0503,
      "theoretical_loss": 3.5893904357224047,
      "tokens_seen": 1193017344
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003224874623871615,
      "loss": 2.9248,
      "theoretical_loss": 3.5893722789124825,
      "tokens_seen": 1193082880
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003224774322968907,
      "loss": 3.0714,
      "theoretical_loss": 3.5893541233791275,
      "tokens_seen": 1193148416
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032246740220661984,
      "loss": 2.8277,
      "theoretical_loss": 3.589335969122179,
      "tokens_seen": 1193213952
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003224573721163491,
      "loss": 2.8413,
      "theoretical_loss": 3.5893178161414783,
      "tokens_seen": 1193279488
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003224473420260782,
      "loss": 2.8892,
      "theoretical_loss": 3.5892996644368647,
      "tokens_seen": 1193345024
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032243731193580744,
      "loss": 2.8821,
      "theoretical_loss": 3.589281514008179,
      "tokens_seen": 1193410560
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003224272818455366,
      "loss": 2.7829,
      "theoretical_loss": 3.5892633648552605,
      "tokens_seen": 1193476096
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003224172517552658,
      "loss": 2.6571,
      "theoretical_loss": 3.5892452169779503,
      "tokens_seen": 1193541632
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.000322407221664995,
      "loss": 2.8459,
      "theoretical_loss": 3.589227070376089,
      "tokens_seen": 1193607168
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032239719157472417,
      "loss": 2.7455,
      "theoretical_loss": 3.5892089250495163,
      "tokens_seen": 1193672704
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032238716148445335,
      "loss": 2.6309,
      "theoretical_loss": 3.5891907809980723,
      "tokens_seen": 1193738240
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003223771313941826,
      "loss": 2.4785,
      "theoretical_loss": 3.589172638221598,
      "tokens_seen": 1193803776
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003223671013039117,
      "loss": 2.8832,
      "theoretical_loss": 3.589154496719934,
      "tokens_seen": 1193869312
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032235707121364095,
      "loss": 2.7813,
      "theoretical_loss": 3.58913635649292,
      "tokens_seen": 1193934848
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003223470411233701,
      "loss": 2.8885,
      "theoretical_loss": 3.5891182175403973,
      "tokens_seen": 1194000384
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003223370110330993,
      "loss": 2.8735,
      "theoretical_loss": 3.5891000798622055,
      "tokens_seen": 1194065920
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003223269809428285,
      "loss": 2.797,
      "theoretical_loss": 3.5890819434581855,
      "tokens_seen": 1194131456
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032231695085255767,
      "loss": 2.8135,
      "theoretical_loss": 3.589063808328179,
      "tokens_seen": 1194196992
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032230692076228685,
      "loss": 3.0291,
      "theoretical_loss": 3.5890456744720245,
      "tokens_seen": 1194262528
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003222968906720161,
      "loss": 2.8106,
      "theoretical_loss": 3.589027541889564,
      "tokens_seen": 1194328064
    },
    {
      "epoch": 0.36,
      "objective/train/advantage_avg": 0.49322161078453064,
      "objective/train/docs_used": 679445,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.741780996322632,
      "objective/train/original_loss": 2.741780996322632,
      "objective/train/theoretical_loss": 3.5890094105806383,
      "objective/train/tokens_used": 1214853600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2434299737215042,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505595207214355,
      "objective/train/weighted_lm_loss": 2.8804235458374023,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0387312173843384,
      "theoretical_loss": 3.5890094105806383,
      "tokens_seen": 1194393600
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003222868605817452,
      "loss": 2.7164,
      "theoretical_loss": 3.5890094105806383,
      "tokens_seen": 1194393600
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032227683049147445,
      "loss": 2.726,
      "theoretical_loss": 3.5889912805450876,
      "tokens_seen": 1194459136
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003222668004012036,
      "loss": 2.884,
      "theoretical_loss": 3.588973151782752,
      "tokens_seen": 1194524672
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003222567703109328,
      "loss": 2.9713,
      "theoretical_loss": 3.5889550242934734,
      "tokens_seen": 1194590208
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.000322246740220662,
      "loss": 2.7897,
      "theoretical_loss": 3.588936898077092,
      "tokens_seen": 1194655744
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032223671013039117,
      "loss": 2.6956,
      "theoretical_loss": 3.5889187731334493,
      "tokens_seen": 1194721280
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032222668004012035,
      "loss": 2.9113,
      "theoretical_loss": 3.588900649462385,
      "tokens_seen": 1194786816
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032221664994984953,
      "loss": 2.7449,
      "theoretical_loss": 3.58888252706374,
      "tokens_seen": 1194852352
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003222066198595787,
      "loss": 2.6664,
      "theoretical_loss": 3.588864405937356,
      "tokens_seen": 1194917888
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032219658976930795,
      "loss": 3.0441,
      "theoretical_loss": 3.588846286083074,
      "tokens_seen": 1194983424
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003221865596790371,
      "loss": 2.9909,
      "theoretical_loss": 3.5888281675007336,
      "tokens_seen": 1195048960
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003221765295887663,
      "loss": 2.9767,
      "theoretical_loss": 3.5888100501901774,
      "tokens_seen": 1195114496
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003221664994984955,
      "loss": 2.839,
      "theoretical_loss": 3.5887919341512453,
      "tokens_seen": 1195180032
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003221564694082247,
      "loss": 2.8192,
      "theoretical_loss": 3.5887738193837793,
      "tokens_seen": 1195245568
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032214643931795386,
      "loss": 2.792,
      "theoretical_loss": 3.588755705887619,
      "tokens_seen": 1195311104
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032213640922768304,
      "loss": 2.7711,
      "theoretical_loss": 3.5887375936626067,
      "tokens_seen": 1195376640
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003221263791374122,
      "loss": 2.6231,
      "theoretical_loss": 3.588719482708583,
      "tokens_seen": 1195442176
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032211634904714145,
      "loss": 2.7752,
      "theoretical_loss": 3.5887013730253896,
      "tokens_seen": 1195507712
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003221063189568706,
      "loss": 3.1171,
      "theoretical_loss": 3.588683264612867,
      "tokens_seen": 1195573248
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003220962888665998,
      "loss": 2.7756,
      "theoretical_loss": 3.5886651574708566,
      "tokens_seen": 1195638784
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032208625877632894,
      "loss": 3.0133,
      "theoretical_loss": 3.5886470515991995,
      "tokens_seen": 1195704320
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003220762286860582,
      "loss": 2.7257,
      "theoretical_loss": 3.588628946997737,
      "tokens_seen": 1195769856
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003220661985957874,
      "loss": 2.8388,
      "theoretical_loss": 3.588610843666311,
      "tokens_seen": 1195835392
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032205616850551654,
      "loss": 2.7967,
      "theoretical_loss": 3.588592741604762,
      "tokens_seen": 1195900928
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003220461384152458,
      "loss": 2.9647,
      "theoretical_loss": 3.588574640812932,
      "tokens_seen": 1195966464
    },
    {
      "epoch": 0.36,
      "objective/train/advantage_avg": 0.49371182918548584,
      "objective/train/docs_used": 680084,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.2709717750549316,
      "objective/train/original_loss": 3.2709717750549316,
      "objective/train/theoretical_loss": 3.5885565412906617,
      "objective/train/tokens_used": 1216492000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2438286989927292,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506106615066528,
      "objective/train/weighted_lm_loss": 3.4364800453186035,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0442860126495361,
      "theoretical_loss": 3.5885565412906617,
      "tokens_seen": 1196032000
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003220361083249749,
      "loss": 3.0229,
      "theoretical_loss": 3.5885565412906617,
      "tokens_seen": 1196032000
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032202607823470414,
      "loss": 2.7002,
      "theoretical_loss": 3.588538443037793,
      "tokens_seen": 1196097536
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003220160481444333,
      "loss": 3.0966,
      "theoretical_loss": 3.588520346054167,
      "tokens_seen": 1196163072
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003220060180541625,
      "loss": 2.9254,
      "theoretical_loss": 3.5885022503396256,
      "tokens_seen": 1196228608
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003219959879638917,
      "loss": 2.8122,
      "theoretical_loss": 3.58848415589401,
      "tokens_seen": 1196294144
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003219859578736209,
      "loss": 2.6793,
      "theoretical_loss": 3.588466062717161,
      "tokens_seen": 1196359680
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032197592778335004,
      "loss": 2.8455,
      "theoretical_loss": 3.5884479708089216,
      "tokens_seen": 1196425216
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003219658976930793,
      "loss": 2.8135,
      "theoretical_loss": 3.5884298801691323,
      "tokens_seen": 1196490752
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003219558676028084,
      "loss": 2.7936,
      "theoretical_loss": 3.588411790797635,
      "tokens_seen": 1196556288
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032194583751253764,
      "loss": 2.899,
      "theoretical_loss": 3.5883937026942716,
      "tokens_seen": 1196621824
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003219358074222668,
      "loss": 2.9837,
      "theoretical_loss": 3.588375615858883,
      "tokens_seen": 1196687360
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.000321925777331996,
      "loss": 2.9344,
      "theoretical_loss": 3.588357530291312,
      "tokens_seen": 1196752896
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003219157472417252,
      "loss": 2.5053,
      "theoretical_loss": 3.5883394459913993,
      "tokens_seen": 1196818432
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032190571715145437,
      "loss": 2.766,
      "theoretical_loss": 3.588321362958987,
      "tokens_seen": 1196883968
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032189568706118355,
      "loss": 2.9095,
      "theoretical_loss": 3.5883032811939173,
      "tokens_seen": 1196949504
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003218856569709128,
      "loss": 2.6891,
      "theoretical_loss": 3.5882852006960313,
      "tokens_seen": 1197015040
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003218756268806419,
      "loss": 2.7581,
      "theoretical_loss": 3.588267121465171,
      "tokens_seen": 1197080576
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032186559679037115,
      "loss": 2.8232,
      "theoretical_loss": 3.588249043501178,
      "tokens_seen": 1197146112
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003218555667001003,
      "loss": 2.954,
      "theoretical_loss": 3.588230966803895,
      "tokens_seen": 1197211648
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003218455366098295,
      "loss": 2.8557,
      "theoretical_loss": 3.5882128913731637,
      "tokens_seen": 1197277184
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003218355065195587,
      "loss": 3.1513,
      "theoretical_loss": 3.5881948172088256,
      "tokens_seen": 1197342720
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032182547642928787,
      "loss": 2.6265,
      "theoretical_loss": 3.5881767443107226,
      "tokens_seen": 1197408256
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032181544633901705,
      "loss": 2.8073,
      "theoretical_loss": 3.5881586726786976,
      "tokens_seen": 1197473792
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003218054162487463,
      "loss": 2.7753,
      "theoretical_loss": 3.588140602312591,
      "tokens_seen": 1197539328
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003217953861584754,
      "loss": 2.9917,
      "theoretical_loss": 3.5881225332122466,
      "tokens_seen": 1197604864
    },
    {
      "epoch": 0.36,
      "objective/train/advantage_avg": 0.493073433637619,
      "objective/train/docs_used": 681502,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.356417655944824,
      "objective/train/original_loss": 2.356417655944824,
      "objective/train/theoretical_loss": 3.5881044653775054,
      "objective/train/tokens_used": 1218130400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2432330697774887,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505437850952148,
      "objective/train/weighted_lm_loss": 2.4756102561950684,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0437887907028198,
      "theoretical_loss": 3.5881044653775054,
      "tokens_seen": 1197670400
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032178535606820465,
      "loss": 2.8071,
      "theoretical_loss": 3.5881044653775054,
      "tokens_seen": 1197670400
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003217753259779338,
      "loss": 2.6779,
      "theoretical_loss": 3.58808639880821,
      "tokens_seen": 1197735936
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.000321765295887663,
      "loss": 2.8336,
      "theoretical_loss": 3.588068333504202,
      "tokens_seen": 1197801472
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003217552657973922,
      "loss": 2.8199,
      "theoretical_loss": 3.588050269465324,
      "tokens_seen": 1197867008
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032174523570712137,
      "loss": 2.8797,
      "theoretical_loss": 3.5880322066914183,
      "tokens_seen": 1197932544
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032173520561685055,
      "loss": 2.6987,
      "theoretical_loss": 3.588014145182327,
      "tokens_seen": 1197998080
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032172517552657974,
      "loss": 2.8313,
      "theoretical_loss": 3.587996084937892,
      "tokens_seen": 1198063616
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003217151454363089,
      "loss": 2.67,
      "theoretical_loss": 3.587978025957956,
      "tokens_seen": 1198129152
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032170511534603815,
      "loss": 2.7176,
      "theoretical_loss": 3.587959968242361,
      "tokens_seen": 1198194688
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003216950852557673,
      "loss": 2.9378,
      "theoretical_loss": 3.5879419117909497,
      "tokens_seen": 1198260224
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003216850551654965,
      "loss": 2.9611,
      "theoretical_loss": 3.587923856603564,
      "tokens_seen": 1198325760
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003216750250752257,
      "loss": 2.6939,
      "theoretical_loss": 3.587905802680047,
      "tokens_seen": 1198391296
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003216649949849549,
      "loss": 2.7802,
      "theoretical_loss": 3.587887750020241,
      "tokens_seen": 1198456832
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032165496489468406,
      "loss": 2.7035,
      "theoretical_loss": 3.587869698623987,
      "tokens_seen": 1198522368
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032164493480441324,
      "loss": 2.7057,
      "theoretical_loss": 3.587851648491129,
      "tokens_seen": 1198587904
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003216349047141424,
      "loss": 2.6861,
      "theoretical_loss": 3.5878335996215096,
      "tokens_seen": 1198653440
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032162487462387165,
      "loss": 2.8873,
      "theoretical_loss": 3.5878155520149706,
      "tokens_seen": 1198718976
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003216148445336008,
      "loss": 2.6793,
      "theoretical_loss": 3.5877975056713547,
      "tokens_seen": 1198784512
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032160481444333,
      "loss": 2.8358,
      "theoretical_loss": 3.5877794605905042,
      "tokens_seen": 1198850048
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032159478435305914,
      "loss": 2.8811,
      "theoretical_loss": 3.5877614167722625,
      "tokens_seen": 1198915584
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003215847542627884,
      "loss": 3.0254,
      "theoretical_loss": 3.5877433742164717,
      "tokens_seen": 1198981120
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032157472417251756,
      "loss": 2.8406,
      "theoretical_loss": 3.587725332922975,
      "tokens_seen": 1199046656
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032156469408224674,
      "loss": 2.8718,
      "theoretical_loss": 3.587707292891614,
      "tokens_seen": 1199112192
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003215546639919759,
      "loss": 2.9169,
      "theoretical_loss": 3.5876892541222327,
      "tokens_seen": 1199177728
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003215446339017051,
      "loss": 2.8994,
      "theoretical_loss": 3.587671216614673,
      "tokens_seen": 1199243264
    },
    {
      "epoch": 0.36,
      "objective/train/advantage_avg": 0.494668185710907,
      "objective/train/docs_used": 682215,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.711491584777832,
      "objective/train/original_loss": 2.711491346359253,
      "objective/train/theoretical_loss": 3.5876531803687786,
      "objective/train/tokens_used": 1219768800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24474357068538666,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050710916519165,
      "objective/train/weighted_lm_loss": 2.848935842514038,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0465744733810425,
      "theoretical_loss": 3.5876531803687786,
      "tokens_seen": 1199308800
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003215346038114343,
      "loss": 3.1007,
      "theoretical_loss": 3.5876531803687786,
      "tokens_seen": 1199308800
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003215245737211635,
      "loss": 2.8218,
      "theoretical_loss": 3.5876351453843913,
      "tokens_seen": 1199374336
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032151454363089265,
      "loss": 2.7335,
      "theoretical_loss": 3.5876171116613547,
      "tokens_seen": 1199439872
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003215045135406219,
      "loss": 2.9252,
      "theoretical_loss": 3.5875990791995105,
      "tokens_seen": 1199505408
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032149448345035106,
      "loss": 2.8222,
      "theoretical_loss": 3.587581047998703,
      "tokens_seen": 1199570944
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032148445336008024,
      "loss": 2.8956,
      "theoretical_loss": 3.5875630180587743,
      "tokens_seen": 1199636480
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003214744232698094,
      "loss": 2.827,
      "theoretical_loss": 3.587544989379568,
      "tokens_seen": 1199702016
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003214643931795386,
      "loss": 2.8241,
      "theoretical_loss": 3.5875269619609265,
      "tokens_seen": 1199767552
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003214543630892678,
      "loss": 2.8059,
      "theoretical_loss": 3.587508935802693,
      "tokens_seen": 1199833088
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.000321444332998997,
      "loss": 2.7452,
      "theoretical_loss": 3.58749091090471,
      "tokens_seen": 1199898624
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032143430290872615,
      "loss": 2.7979,
      "theoretical_loss": 3.587472887266822,
      "tokens_seen": 1199964160
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003214242728184554,
      "loss": 2.8647,
      "theoretical_loss": 3.5874548648888704,
      "tokens_seen": 1200029696
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003214142427281845,
      "loss": 2.8791,
      "theoretical_loss": 3.5874368437706994,
      "tokens_seen": 1200095232
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032140421263791375,
      "loss": 3.0584,
      "theoretical_loss": 3.5874188239121523,
      "tokens_seen": 1200160768
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032139418254764293,
      "loss": 2.8536,
      "theoretical_loss": 3.587400805313071,
      "tokens_seen": 1200226304
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003213841524573721,
      "loss": 2.8928,
      "theoretical_loss": 3.5873827879733007,
      "tokens_seen": 1200291840
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003213741223671013,
      "loss": 2.9507,
      "theoretical_loss": 3.587364771892682,
      "tokens_seen": 1200357376
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003213640922768305,
      "loss": 2.9453,
      "theoretical_loss": 3.587346757071061,
      "tokens_seen": 1200422912
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032135406218655965,
      "loss": 2.6628,
      "theoretical_loss": 3.5873287435082792,
      "tokens_seen": 1200488448
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003213440320962889,
      "loss": 2.5907,
      "theoretical_loss": 3.5873107312041803,
      "tokens_seen": 1200553984
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.000321334002006018,
      "loss": 2.9582,
      "theoretical_loss": 3.587292720158608,
      "tokens_seen": 1200619520
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032132397191574725,
      "loss": 2.5791,
      "theoretical_loss": 3.587274710371405,
      "tokens_seen": 1200685056
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003213139418254765,
      "loss": 2.7454,
      "theoretical_loss": 3.5872567018424153,
      "tokens_seen": 1200750592
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003213039117352056,
      "loss": 3.0935,
      "theoretical_loss": 3.5872386945714823,
      "tokens_seen": 1200816128
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032129388164493485,
      "loss": 2.8884,
      "theoretical_loss": 3.587220688558449,
      "tokens_seen": 1200881664
    },
    {
      "epoch": 0.36,
      "objective/train/advantage_avg": 0.49338802695274353,
      "objective/train/docs_used": 683386,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.1219968795776367,
      "objective/train/original_loss": 2.1219966411590576,
      "objective/train/theoretical_loss": 3.5872026838031594,
      "objective/train/tokens_used": 1221407200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24354267120361328,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505766868591309,
      "objective/train/weighted_lm_loss": 2.229297399520874,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0408108234405518,
      "theoretical_loss": 3.5872026838031594,
      "tokens_seen": 1200947200
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.000321283851554664,
      "loss": 2.7625,
      "theoretical_loss": 3.5872026838031594,
      "tokens_seen": 1200947200
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003212738214643932,
      "loss": 3.0051,
      "theoretical_loss": 3.587184680305457,
      "tokens_seen": 1201012736
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003212637913741224,
      "loss": 2.8319,
      "theoretical_loss": 3.5871666780651847,
      "tokens_seen": 1201078272
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003212537612838516,
      "loss": 3.0602,
      "theoretical_loss": 3.5871486770821868,
      "tokens_seen": 1201143808
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032124373119358075,
      "loss": 3.1134,
      "theoretical_loss": 3.5871306773563068,
      "tokens_seen": 1201209344
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032123370110330994,
      "loss": 2.7597,
      "theoretical_loss": 3.587112678887388,
      "tokens_seen": 1201274880
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003212236710130391,
      "loss": 2.8192,
      "theoretical_loss": 3.587094681675274,
      "tokens_seen": 1201340416
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032121364092276835,
      "loss": 2.972,
      "theoretical_loss": 3.5870766857198086,
      "tokens_seen": 1201405952
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003212036108324975,
      "loss": 2.8996,
      "theoretical_loss": 3.587058691020836,
      "tokens_seen": 1201471488
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003211935807422267,
      "loss": 2.7257,
      "theoretical_loss": 3.5870406975781997,
      "tokens_seen": 1201537024
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003211835506519559,
      "loss": 2.7861,
      "theoretical_loss": 3.5870227053917425,
      "tokens_seen": 1201602560
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003211735205616851,
      "loss": 2.8208,
      "theoretical_loss": 3.58700471446131,
      "tokens_seen": 1201668096
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032116349047141426,
      "loss": 2.7987,
      "theoretical_loss": 3.5869867247867444,
      "tokens_seen": 1201733632
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032115346038114344,
      "loss": 3.1218,
      "theoretical_loss": 3.5869687363678904,
      "tokens_seen": 1201799168
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003211434302908726,
      "loss": 3.057,
      "theoretical_loss": 3.5869507492045916,
      "tokens_seen": 1201864704
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032113340020060185,
      "loss": 2.8908,
      "theoretical_loss": 3.586932763296692,
      "tokens_seen": 1201930240
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.000321123370110331,
      "loss": 2.8413,
      "theoretical_loss": 3.5869147786440356,
      "tokens_seen": 1201995776
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003211133400200602,
      "loss": 2.7157,
      "theoretical_loss": 3.5868967952464663,
      "tokens_seen": 1202061312
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032110330992978934,
      "loss": 2.9363,
      "theoretical_loss": 3.5868788131038283,
      "tokens_seen": 1202126848
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003210932798395186,
      "loss": 3.0315,
      "theoretical_loss": 3.5868608322159647,
      "tokens_seen": 1202192384
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032108324974924776,
      "loss": 2.7667,
      "theoretical_loss": 3.5868428525827207,
      "tokens_seen": 1202257920
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032107321965897694,
      "loss": 2.9668,
      "theoretical_loss": 3.58682487420394,
      "tokens_seen": 1202323456
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003210631895687061,
      "loss": 2.7754,
      "theoretical_loss": 3.586806897079466,
      "tokens_seen": 1202388992
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003210531594784353,
      "loss": 2.7893,
      "theoretical_loss": 3.586788921209144,
      "tokens_seen": 1202454528
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003210431293881645,
      "loss": 2.564,
      "theoretical_loss": 3.586770946592818,
      "tokens_seen": 1202520064
    },
    {
      "epoch": 0.36,
      "objective/train/advantage_avg": 0.49413520097732544,
      "objective/train/docs_used": 683937,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.8437411785125732,
      "objective/train/original_loss": 2.843740940093994,
      "objective/train/theoretical_loss": 3.5867529732303307,
      "objective/train/tokens_used": 1223045600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24424895644187927,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506551265716553,
      "objective/train/weighted_lm_loss": 2.9879066944122314,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0304546356201172,
      "theoretical_loss": 3.5867529732303307,
      "tokens_seen": 1202585600
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003210330992978937,
      "loss": 2.7965,
      "theoretical_loss": 3.5867529732303307,
      "tokens_seen": 1202585600
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032102306920762285,
      "loss": 2.7324,
      "theoretical_loss": 3.586735001121528,
      "tokens_seen": 1202651136
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003210130391173521,
      "loss": 2.8917,
      "theoretical_loss": 3.5867170302662537,
      "tokens_seen": 1202716672
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032100300902708126,
      "loss": 2.8452,
      "theoretical_loss": 3.586699060664352,
      "tokens_seen": 1202782208
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032099297893681044,
      "loss": 3.0703,
      "theoretical_loss": 3.5866810923156667,
      "tokens_seen": 1202847744
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003209829488465396,
      "loss": 2.8001,
      "theoretical_loss": 3.586663125220043,
      "tokens_seen": 1202913280
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003209729187562688,
      "loss": 2.9689,
      "theoretical_loss": 3.5866451593773245,
      "tokens_seen": 1202978816
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.000320962888665998,
      "loss": 2.941,
      "theoretical_loss": 3.586627194787356,
      "tokens_seen": 1203044352
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003209528585757272,
      "loss": 2.8425,
      "theoretical_loss": 3.586609231449982,
      "tokens_seen": 1203109888
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032094282848545635,
      "loss": 2.9357,
      "theoretical_loss": 3.586591269365046,
      "tokens_seen": 1203175424
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003209327983951856,
      "loss": 2.6605,
      "theoretical_loss": 3.586573308532394,
      "tokens_seen": 1203240960
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003209227683049147,
      "loss": 2.6424,
      "theoretical_loss": 3.5865553489518693,
      "tokens_seen": 1203306496
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032091273821464395,
      "loss": 2.7287,
      "theoretical_loss": 3.586537390623317,
      "tokens_seen": 1203372032
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032090270812437313,
      "loss": 3.0751,
      "theoretical_loss": 3.5865194335465818,
      "tokens_seen": 1203437568
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003208926780341023,
      "loss": 2.9636,
      "theoretical_loss": 3.5865014777215074,
      "tokens_seen": 1203503104
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003208826479438315,
      "loss": 2.9016,
      "theoretical_loss": 3.586483523147939,
      "tokens_seen": 1203568640
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003208726178535607,
      "loss": 2.897,
      "theoretical_loss": 3.5864655698257213,
      "tokens_seen": 1203634176
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032086258776328985,
      "loss": 3.032,
      "theoretical_loss": 3.5864476177546987,
      "tokens_seen": 1203699712
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003208525576730191,
      "loss": 2.9931,
      "theoretical_loss": 3.586429666934716,
      "tokens_seen": 1203765248
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003208425275827482,
      "loss": 3.0208,
      "theoretical_loss": 3.586411717365618,
      "tokens_seen": 1203830784
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032083249749247745,
      "loss": 2.9596,
      "theoretical_loss": 3.5863937690472496,
      "tokens_seen": 1203896320
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032082246740220663,
      "loss": 3.0762,
      "theoretical_loss": 3.586375821979455,
      "tokens_seen": 1203961856
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003208124373119358,
      "loss": 2.9868,
      "theoretical_loss": 3.5863578761620794,
      "tokens_seen": 1204027392
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.000320802407221665,
      "loss": 2.9017,
      "theoretical_loss": 3.586339931594968,
      "tokens_seen": 1204092928
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003207923771313942,
      "loss": 2.798,
      "theoretical_loss": 3.5863219882779647,
      "tokens_seen": 1204158464
    },
    {
      "epoch": 0.36,
      "objective/train/advantage_avg": 0.4946693778038025,
      "objective/train/docs_used": 684175,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.6254286766052246,
      "objective/train/original_loss": 2.6254281997680664,
      "objective/train/theoretical_loss": 3.5863040462109153,
      "objective/train/tokens_used": 1224684000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24485968053340912,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507115125656128,
      "objective/train/weighted_lm_loss": 2.758524179458618,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0281672477722168,
      "theoretical_loss": 3.5863040462109153,
      "tokens_seen": 1204224000
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032078234704112336,
      "loss": 2.9131,
      "theoretical_loss": 3.5863040462109153,
      "tokens_seen": 1204224000
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003207723169508526,
      "loss": 3.0719,
      "theoretical_loss": 3.586286105393664,
      "tokens_seen": 1204289536
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003207622868605817,
      "loss": 2.9748,
      "theoretical_loss": 3.5862681658260565,
      "tokens_seen": 1204355072
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.00032075225677031095,
      "loss": 2.6987,
      "theoretical_loss": 3.5862502275079366,
      "tokens_seen": 1204420608
    },
    {
      "epoch": 0.36,
      "learning_rate": 0.0003207422266800401,
      "loss": 3.1864,
      "theoretical_loss": 3.586232290439151,
      "tokens_seen": 1204486144
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003207321965897693,
      "loss": 2.7452,
      "theoretical_loss": 3.5862143546195426,
      "tokens_seen": 1204551680
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003207221664994985,
      "loss": 2.8812,
      "theoretical_loss": 3.5861964200489584,
      "tokens_seen": 1204617216
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003207121364092277,
      "loss": 2.8685,
      "theoretical_loss": 3.5861784867272424,
      "tokens_seen": 1204682752
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00032070210631895686,
      "loss": 2.8826,
      "theoretical_loss": 3.5861605546542403,
      "tokens_seen": 1204748288
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003206920762286861,
      "loss": 2.7916,
      "theoretical_loss": 3.5861426238297964,
      "tokens_seen": 1204813824
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003206820461384152,
      "loss": 2.6645,
      "theoretical_loss": 3.5861246942537566,
      "tokens_seen": 1204879360
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00032067201604814446,
      "loss": 2.8439,
      "theoretical_loss": 3.5861067659259653,
      "tokens_seen": 1204944896
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003206619859578736,
      "loss": 2.8954,
      "theoretical_loss": 3.586088838846269,
      "tokens_seen": 1205010432
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003206519558676028,
      "loss": 2.9732,
      "theoretical_loss": 3.5860709130145114,
      "tokens_seen": 1205075968
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.000320641925777332,
      "loss": 3.0362,
      "theoretical_loss": 3.586052988430539,
      "tokens_seen": 1205141504
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003206318956870612,
      "loss": 2.8833,
      "theoretical_loss": 3.586035065094197,
      "tokens_seen": 1205207040
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00032062186559679036,
      "loss": 3.1382,
      "theoretical_loss": 3.5860171430053303,
      "tokens_seen": 1205272576
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00032061183550651954,
      "loss": 3.256,
      "theoretical_loss": 3.5859992221637835,
      "tokens_seen": 1205338112
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003206018054162487,
      "loss": 3.0581,
      "theoretical_loss": 3.5859813025694036,
      "tokens_seen": 1205403648
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00032059177532597796,
      "loss": 2.9431,
      "theoretical_loss": 3.585963384222035,
      "tokens_seen": 1205469184
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003205817452357071,
      "loss": 2.8773,
      "theoretical_loss": 3.5859454671215234,
      "tokens_seen": 1205534720
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003205717151454363,
      "loss": 2.9371,
      "theoretical_loss": 3.585927551267714,
      "tokens_seen": 1205600256
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003205616850551655,
      "loss": 2.9334,
      "theoretical_loss": 3.585909636660453,
      "tokens_seen": 1205665792
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003205516549648947,
      "loss": 2.9477,
      "theoretical_loss": 3.585891723299585,
      "tokens_seen": 1205731328
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003205416248746239,
      "loss": 3.0263,
      "theoretical_loss": 3.585873811184956,
      "tokens_seen": 1205796864
    },
    {
      "epoch": 0.37,
      "objective/train/advantage_avg": 0.4941776990890503,
      "objective/train/docs_used": 684684,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.0341503620147705,
      "objective/train/original_loss": 3.0341501235961914,
      "objective/train/theoretical_loss": 3.585855900316411,
      "objective/train/tokens_used": 1226322400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24426716566085815,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506595373153687,
      "objective/train/weighted_lm_loss": 3.1877570152282715,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0425469875335693,
      "theoretical_loss": 3.585855900316411,
      "tokens_seen": 1205862400
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00032053159478435305,
      "loss": 2.9313,
      "theoretical_loss": 3.585855900316411,
      "tokens_seen": 1205862400
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003205215646940823,
      "loss": 3.1106,
      "theoretical_loss": 3.585837990693797,
      "tokens_seen": 1205927936
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00032051153460381146,
      "loss": 2.7987,
      "theoretical_loss": 3.5858200823169586,
      "tokens_seen": 1205993472
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00032050150451354064,
      "loss": 2.8481,
      "theoretical_loss": 3.585802175185741,
      "tokens_seen": 1206059008
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003204914744232698,
      "loss": 2.8217,
      "theoretical_loss": 3.5857842692999915,
      "tokens_seen": 1206124544
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.000320481444332999,
      "loss": 2.9263,
      "theoretical_loss": 3.5857663646595537,
      "tokens_seen": 1206190080
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003204714142427282,
      "loss": 2.9477,
      "theoretical_loss": 3.5857484612642754,
      "tokens_seen": 1206255616
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003204613841524574,
      "loss": 3.1476,
      "theoretical_loss": 3.5857305591140007,
      "tokens_seen": 1206321152
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00032045135406218655,
      "loss": 2.8754,
      "theoretical_loss": 3.5857126582085765,
      "tokens_seen": 1206386688
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003204413239719158,
      "loss": 2.7896,
      "theoretical_loss": 3.585694758547848,
      "tokens_seen": 1206452224
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003204312938816449,
      "loss": 2.8301,
      "theoretical_loss": 3.5856768601316613,
      "tokens_seen": 1206517760
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00032042126379137415,
      "loss": 3.0925,
      "theoretical_loss": 3.585658962959862,
      "tokens_seen": 1206583296
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00032041123370110333,
      "loss": 2.7334,
      "theoretical_loss": 3.585641067032297,
      "tokens_seen": 1206648832
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003204012036108325,
      "loss": 2.7519,
      "theoretical_loss": 3.585623172348811,
      "tokens_seen": 1206714368
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003203911735205617,
      "loss": 2.8938,
      "theoretical_loss": 3.58560527890925,
      "tokens_seen": 1206779904
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003203811434302909,
      "loss": 3.117,
      "theoretical_loss": 3.5855873867134607,
      "tokens_seen": 1206845440
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00032037111334002005,
      "loss": 2.8608,
      "theoretical_loss": 3.5855694957612894,
      "tokens_seen": 1206910976
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003203610832497493,
      "loss": 3.0195,
      "theoretical_loss": 3.5855516060525807,
      "tokens_seen": 1206976512
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003203510531594784,
      "loss": 2.8291,
      "theoretical_loss": 3.585533717587182,
      "tokens_seen": 1207042048
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00032034102306920765,
      "loss": 3.026,
      "theoretical_loss": 3.5855158303649386,
      "tokens_seen": 1207107584
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00032033099297893683,
      "loss": 3.094,
      "theoretical_loss": 3.585497944385697,
      "tokens_seen": 1207173120
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.000320320962888666,
      "loss": 2.8886,
      "theoretical_loss": 3.5854800596493037,
      "tokens_seen": 1207238656
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003203109327983952,
      "loss": 2.7939,
      "theoretical_loss": 3.585462176155604,
      "tokens_seen": 1207304192
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003203009027081244,
      "loss": 2.9255,
      "theoretical_loss": 3.5854442939044446,
      "tokens_seen": 1207369728
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00032029087261785356,
      "loss": 3.2163,
      "theoretical_loss": 3.5854264128956723,
      "tokens_seen": 1207435264
    },
    {
      "epoch": 0.37,
      "objective/train/advantage_avg": 0.49277618527412415,
      "objective/train/docs_used": 686035,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.163205623626709,
      "objective/train/original_loss": 3.16320538520813,
      "objective/train/theoretical_loss": 3.585408533129132,
      "objective/train/tokens_used": 1227960800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2429347187280655,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505125522613525,
      "objective/train/weighted_lm_loss": 3.3227405548095703,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 3.585408533129132,
      "tokens_seen": 1207500800
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003202808425275828,
      "loss": 3.1358,
      "theoretical_loss": 3.585408533129132,
      "tokens_seen": 1207500800
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003202708124373119,
      "loss": 3.0835,
      "theoretical_loss": 3.585390654604671,
      "tokens_seen": 1207566336
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00032026078234704115,
      "loss": 3.0764,
      "theoretical_loss": 3.585372777322135,
      "tokens_seen": 1207631872
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003202507522567703,
      "loss": 3.0948,
      "theoretical_loss": 3.5853549012813715,
      "tokens_seen": 1207697408
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003202407221664995,
      "loss": 2.9913,
      "theoretical_loss": 3.5853370264822253,
      "tokens_seen": 1207762944
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003202306920762287,
      "loss": 3.0443,
      "theoretical_loss": 3.5853191529245443,
      "tokens_seen": 1207828480
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003202206619859579,
      "loss": 3.002,
      "theoretical_loss": 3.5853012806081734,
      "tokens_seen": 1207894016
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00032021063189568706,
      "loss": 2.8749,
      "theoretical_loss": 3.58528340953296,
      "tokens_seen": 1207959552
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003202006018054163,
      "loss": 2.6829,
      "theoretical_loss": 3.5852655396987507,
      "tokens_seen": 1208025088
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003201905717151454,
      "loss": 3.2425,
      "theoretical_loss": 3.5852476711053916,
      "tokens_seen": 1208090624
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00032018054162487466,
      "loss": 2.9463,
      "theoretical_loss": 3.5852298037527293,
      "tokens_seen": 1208156160
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003201705115346038,
      "loss": 2.9625,
      "theoretical_loss": 3.5852119376406106,
      "tokens_seen": 1208221696
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.000320160481444333,
      "loss": 2.752,
      "theoretical_loss": 3.585194072768881,
      "tokens_seen": 1208287232
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003201504513540622,
      "loss": 2.7807,
      "theoretical_loss": 3.585176209137389,
      "tokens_seen": 1208352768
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003201404212637914,
      "loss": 2.8146,
      "theoretical_loss": 3.58515834674598,
      "tokens_seen": 1208418304
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00032013039117352056,
      "loss": 2.9341,
      "theoretical_loss": 3.5851404855945006,
      "tokens_seen": 1208483840
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00032012036108324974,
      "loss": 3.1156,
      "theoretical_loss": 3.585122625682798,
      "tokens_seen": 1208549376
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003201103309929789,
      "loss": 2.7974,
      "theoretical_loss": 3.585104767010719,
      "tokens_seen": 1208614912
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00032010030090270816,
      "loss": 3.0401,
      "theoretical_loss": 3.5850869095781093,
      "tokens_seen": 1208680448
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003200902708124373,
      "loss": 2.8388,
      "theoretical_loss": 3.5850690533848173,
      "tokens_seen": 1208745984
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003200802407221665,
      "loss": 2.8455,
      "theoretical_loss": 3.5850511984306888,
      "tokens_seen": 1208811520
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00032007021063189565,
      "loss": 3.1456,
      "theoretical_loss": 3.58503334471557,
      "tokens_seen": 1208877056
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003200601805416249,
      "loss": 2.97,
      "theoretical_loss": 3.585015492239309,
      "tokens_seen": 1208942592
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00032005015045135407,
      "loss": 2.8458,
      "theoretical_loss": 3.5849976410017526,
      "tokens_seen": 1209008128
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00032004012036108325,
      "loss": 2.9733,
      "theoretical_loss": 3.584979791002747,
      "tokens_seen": 1209073664
    },
    {
      "epoch": 0.37,
      "objective/train/advantage_avg": 0.493289589881897,
      "objective/train/docs_used": 686549,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.96972393989563,
      "objective/train/original_loss": 2.969724178314209,
      "objective/train/theoretical_loss": 3.5849619422421393,
      "objective/train/tokens_used": 1229599200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2434336245059967,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505664348602295,
      "objective/train/weighted_lm_loss": 3.1199090480804443,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0338950157165527,
      "theoretical_loss": 3.5849619422421393,
      "tokens_seen": 1209139200
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00032003009027081243,
      "loss": 2.9611,
      "theoretical_loss": 3.5849619422421393,
      "tokens_seen": 1209139200
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00032002006018054166,
      "loss": 2.8798,
      "theoretical_loss": 3.584944094719777,
      "tokens_seen": 1209204736
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003200100300902708,
      "loss": 2.9356,
      "theoretical_loss": 3.5849262484355067,
      "tokens_seen": 1209270272
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00032,
      "loss": 3.0147,
      "theoretical_loss": 3.584908403389176,
      "tokens_seen": 1209335808
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031998996990972915,
      "loss": 3.0329,
      "theoretical_loss": 3.5848905595806304,
      "tokens_seen": 1209401344
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003199799398194584,
      "loss": 2.9072,
      "theoretical_loss": 3.584872717009718,
      "tokens_seen": 1209466880
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031996990972918757,
      "loss": 2.7499,
      "theoretical_loss": 3.5848548756762866,
      "tokens_seen": 1209532416
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031995987963891675,
      "loss": 2.9899,
      "theoretical_loss": 3.584837035580182,
      "tokens_seen": 1209597952
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031994984954864593,
      "loss": 2.8698,
      "theoretical_loss": 3.5848191967212526,
      "tokens_seen": 1209663488
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003199398194583751,
      "loss": 2.8148,
      "theoretical_loss": 3.5848013590993446,
      "tokens_seen": 1209729024
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003199297893681043,
      "loss": 3.0905,
      "theoretical_loss": 3.5847835227143054,
      "tokens_seen": 1209794560
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031991975927783353,
      "loss": 2.8505,
      "theoretical_loss": 3.584765687565983,
      "tokens_seen": 1209860096
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031990972918756266,
      "loss": 2.8882,
      "theoretical_loss": 3.5847478536542234,
      "tokens_seen": 1209925632
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003198996990972919,
      "loss": 3.0227,
      "theoretical_loss": 3.584730020978875,
      "tokens_seen": 1209991168
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.000319889669007021,
      "loss": 2.994,
      "theoretical_loss": 3.5847121895397844,
      "tokens_seen": 1210056704
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031987963891675025,
      "loss": 3.0572,
      "theoretical_loss": 3.5846943593367997,
      "tokens_seen": 1210122240
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031986960882647943,
      "loss": 3.0162,
      "theoretical_loss": 3.5846765303697676,
      "tokens_seen": 1210187776
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003198595787362086,
      "loss": 2.8967,
      "theoretical_loss": 3.5846587026385355,
      "tokens_seen": 1210253312
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003198495486459378,
      "loss": 2.8837,
      "theoretical_loss": 3.5846408761429513,
      "tokens_seen": 1210318848
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031983951855566703,
      "loss": 2.9354,
      "theoretical_loss": 3.584623050882862,
      "tokens_seen": 1210384384
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031982948846539616,
      "loss": 2.7457,
      "theoretical_loss": 3.5846052268581157,
      "tokens_seen": 1210449920
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003198194583751254,
      "loss": 2.7918,
      "theoretical_loss": 3.584587404068559,
      "tokens_seen": 1210515456
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003198094282848546,
      "loss": 2.8931,
      "theoretical_loss": 3.5845695825140407,
      "tokens_seen": 1210580992
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031979939819458376,
      "loss": 2.8211,
      "theoretical_loss": 3.5845517621944065,
      "tokens_seen": 1210646528
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.000319789368104313,
      "loss": 2.6835,
      "theoretical_loss": 3.584533943109506,
      "tokens_seen": 1210712064
    },
    {
      "epoch": 0.37,
      "objective/train/advantage_avg": 0.4944123327732086,
      "objective/train/docs_used": 687794,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.987743854522705,
      "objective/train/original_loss": 2.987744092941284,
      "objective/train/theoretical_loss": 3.5845161252591855,
      "objective/train/tokens_used": 1231237600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24449171125888824,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506839752197266,
      "objective/train/weighted_lm_loss": 3.1392366886138916,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0442860126495361,
      "theoretical_loss": 3.5845161252591855,
      "tokens_seen": 1210777600
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003197793380140421,
      "loss": 2.891,
      "theoretical_loss": 3.5845161252591855,
      "tokens_seen": 1210777600
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031976930792377135,
      "loss": 3.2132,
      "theoretical_loss": 3.584498308643293,
      "tokens_seen": 1210843136
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003197592778335005,
      "loss": 2.8424,
      "theoretical_loss": 3.5844804932616765,
      "tokens_seen": 1210908672
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003197492477432297,
      "loss": 3.0229,
      "theoretical_loss": 3.5844626791141834,
      "tokens_seen": 1210974208
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003197392176529589,
      "loss": 2.9564,
      "theoretical_loss": 3.5844448662006614,
      "tokens_seen": 1211039744
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003197291875626881,
      "loss": 3.162,
      "theoretical_loss": 3.5844270545209582,
      "tokens_seen": 1211105280
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031971915747241726,
      "loss": 2.9476,
      "theoretical_loss": 3.584409244074922,
      "tokens_seen": 1211170816
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003197091273821465,
      "loss": 2.8051,
      "theoretical_loss": 3.5843914348623995,
      "tokens_seen": 1211236352
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003196990972918756,
      "loss": 2.982,
      "theoretical_loss": 3.5843736268832402,
      "tokens_seen": 1211301888
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031968906720160486,
      "loss": 2.99,
      "theoretical_loss": 3.584355820137291,
      "tokens_seen": 1211367424
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.000319679037111334,
      "loss": 2.8351,
      "theoretical_loss": 3.5843380146243993,
      "tokens_seen": 1211432960
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003196690070210632,
      "loss": 2.7895,
      "theoretical_loss": 3.584320210344414,
      "tokens_seen": 1211498496
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003196589769307924,
      "loss": 2.8176,
      "theoretical_loss": 3.5843024072971827,
      "tokens_seen": 1211564032
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003196489468405216,
      "loss": 2.849,
      "theoretical_loss": 3.5842846054825532,
      "tokens_seen": 1211629568
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031963891675025076,
      "loss": 2.8927,
      "theoretical_loss": 3.584266804900374,
      "tokens_seen": 1211695104
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031962888665997994,
      "loss": 2.8451,
      "theoretical_loss": 3.5842490055504923,
      "tokens_seen": 1211760640
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003196188565697091,
      "loss": 2.8379,
      "theoretical_loss": 3.5842312074327567,
      "tokens_seen": 1211826176
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031960882647943836,
      "loss": 3.0347,
      "theoretical_loss": 3.584213410547015,
      "tokens_seen": 1211891712
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003195987963891675,
      "loss": 2.8775,
      "theoretical_loss": 3.5841956148931153,
      "tokens_seen": 1211957248
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003195887662988967,
      "loss": 2.9503,
      "theoretical_loss": 3.5841778204709067,
      "tokens_seen": 1212022784
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031957873620862585,
      "loss": 2.9812,
      "theoretical_loss": 3.5841600272802356,
      "tokens_seen": 1212088320
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003195687061183551,
      "loss": 2.9546,
      "theoretical_loss": 3.584142235320952,
      "tokens_seen": 1212153856
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031955867602808427,
      "loss": 2.9968,
      "theoretical_loss": 3.584124444592903,
      "tokens_seen": 1212219392
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031954864593781345,
      "loss": 3.0412,
      "theoretical_loss": 3.584106655095937,
      "tokens_seen": 1212284928
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031953861584754263,
      "loss": 3.0003,
      "theoretical_loss": 3.584088866829902,
      "tokens_seen": 1212350464
    },
    {
      "epoch": 0.37,
      "objective/train/advantage_avg": 0.493076890707016,
      "objective/train/docs_used": 688453,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.25057315826416,
      "objective/train/original_loss": 2.2505733966827393,
      "objective/train/theoretical_loss": 3.584071079794647,
      "objective/train/tokens_used": 1232876000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24319201707839966,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505437850952148,
      "objective/train/weighted_lm_loss": 2.364032506942749,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0451053380966187,
      "theoretical_loss": 3.584071079794647,
      "tokens_seen": 1212416000
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031952858575727186,
      "loss": 2.7448,
      "theoretical_loss": 3.584071079794647,
      "tokens_seen": 1212416000
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.000319518555667001,
      "loss": 2.8734,
      "theoretical_loss": 3.58405329399002,
      "tokens_seen": 1212481536
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003195085255767302,
      "loss": 2.8651,
      "theoretical_loss": 3.5840355094158696,
      "tokens_seen": 1212547072
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031949849548645935,
      "loss": 2.9898,
      "theoretical_loss": 3.5840177260720436,
      "tokens_seen": 1212612608
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003194884653961886,
      "loss": 2.8683,
      "theoretical_loss": 3.5839999439583905,
      "tokens_seen": 1212678144
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031947843530591777,
      "loss": 2.7027,
      "theoretical_loss": 3.583982163074759,
      "tokens_seen": 1212743680
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031946840521564695,
      "loss": 2.8909,
      "theoretical_loss": 3.5839643834209975,
      "tokens_seen": 1212809216
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031945837512537613,
      "loss": 2.7257,
      "theoretical_loss": 3.583946604996955,
      "tokens_seen": 1212874752
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003194483450351053,
      "loss": 2.7724,
      "theoretical_loss": 3.5839288278024783,
      "tokens_seen": 1212940288
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003194383149448345,
      "loss": 2.9691,
      "theoretical_loss": 3.583911051837418,
      "tokens_seen": 1213005824
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031942828485456373,
      "loss": 3.0101,
      "theoretical_loss": 3.5838932771016214,
      "tokens_seen": 1213071360
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031941825476429286,
      "loss": 2.9396,
      "theoretical_loss": 3.583875503594938,
      "tokens_seen": 1213136896
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003194082246740221,
      "loss": 2.9119,
      "theoretical_loss": 3.5838577313172157,
      "tokens_seen": 1213202432
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003193981945837512,
      "loss": 2.7619,
      "theoretical_loss": 3.583839960268303,
      "tokens_seen": 1213267968
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031938816449348045,
      "loss": 3.0171,
      "theoretical_loss": 3.583822190448049,
      "tokens_seen": 1213333504
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031937813440320964,
      "loss": 3.1015,
      "theoretical_loss": 3.583804421856302,
      "tokens_seen": 1213399040
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003193681043129388,
      "loss": 2.6702,
      "theoretical_loss": 3.5837866544929113,
      "tokens_seen": 1213464576
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.000319358074222668,
      "loss": 2.8442,
      "theoretical_loss": 3.5837688883577252,
      "tokens_seen": 1213530112
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031934804413239723,
      "loss": 3.0802,
      "theoretical_loss": 3.5837511234505928,
      "tokens_seen": 1213595648
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031933801404212636,
      "loss": 2.9608,
      "theoretical_loss": 3.5837333597713625,
      "tokens_seen": 1213661184
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003193279839518556,
      "loss": 2.66,
      "theoretical_loss": 3.5837155973198835,
      "tokens_seen": 1213726720
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003193179538615847,
      "loss": 2.7638,
      "theoretical_loss": 3.583697836096005,
      "tokens_seen": 1213792256
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031930792377131396,
      "loss": 2.9624,
      "theoretical_loss": 3.5836800760995744,
      "tokens_seen": 1213857792
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031929789368104314,
      "loss": 2.905,
      "theoretical_loss": 3.5836623173304423,
      "tokens_seen": 1213923328
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003192878635907723,
      "loss": 2.8342,
      "theoretical_loss": 3.583644559788457,
      "tokens_seen": 1213988864
    },
    {
      "epoch": 0.37,
      "objective/train/advantage_avg": 0.4929000437259674,
      "objective/train/docs_used": 689723,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.400998830795288,
      "objective/train/original_loss": 2.400998830795288,
      "objective/train/theoretical_loss": 3.583626803473467,
      "objective/train/tokens_used": 1234514400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24302543699741364,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050525426864624,
      "objective/train/weighted_lm_loss": 2.5221362113952637,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0417574644088745,
      "theoretical_loss": 3.583626803473467,
      "tokens_seen": 1214054400
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003192778335005015,
      "loss": 2.6496,
      "theoretical_loss": 3.583626803473467,
      "tokens_seen": 1214054400
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003192678034102307,
      "loss": 2.9228,
      "theoretical_loss": 3.5836090483853216,
      "tokens_seen": 1214119936
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031925777331995986,
      "loss": 3.0045,
      "theoretical_loss": 3.5835912945238704,
      "tokens_seen": 1214185472
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003192477432296891,
      "loss": 3.0679,
      "theoretical_loss": 3.5835735418889616,
      "tokens_seen": 1214251008
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003192377131394182,
      "loss": 3.0291,
      "theoretical_loss": 3.583555790480445,
      "tokens_seen": 1214316544
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031922768304914746,
      "loss": 2.8142,
      "theoretical_loss": 3.583538040298169,
      "tokens_seen": 1214382080
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003192176529588766,
      "loss": 2.9329,
      "theoretical_loss": 3.583520291341983,
      "tokens_seen": 1214447616
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003192076228686058,
      "loss": 2.8979,
      "theoretical_loss": 3.5835025436117363,
      "tokens_seen": 1214513152
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.000319197592778335,
      "loss": 2.8586,
      "theoretical_loss": 3.5834847971072783,
      "tokens_seen": 1214578688
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003191875626880642,
      "loss": 2.8014,
      "theoretical_loss": 3.5834670518284577,
      "tokens_seen": 1214644224
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031917753259779337,
      "loss": 2.8034,
      "theoretical_loss": 3.5834493077751235,
      "tokens_seen": 1214709760
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003191675025075226,
      "loss": 2.9528,
      "theoretical_loss": 3.5834315649471256,
      "tokens_seen": 1214775296
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031915747241725173,
      "loss": 2.9849,
      "theoretical_loss": 3.583413823344313,
      "tokens_seen": 1214840832
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031914744232698096,
      "loss": 2.9756,
      "theoretical_loss": 3.5833960829665354,
      "tokens_seen": 1214906368
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003191374122367101,
      "loss": 2.9389,
      "theoretical_loss": 3.5833783438136417,
      "tokens_seen": 1214971904
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003191273821464393,
      "loss": 2.94,
      "theoretical_loss": 3.583360605885481,
      "tokens_seen": 1215037440
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003191173520561685,
      "loss": 2.949,
      "theoretical_loss": 3.5833428691819034,
      "tokens_seen": 1215102976
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003191073219658977,
      "loss": 2.913,
      "theoretical_loss": 3.583325133702758,
      "tokens_seen": 1215168512
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031909729187562687,
      "loss": 3.0528,
      "theoretical_loss": 3.583307399447894,
      "tokens_seen": 1215234048
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031908726178535605,
      "loss": 2.7662,
      "theoretical_loss": 3.583289666417161,
      "tokens_seen": 1215299584
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031907723169508523,
      "loss": 2.7871,
      "theoretical_loss": 3.5832719346104094,
      "tokens_seen": 1215365120
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031906720160481447,
      "loss": 2.6911,
      "theoretical_loss": 3.5832542040274875,
      "tokens_seen": 1215430656
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031905717151454365,
      "loss": 2.6634,
      "theoretical_loss": 3.5832364746682446,
      "tokens_seen": 1215496192
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031904714142427283,
      "loss": 2.8101,
      "theoretical_loss": 3.5832187465325314,
      "tokens_seen": 1215561728
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031903711133400206,
      "loss": 2.7484,
      "theoretical_loss": 3.5832010196201973,
      "tokens_seen": 1215627264
    },
    {
      "epoch": 0.37,
      "objective/train/advantage_avg": 0.49280720949172974,
      "objective/train/docs_used": 690230,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.2011613845825195,
      "objective/train/original_loss": 2.2011613845825195,
      "objective/train/theoretical_loss": 3.583183293931091,
      "objective/train/tokens_used": 1236152800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24301664531230927,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050516128540039,
      "objective/train/weighted_lm_loss": 2.312344789505005,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 3.583183293931091,
      "tokens_seen": 1215692800
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003190270812437312,
      "loss": 2.6929,
      "theoretical_loss": 3.583183293931091,
      "tokens_seen": 1215692800
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003190170511534604,
      "loss": 2.9149,
      "theoretical_loss": 3.5831655694650637,
      "tokens_seen": 1215758336
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031900702106318955,
      "loss": 2.7961,
      "theoretical_loss": 3.583147846221964,
      "tokens_seen": 1215823872
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003189969909729188,
      "loss": 3.0062,
      "theoretical_loss": 3.583130124201641,
      "tokens_seen": 1215889408
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031898696088264797,
      "loss": 2.8007,
      "theoretical_loss": 3.583112403403946,
      "tokens_seen": 1215954944
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031897693079237715,
      "loss": 2.8099,
      "theoretical_loss": 3.583094683828728,
      "tokens_seen": 1216020480
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031896690070210633,
      "loss": 2.853,
      "theoretical_loss": 3.5830769654758368,
      "tokens_seen": 1216086016
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003189568706118355,
      "loss": 2.8227,
      "theoretical_loss": 3.583059248345122,
      "tokens_seen": 1216151552
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003189468405215647,
      "loss": 2.9501,
      "theoretical_loss": 3.5830415324364338,
      "tokens_seen": 1216217088
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031893681043129393,
      "loss": 2.8237,
      "theoretical_loss": 3.5830238177496216,
      "tokens_seen": 1216282624
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031892678034102306,
      "loss": 3.0162,
      "theoretical_loss": 3.5830061042845363,
      "tokens_seen": 1216348160
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003189167502507523,
      "loss": 2.8146,
      "theoretical_loss": 3.5829883920410266,
      "tokens_seen": 1216413696
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003189067201604814,
      "loss": 2.7427,
      "theoretical_loss": 3.582970681018943,
      "tokens_seen": 1216479232
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031889669007021065,
      "loss": 3.0124,
      "theoretical_loss": 3.582952971218136,
      "tokens_seen": 1216544768
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031888665997993984,
      "loss": 2.8706,
      "theoretical_loss": 3.582935262638455,
      "tokens_seen": 1216610304
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.000318876629889669,
      "loss": 2.7741,
      "theoretical_loss": 3.58291755527975,
      "tokens_seen": 1216675840
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003188665997993982,
      "loss": 2.9664,
      "theoretical_loss": 3.582899849141871,
      "tokens_seen": 1216741376
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031885656970912743,
      "loss": 3.0728,
      "theoretical_loss": 3.5828821442246683,
      "tokens_seen": 1216806912
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031884653961885656,
      "loss": 3.0198,
      "theoretical_loss": 3.5828644405279917,
      "tokens_seen": 1216872448
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003188365095285858,
      "loss": 3.0204,
      "theoretical_loss": 3.5828467380516917,
      "tokens_seen": 1216937984
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003188264794383149,
      "loss": 2.7647,
      "theoretical_loss": 3.582829036795619,
      "tokens_seen": 1217003520
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031881644934804416,
      "loss": 2.8981,
      "theoretical_loss": 3.5828113367596224,
      "tokens_seen": 1217069056
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031880641925777334,
      "loss": 2.9846,
      "theoretical_loss": 3.582793637943553,
      "tokens_seen": 1217134592
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003187963891675025,
      "loss": 2.9085,
      "theoretical_loss": 3.582775940347261,
      "tokens_seen": 1217200128
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003187863590772317,
      "loss": 3.16,
      "theoretical_loss": 3.5827582439705967,
      "tokens_seen": 1217265664
    },
    {
      "epoch": 0.37,
      "objective/train/advantage_avg": 0.49530959129333496,
      "objective/train/docs_used": 691368,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.5830583572387695,
      "objective/train/original_loss": 2.5830588340759277,
      "objective/train/theoretical_loss": 3.58274054881341,
      "objective/train/tokens_used": 1237791200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24537278711795807,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507783889770508,
      "objective/train/weighted_lm_loss": 2.7141616344451904,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0425469875335693,
      "theoretical_loss": 3.58274054881341,
      "tokens_seen": 1217331200
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003187763289869609,
      "loss": 2.7049,
      "theoretical_loss": 3.58274054881341,
      "tokens_seen": 1217331200
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031876629889669006,
      "loss": 2.9855,
      "theoretical_loss": 3.582722854875552,
      "tokens_seen": 1217396736
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003187562688064193,
      "loss": 2.9424,
      "theoretical_loss": 3.5827051621568717,
      "tokens_seen": 1217462272
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003187462387161484,
      "loss": 2.8853,
      "theoretical_loss": 3.582687470657221,
      "tokens_seen": 1217527808
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031873620862587766,
      "loss": 2.9246,
      "theoretical_loss": 3.5826697803764493,
      "tokens_seen": 1217593344
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003187261785356068,
      "loss": 3.0946,
      "theoretical_loss": 3.5826520913144075,
      "tokens_seen": 1217658880
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.000318716148445336,
      "loss": 3.0397,
      "theoretical_loss": 3.5826344034709456,
      "tokens_seen": 1217724416
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003187061183550652,
      "loss": 2.981,
      "theoretical_loss": 3.5826167168459144,
      "tokens_seen": 1217789952
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003186960882647944,
      "loss": 2.8306,
      "theoretical_loss": 3.5825990314391642,
      "tokens_seen": 1217855488
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031868605817452357,
      "loss": 3.1044,
      "theoretical_loss": 3.5825813472505463,
      "tokens_seen": 1217921024
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003186760280842528,
      "loss": 2.7342,
      "theoretical_loss": 3.58256366427991,
      "tokens_seen": 1217986560
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031866599799398193,
      "loss": 3.0069,
      "theoretical_loss": 3.582545982527107,
      "tokens_seen": 1218052096
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031865596790371116,
      "loss": 2.8138,
      "theoretical_loss": 3.582528301991987,
      "tokens_seen": 1218117632
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003186459378134403,
      "loss": 2.8242,
      "theoretical_loss": 3.582510622674401,
      "tokens_seen": 1218183168
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003186359077231695,
      "loss": 2.883,
      "theoretical_loss": 3.5824929445742004,
      "tokens_seen": 1218248704
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003186258776328987,
      "loss": 3.0886,
      "theoretical_loss": 3.5824752676912346,
      "tokens_seen": 1218314240
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003186158475426279,
      "loss": 2.7919,
      "theoretical_loss": 3.582457592025355,
      "tokens_seen": 1218379776
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031860581745235707,
      "loss": 3.0245,
      "theoretical_loss": 3.5824399175764126,
      "tokens_seen": 1218445312
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031859578736208625,
      "loss": 2.908,
      "theoretical_loss": 3.582422244344258,
      "tokens_seen": 1218510848
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031858575727181543,
      "loss": 2.8078,
      "theoretical_loss": 3.582404572328741,
      "tokens_seen": 1218576384
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031857572718154467,
      "loss": 2.9898,
      "theoretical_loss": 3.582386901529714,
      "tokens_seen": 1218641920
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003185656970912738,
      "loss": 2.8701,
      "theoretical_loss": 3.582369231947027,
      "tokens_seen": 1218707456
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031855566700100303,
      "loss": 2.9237,
      "theoretical_loss": 3.5823515635805308,
      "tokens_seen": 1218772992
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003185456369107322,
      "loss": 2.8051,
      "theoretical_loss": 3.582333896430076,
      "tokens_seen": 1218838528
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003185356068204614,
      "loss": 2.7019,
      "theoretical_loss": 3.5823162304955147,
      "tokens_seen": 1218904064
    },
    {
      "epoch": 0.37,
      "objective/train/advantage_avg": 0.4930156469345093,
      "objective/train/docs_used": 692060,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.961686134338379,
      "objective/train/original_loss": 2.9616858959198,
      "objective/train/theoretical_loss": 3.5822985657766973,
      "objective/train/tokens_used": 1239429600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2431536167860031,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505375862121582,
      "objective/train/weighted_lm_loss": 3.111255168914795,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.044721245765686,
      "theoretical_loss": 3.5822985657766973,
      "tokens_seen": 1218969600
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031852557673019057,
      "loss": 3.1432,
      "theoretical_loss": 3.5822985657766973,
      "tokens_seen": 1218969600
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031851554663991975,
      "loss": 2.6989,
      "theoretical_loss": 3.5822809022734745,
      "tokens_seen": 1219035136
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031850551654964893,
      "loss": 2.9831,
      "theoretical_loss": 3.582263239985697,
      "tokens_seen": 1219100672
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031849548645937817,
      "loss": 2.8759,
      "theoretical_loss": 3.5822455789132164,
      "tokens_seen": 1219166208
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003184854563691073,
      "loss": 2.8288,
      "theoretical_loss": 3.582227919055884,
      "tokens_seen": 1219231744
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031847542627883653,
      "loss": 2.7167,
      "theoretical_loss": 3.58221026041355,
      "tokens_seen": 1219297280
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031846539618856566,
      "loss": 2.8903,
      "theoretical_loss": 3.5821926029860665,
      "tokens_seen": 1219362816
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003184553660982949,
      "loss": 3.117,
      "theoretical_loss": 3.582174946773284,
      "tokens_seen": 1219428352
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003184453360080241,
      "loss": 2.6791,
      "theoretical_loss": 3.5821572917750535,
      "tokens_seen": 1219493888
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031843530591775326,
      "loss": 2.7119,
      "theoretical_loss": 3.582139637991227,
      "tokens_seen": 1219559424
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031842527582748244,
      "loss": 2.9566,
      "theoretical_loss": 3.5821219854216553,
      "tokens_seen": 1219624960
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003184152457372116,
      "loss": 2.9094,
      "theoretical_loss": 3.582104334066189,
      "tokens_seen": 1219690496
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003184052156469408,
      "loss": 3.0037,
      "theoretical_loss": 3.5820866839246803,
      "tokens_seen": 1219756032
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031839518555667004,
      "loss": 3.1917,
      "theoretical_loss": 3.58206903499698,
      "tokens_seen": 1219821568
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031838515546639916,
      "loss": 2.8739,
      "theoretical_loss": 3.58205138728294,
      "tokens_seen": 1219887104
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003183751253761284,
      "loss": 2.6709,
      "theoretical_loss": 3.5820337407824105,
      "tokens_seen": 1219952640
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003183650952858576,
      "loss": 2.9318,
      "theoretical_loss": 3.582016095495244,
      "tokens_seen": 1220018176
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031835506519558676,
      "loss": 2.8496,
      "theoretical_loss": 3.581998451421291,
      "tokens_seen": 1220083712
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031834503510531594,
      "loss": 2.9362,
      "theoretical_loss": 3.581980808560404,
      "tokens_seen": 1220149248
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003183350050150451,
      "loss": 2.9567,
      "theoretical_loss": 3.581963166912433,
      "tokens_seen": 1220214784
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003183249749247743,
      "loss": 2.6683,
      "theoretical_loss": 3.5819455264772317,
      "tokens_seen": 1220280320
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031831494483450354,
      "loss": 2.9717,
      "theoretical_loss": 3.581927887254649,
      "tokens_seen": 1220345856
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003183049147442327,
      "loss": 2.7359,
      "theoretical_loss": 3.581910249244538,
      "tokens_seen": 1220411392
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003182948846539619,
      "loss": 2.8551,
      "theoretical_loss": 3.58189261244675,
      "tokens_seen": 1220476928
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003182848545636911,
      "loss": 2.8479,
      "theoretical_loss": 3.5818749768611364,
      "tokens_seen": 1220542464
    },
    {
      "epoch": 0.37,
      "objective/train/advantage_avg": 0.49461600184440613,
      "objective/train/docs_used": 693613,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.3325247764587402,
      "objective/train/original_loss": 2.332524538040161,
      "objective/train/theoretical_loss": 3.581857342487549,
      "objective/train/tokens_used": 1241068000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24469751119613647,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507054328918457,
      "objective/train/weighted_lm_loss": 2.4508261680603027,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0363596677780151,
      "theoretical_loss": 3.581857342487549,
      "tokens_seen": 1220608000
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031827482447342026,
      "loss": 2.7874,
      "theoretical_loss": 3.581857342487549,
      "tokens_seen": 1220608000
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003182647943831495,
      "loss": 2.894,
      "theoretical_loss": 3.5818397093258394,
      "tokens_seen": 1220673536
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003182547642928786,
      "loss": 2.8991,
      "theoretical_loss": 3.5818220773758584,
      "tokens_seen": 1220739072
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031824473420260786,
      "loss": 2.8508,
      "theoretical_loss": 3.5818044466374594,
      "tokens_seen": 1220804608
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.000318234704112337,
      "loss": 3.0083,
      "theoretical_loss": 3.581786817110493,
      "tokens_seen": 1220870144
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003182246740220662,
      "loss": 2.7952,
      "theoretical_loss": 3.581769188794811,
      "tokens_seen": 1220935680
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003182146439317954,
      "loss": 2.9842,
      "theoretical_loss": 3.581751561690265,
      "tokens_seen": 1221001216
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003182046138415246,
      "loss": 2.5487,
      "theoretical_loss": 3.5817339357967075,
      "tokens_seen": 1221066752
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031819458375125377,
      "loss": 2.9104,
      "theoretical_loss": 3.58171631111399,
      "tokens_seen": 1221132288
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.000318184553660983,
      "loss": 2.591,
      "theoretical_loss": 3.5816986876419636,
      "tokens_seen": 1221197824
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031817452357071213,
      "loss": 2.8631,
      "theoretical_loss": 3.581681065380481,
      "tokens_seen": 1221263360
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031816449348044136,
      "loss": 2.9364,
      "theoretical_loss": 3.5816634443293944,
      "tokens_seen": 1221328896
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003181544633901705,
      "loss": 3.1533,
      "theoretical_loss": 3.5816458244885547,
      "tokens_seen": 1221394432
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003181444332998997,
      "loss": 2.7861,
      "theoretical_loss": 3.5816282058578146,
      "tokens_seen": 1221459968
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003181344032096289,
      "loss": 2.7903,
      "theoretical_loss": 3.5816105884370257,
      "tokens_seen": 1221525504
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003181243731193581,
      "loss": 2.8759,
      "theoretical_loss": 3.5815929722260402,
      "tokens_seen": 1221591040
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031811434302908727,
      "loss": 2.7841,
      "theoretical_loss": 3.58157535722471,
      "tokens_seen": 1221656576
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031810431293881645,
      "loss": 2.84,
      "theoretical_loss": 3.5815577434328874,
      "tokens_seen": 1221722112
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031809428284854563,
      "loss": 2.7862,
      "theoretical_loss": 3.581540130850424,
      "tokens_seen": 1221787648
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031808425275827487,
      "loss": 2.5297,
      "theoretical_loss": 3.5815225194771725,
      "tokens_seen": 1221853184
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.000318074222668004,
      "loss": 2.914,
      "theoretical_loss": 3.5815049093129847,
      "tokens_seen": 1221918720
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031806419257773323,
      "loss": 2.8949,
      "theoretical_loss": 3.581487300357712,
      "tokens_seen": 1221984256
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003180541624874624,
      "loss": 2.7751,
      "theoretical_loss": 3.5814696926112086,
      "tokens_seen": 1222049792
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003180441323971916,
      "loss": 2.7165,
      "theoretical_loss": 3.5814520860733245,
      "tokens_seen": 1222115328
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031803410230692077,
      "loss": 2.9741,
      "theoretical_loss": 3.5814344807439134,
      "tokens_seen": 1222180864
    },
    {
      "epoch": 0.37,
      "objective/train/advantage_avg": 0.49362555146217346,
      "objective/train/docs_used": 694027,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.877548933029175,
      "objective/train/original_loss": 2.8775486946105957,
      "objective/train/theoretical_loss": 3.5814168766228267,
      "objective/train/tokens_used": 1242706400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24380287528038025,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506019592285156,
      "objective/train/weighted_lm_loss": 3.023139238357544,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 3.5814168766228267,
      "tokens_seen": 1222246400
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031802407221664995,
      "loss": 3.0081,
      "theoretical_loss": 3.5814168766228267,
      "tokens_seen": 1222246400
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031801404212637913,
      "loss": 2.9164,
      "theoretical_loss": 3.581399273709917,
      "tokens_seen": 1222311936
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031800401203610837,
      "loss": 2.7648,
      "theoretical_loss": 3.5813816720050373,
      "tokens_seen": 1222377472
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003179939819458375,
      "loss": 2.8332,
      "theoretical_loss": 3.581364071508039,
      "tokens_seen": 1222443008
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031798395185556673,
      "loss": 2.7724,
      "theoretical_loss": 3.5813464722187742,
      "tokens_seen": 1222508544
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031797392176529586,
      "loss": 2.8354,
      "theoretical_loss": 3.5813288741370966,
      "tokens_seen": 1222574080
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003179638916750251,
      "loss": 2.9493,
      "theoretical_loss": 3.5813112772628575,
      "tokens_seen": 1222639616
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003179538615847543,
      "loss": 3.0451,
      "theoretical_loss": 3.5812936815959096,
      "tokens_seen": 1222705152
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031794383149448346,
      "loss": 2.9811,
      "theoretical_loss": 3.5812760871361053,
      "tokens_seen": 1222770688
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031793380140421264,
      "loss": 2.5742,
      "theoretical_loss": 3.581258493883298,
      "tokens_seen": 1222836224
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003179237713139418,
      "loss": 2.7586,
      "theoretical_loss": 3.581240901837339,
      "tokens_seen": 1222901760
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.000317913741223671,
      "loss": 2.884,
      "theoretical_loss": 3.581223310998081,
      "tokens_seen": 1222967296
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031790371113340024,
      "loss": 2.7587,
      "theoretical_loss": 3.5812057213653774,
      "tokens_seen": 1223032832
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031789368104312936,
      "loss": 2.833,
      "theoretical_loss": 3.5811881329390802,
      "tokens_seen": 1223098368
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003178836509528586,
      "loss": 2.9403,
      "theoretical_loss": 3.5811705457190417,
      "tokens_seen": 1223163904
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003178736208625878,
      "loss": 2.8231,
      "theoretical_loss": 3.5811529597051153,
      "tokens_seen": 1223229440
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031786359077231696,
      "loss": 2.8883,
      "theoretical_loss": 3.581135374897153,
      "tokens_seen": 1223294976
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031785356068204614,
      "loss": 2.7325,
      "theoretical_loss": 3.5811177912950085,
      "tokens_seen": 1223360512
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003178435305917753,
      "loss": 3.0974,
      "theoretical_loss": 3.5811002088985333,
      "tokens_seen": 1223426048
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003178335005015045,
      "loss": 2.8497,
      "theoretical_loss": 3.5810826277075805,
      "tokens_seen": 1223491584
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031782347041123374,
      "loss": 3.0737,
      "theoretical_loss": 3.581065047722004,
      "tokens_seen": 1223557120
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031781344032096287,
      "loss": 3.1281,
      "theoretical_loss": 3.581047468941655,
      "tokens_seen": 1223622656
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003178034102306921,
      "loss": 2.8219,
      "theoretical_loss": 3.581029891366387,
      "tokens_seen": 1223688192
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031779338014042123,
      "loss": 2.9045,
      "theoretical_loss": 3.581012314996053,
      "tokens_seen": 1223753728
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031778335005015046,
      "loss": 2.8626,
      "theoretical_loss": 3.5809947398305058,
      "tokens_seen": 1223819264
    },
    {
      "epoch": 0.37,
      "objective/train/advantage_avg": 0.49368706345558167,
      "objective/train/docs_used": 695222,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.486943483352661,
      "objective/train/original_loss": 2.486943006515503,
      "objective/train/theoretical_loss": 3.580977165869598,
      "objective/train/tokens_used": 1244344800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24388079345226288,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050608515739441,
      "objective/train/weighted_lm_loss": 2.6129512786865234,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0363596677780151,
      "theoretical_loss": 3.580977165869598,
      "tokens_seen": 1223884800
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031777331995987964,
      "loss": 2.798,
      "theoretical_loss": 3.580977165869598,
      "tokens_seen": 1223884800
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003177632898696088,
      "loss": 3.0927,
      "theoretical_loss": 3.580959593113183,
      "tokens_seen": 1223950336
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.000317753259779338,
      "loss": 2.6978,
      "theoretical_loss": 3.580942021561113,
      "tokens_seen": 1224015872
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003177432296890672,
      "loss": 2.7801,
      "theoretical_loss": 3.5809244512132423,
      "tokens_seen": 1224081408
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031773319959879637,
      "loss": 3.0189,
      "theoretical_loss": 3.5809068820694234,
      "tokens_seen": 1224146944
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003177231695085256,
      "loss": 2.8246,
      "theoretical_loss": 3.580889314129508,
      "tokens_seen": 1224212480
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031771313941825473,
      "loss": 2.8294,
      "theoretical_loss": 3.580871747393351,
      "tokens_seen": 1224278016
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031770310932798397,
      "loss": 2.8176,
      "theoretical_loss": 3.5808541818608046,
      "tokens_seen": 1224343552
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031769307923771315,
      "loss": 3.0475,
      "theoretical_loss": 3.5808366175317223,
      "tokens_seen": 1224409088
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031768304914744233,
      "loss": 2.942,
      "theoretical_loss": 3.580819054405956,
      "tokens_seen": 1224474624
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003176730190571715,
      "loss": 2.8286,
      "theoretical_loss": 3.580801492483361,
      "tokens_seen": 1224540160
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003176629889669007,
      "loss": 2.783,
      "theoretical_loss": 3.580783931763789,
      "tokens_seen": 1224605696
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031765295887662987,
      "loss": 2.8959,
      "theoretical_loss": 3.5807663722470933,
      "tokens_seen": 1224671232
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003176429287863591,
      "loss": 2.6541,
      "theoretical_loss": 3.5807488139331274,
      "tokens_seen": 1224736768
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031763289869608823,
      "loss": 2.884,
      "theoretical_loss": 3.5807312568217444,
      "tokens_seen": 1224802304
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031762286860581747,
      "loss": 2.9139,
      "theoretical_loss": 3.5807137009127983,
      "tokens_seen": 1224867840
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003176128385155466,
      "loss": 2.7903,
      "theoretical_loss": 3.5806961462061415,
      "tokens_seen": 1224933376
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031760280842527583,
      "loss": 3.0043,
      "theoretical_loss": 3.580678592701628,
      "tokens_seen": 1224998912
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.000317592778335005,
      "loss": 2.825,
      "theoretical_loss": 3.5806610403991104,
      "tokens_seen": 1225064448
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003175827482447342,
      "loss": 2.7579,
      "theoretical_loss": 3.580643489298443,
      "tokens_seen": 1225129984
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003175727181544634,
      "loss": 2.7714,
      "theoretical_loss": 3.5806259393994786,
      "tokens_seen": 1225195520
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003175626880641926,
      "loss": 2.7442,
      "theoretical_loss": 3.580608390702071,
      "tokens_seen": 1225261056
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003175526579739218,
      "loss": 3.0607,
      "theoretical_loss": 3.5805908432060733,
      "tokens_seen": 1225326592
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031754262788365097,
      "loss": 2.8833,
      "theoretical_loss": 3.5805732969113393,
      "tokens_seen": 1225392128
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031753259779338015,
      "loss": 2.9283,
      "theoretical_loss": 3.5805557518177227,
      "tokens_seen": 1225457664
    },
    {
      "epoch": 0.37,
      "objective/train/advantage_avg": 0.4949706494808197,
      "objective/train/docs_used": 695595,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.5177462100982666,
      "objective/train/original_loss": 2.5177462100982666,
      "objective/train/theoretical_loss": 3.580538207925077,
      "objective/train/tokens_used": 1245983200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2450331598520279,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050742745399475,
      "objective/train/weighted_lm_loss": 2.6454062461853027,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0432155132293701,
      "theoretical_loss": 3.580538207925077,
      "tokens_seen": 1225523200
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031752256770310933,
      "loss": 2.9374,
      "theoretical_loss": 3.580538207925077,
      "tokens_seen": 1225523200
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031751253761283857,
      "loss": 2.8876,
      "theoretical_loss": 3.580520665233255,
      "tokens_seen": 1225588736
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003175025075225677,
      "loss": 3.015,
      "theoretical_loss": 3.580503123742111,
      "tokens_seen": 1225654272
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031749247743229693,
      "loss": 2.7745,
      "theoretical_loss": 3.5804855834514986,
      "tokens_seen": 1225719808
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031748244734202606,
      "loss": 3.085,
      "theoretical_loss": 3.5804680443612718,
      "tokens_seen": 1225785344
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003174724172517553,
      "loss": 3.0285,
      "theoretical_loss": 3.580450506471283,
      "tokens_seen": 1225850880
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003174623871614845,
      "loss": 2.8652,
      "theoretical_loss": 3.5804329697813873,
      "tokens_seen": 1225916416
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031745235707121366,
      "loss": 2.7957,
      "theoretical_loss": 3.580415434291438,
      "tokens_seen": 1225981952
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031744232698094284,
      "loss": 2.8261,
      "theoretical_loss": 3.580397900001288,
      "tokens_seen": 1226047488
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.000317432296890672,
      "loss": 2.8472,
      "theoretical_loss": 3.5803803669107923,
      "tokens_seen": 1226113024
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003174222668004012,
      "loss": 2.7875,
      "theoretical_loss": 3.5803628350198045,
      "tokens_seen": 1226178560
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031741223671013044,
      "loss": 3.0036,
      "theoretical_loss": 3.580345304328178,
      "tokens_seen": 1226244096
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031740220661985956,
      "loss": 2.6998,
      "theoretical_loss": 3.5803277748357667,
      "tokens_seen": 1226309632
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003173921765295888,
      "loss": 2.8972,
      "theoretical_loss": 3.5803102465424246,
      "tokens_seen": 1226375168
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.000317382146439318,
      "loss": 2.9068,
      "theoretical_loss": 3.580292719448005,
      "tokens_seen": 1226440704
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031737211634904716,
      "loss": 2.974,
      "theoretical_loss": 3.580275193552364,
      "tokens_seen": 1226506240
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031736208625877634,
      "loss": 2.9042,
      "theoretical_loss": 3.5802576688553525,
      "tokens_seen": 1226571776
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003173520561685055,
      "loss": 2.8524,
      "theoretical_loss": 3.580240145356827,
      "tokens_seen": 1226637312
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003173420260782347,
      "loss": 2.8954,
      "theoretical_loss": 3.58022262305664,
      "tokens_seen": 1226702848
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031733199598796394,
      "loss": 2.7435,
      "theoretical_loss": 3.580205101954646,
      "tokens_seen": 1226768384
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031732196589769307,
      "loss": 2.8629,
      "theoretical_loss": 3.5801875820506988,
      "tokens_seen": 1226833920
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003173119358074223,
      "loss": 2.808,
      "theoretical_loss": 3.580170063344654,
      "tokens_seen": 1226899456
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031730190571715143,
      "loss": 2.7941,
      "theoretical_loss": 3.5801525458363637,
      "tokens_seen": 1226964992
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031729187562688066,
      "loss": 2.8675,
      "theoretical_loss": 3.5801350295256826,
      "tokens_seen": 1227030528
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031728184553660984,
      "loss": 2.9246,
      "theoretical_loss": 3.580117514412465,
      "tokens_seen": 1227096064
    },
    {
      "epoch": 0.37,
      "objective/train/advantage_avg": 0.49416568875312805,
      "objective/train/docs_used": 696474,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.2983903884887695,
      "objective/train/original_loss": 2.2983903884887695,
      "objective/train/theoretical_loss": 3.580100000496566,
      "objective/train/tokens_used": 1247621600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24429818987846375,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506583452224731,
      "objective/train/weighted_lm_loss": 2.414680242538452,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0168063640594482,
      "theoretical_loss": 3.580100000496566,
      "tokens_seen": 1227161600
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.000317271815446339,
      "loss": 2.6653,
      "theoretical_loss": 3.580100000496566,
      "tokens_seen": 1227161600
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003172617853560682,
      "loss": 2.7874,
      "theoretical_loss": 3.5800824877778386,
      "tokens_seen": 1227227136
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003172517552657974,
      "loss": 3.0675,
      "theoretical_loss": 3.5800649762561374,
      "tokens_seen": 1227292672
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031724172517552657,
      "loss": 2.973,
      "theoretical_loss": 3.580047465931317,
      "tokens_seen": 1227358208
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003172316950852558,
      "loss": 2.7393,
      "theoretical_loss": 3.580029956803231,
      "tokens_seen": 1227423744
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031722166499498493,
      "loss": 2.8243,
      "theoretical_loss": 3.5800124488717344,
      "tokens_seen": 1227489280
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031721163490471417,
      "loss": 3.0802,
      "theoretical_loss": 3.5799949421366817,
      "tokens_seen": 1227554816
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031720160481444335,
      "loss": 2.7422,
      "theoretical_loss": 3.579977436597926,
      "tokens_seen": 1227620352
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031719157472417253,
      "loss": 2.608,
      "theoretical_loss": 3.5799599322553233,
      "tokens_seen": 1227685888
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003171815446339017,
      "loss": 2.8758,
      "theoretical_loss": 3.579942429108727,
      "tokens_seen": 1227751424
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003171715145436309,
      "loss": 2.9529,
      "theoretical_loss": 3.5799249271579923,
      "tokens_seen": 1227816960
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031716148445336007,
      "loss": 3.012,
      "theoretical_loss": 3.579907426402972,
      "tokens_seen": 1227882496
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003171514543630893,
      "loss": 2.9987,
      "theoretical_loss": 3.579889926843523,
      "tokens_seen": 1227948032
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031714142427281843,
      "loss": 2.8249,
      "theoretical_loss": 3.579872428479498,
      "tokens_seen": 1228013568
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031713139418254767,
      "loss": 2.7289,
      "theoretical_loss": 3.579854931310752,
      "tokens_seen": 1228079104
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003171213640922768,
      "loss": 3.0134,
      "theoretical_loss": 3.57983743533714,
      "tokens_seen": 1228144640
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031711133400200603,
      "loss": 2.9153,
      "theoretical_loss": 3.579819940558517,
      "tokens_seen": 1228210176
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003171013039117352,
      "loss": 3.0391,
      "theoretical_loss": 3.5798024469747363,
      "tokens_seen": 1228275712
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003170912738214644,
      "loss": 3.0178,
      "theoretical_loss": 3.5797849545856533,
      "tokens_seen": 1228341248
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003170812437311936,
      "loss": 2.7611,
      "theoretical_loss": 3.5797674633911223,
      "tokens_seen": 1228406784
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003170712136409228,
      "loss": 2.8314,
      "theoretical_loss": 3.5797499733909985,
      "tokens_seen": 1228472320
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031706118355065194,
      "loss": 2.8043,
      "theoretical_loss": 3.5797324845851364,
      "tokens_seen": 1228537856
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003170511534603812,
      "loss": 2.8785,
      "theoretical_loss": 3.579714996973391,
      "tokens_seen": 1228603392
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003170411233701103,
      "loss": 2.9074,
      "theoretical_loss": 3.579697510555617,
      "tokens_seen": 1228668928
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031703109327983954,
      "loss": 2.6801,
      "theoretical_loss": 3.579680025331668,
      "tokens_seen": 1228734464
    },
    {
      "epoch": 0.37,
      "objective/train/advantage_avg": 0.4931412935256958,
      "objective/train/docs_used": 696846,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.43643856048584,
      "objective/train/original_loss": 2.436438798904419,
      "objective/train/theoretical_loss": 3.579662541301401,
      "objective/train/tokens_used": 1249260000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24326547980308533,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505506992340088,
      "objective/train/weighted_lm_loss": 2.5594592094421387,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0445822477340698,
      "theoretical_loss": 3.579662541301401,
      "tokens_seen": 1228800000
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003170210631895687,
      "loss": 2.707,
      "theoretical_loss": 3.579662541301401,
      "tokens_seen": 1228800000
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003170110330992979,
      "loss": 2.8265,
      "theoretical_loss": 3.579645058464669,
      "tokens_seen": 1228865536
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003170010030090271,
      "loss": 2.9369,
      "theoretical_loss": 3.579627576821328,
      "tokens_seen": 1228931072
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031699097291875626,
      "loss": 2.4609,
      "theoretical_loss": 3.5796100963712325,
      "tokens_seen": 1228996608
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031698094282848544,
      "loss": 2.8783,
      "theoretical_loss": 3.579592617114237,
      "tokens_seen": 1229062144
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003169709127382147,
      "loss": 2.8907,
      "theoretical_loss": 3.5795751390501973,
      "tokens_seen": 1229127680
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003169608826479438,
      "loss": 2.741,
      "theoretical_loss": 3.579557662178968,
      "tokens_seen": 1229193216
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031695085255767304,
      "loss": 2.7332,
      "theoretical_loss": 3.579540186500404,
      "tokens_seen": 1229258752
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031694082246740217,
      "loss": 3.0563,
      "theoretical_loss": 3.5795227120143607,
      "tokens_seen": 1229324288
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003169307923771314,
      "loss": 2.9981,
      "theoretical_loss": 3.5795052387206923,
      "tokens_seen": 1229389824
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003169207622868606,
      "loss": 2.9937,
      "theoretical_loss": 3.5794877666192546,
      "tokens_seen": 1229455360
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031691073219658976,
      "loss": 2.7325,
      "theoretical_loss": 3.579470295709903,
      "tokens_seen": 1229520896
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031690070210631894,
      "loss": 2.9015,
      "theoretical_loss": 3.579452825992492,
      "tokens_seen": 1229586432
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003168906720160482,
      "loss": 2.9157,
      "theoretical_loss": 3.5794353574668767,
      "tokens_seen": 1229651968
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003168806419257773,
      "loss": 2.844,
      "theoretical_loss": 3.5794178901329126,
      "tokens_seen": 1229717504
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031687061183550654,
      "loss": 2.9573,
      "theoretical_loss": 3.579400423990455,
      "tokens_seen": 1229783040
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031686058174523567,
      "loss": 2.9706,
      "theoretical_loss": 3.579382959039359,
      "tokens_seen": 1229848576
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003168505516549649,
      "loss": 3.1509,
      "theoretical_loss": 3.57936549527948,
      "tokens_seen": 1229914112
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003168405215646941,
      "loss": 3.1342,
      "theoretical_loss": 3.579348032710672,
      "tokens_seen": 1229979648
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031683049147442327,
      "loss": 2.6685,
      "theoretical_loss": 3.5793305713327923,
      "tokens_seen": 1230045184
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031682046138415245,
      "loss": 3.0102,
      "theoretical_loss": 3.5793131111456953,
      "tokens_seen": 1230110720
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031681043129388163,
      "loss": 2.703,
      "theoretical_loss": 3.579295652149237,
      "tokens_seen": 1230176256
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031680040120361086,
      "loss": 2.9182,
      "theoretical_loss": 3.579278194343271,
      "tokens_seen": 1230241792
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031679037111334004,
      "loss": 2.9758,
      "theoretical_loss": 3.5792607377276546,
      "tokens_seen": 1230307328
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003167803410230692,
      "loss": 2.7091,
      "theoretical_loss": 3.579243282302243,
      "tokens_seen": 1230372864
    },
    {
      "epoch": 0.37,
      "objective/train/advantage_avg": 0.49500948190689087,
      "objective/train/docs_used": 698095,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.9951109886169434,
      "objective/train/original_loss": 2.9951112270355225,
      "objective/train/theoretical_loss": 3.5792258280668903,
      "objective/train/tokens_used": 1250898400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24508962035179138,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507469177246094,
      "objective/train/weighted_lm_loss": 3.1470413208007812,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0445106029510498,
      "theoretical_loss": 3.5792258280668903,
      "tokens_seen": 1230438400
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003167703109327984,
      "loss": 3.052,
      "theoretical_loss": 3.5792258280668903,
      "tokens_seen": 1230438400
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003167602808425276,
      "loss": 3.0593,
      "theoretical_loss": 3.5792083750214534,
      "tokens_seen": 1230503936
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031675025075225677,
      "loss": 2.6287,
      "theoretical_loss": 3.579190923165787,
      "tokens_seen": 1230569472
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.000316740220661986,
      "loss": 3.0193,
      "theoretical_loss": 3.5791734724997473,
      "tokens_seen": 1230635008
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031673019057171513,
      "loss": 3.1375,
      "theoretical_loss": 3.5791560230231894,
      "tokens_seen": 1230700544
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031672016048144437,
      "loss": 2.7396,
      "theoretical_loss": 3.5791385747359685,
      "tokens_seen": 1230766080
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031671013039117355,
      "loss": 2.8771,
      "theoretical_loss": 3.5791211276379418,
      "tokens_seen": 1230831616
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031670010030090273,
      "loss": 2.7702,
      "theoretical_loss": 3.579103681728963,
      "tokens_seen": 1230897152
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003166900702106319,
      "loss": 2.7424,
      "theoretical_loss": 3.5790862370088887,
      "tokens_seen": 1230962688
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003166800401203611,
      "loss": 2.9012,
      "theoretical_loss": 3.5790687934775747,
      "tokens_seen": 1231028224
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031667001003009027,
      "loss": 3.0245,
      "theoretical_loss": 3.5790513511348765,
      "tokens_seen": 1231093760
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003166599799398195,
      "loss": 2.5261,
      "theoretical_loss": 3.57903390998065,
      "tokens_seen": 1231159296
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031664994984954863,
      "loss": 3.0377,
      "theoretical_loss": 3.57901647001475,
      "tokens_seen": 1231224832
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031663991975927787,
      "loss": 3.0632,
      "theoretical_loss": 3.578999031237034,
      "tokens_seen": 1231290368
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.000316629889669007,
      "loss": 2.7706,
      "theoretical_loss": 3.578981593647357,
      "tokens_seen": 1231355904
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031661985957873623,
      "loss": 2.874,
      "theoretical_loss": 3.578964157245574,
      "tokens_seen": 1231421440
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003166098294884654,
      "loss": 3.0145,
      "theoretical_loss": 3.578946722031542,
      "tokens_seen": 1231486976
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003165997993981946,
      "loss": 2.7195,
      "theoretical_loss": 3.5789292880051162,
      "tokens_seen": 1231552512
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003165897693079238,
      "loss": 3.032,
      "theoretical_loss": 3.578911855166153,
      "tokens_seen": 1231618048
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.000316579739217653,
      "loss": 3.0098,
      "theoretical_loss": 3.5788944235145084,
      "tokens_seen": 1231683584
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031656970912738214,
      "loss": 2.8667,
      "theoretical_loss": 3.578876993050038,
      "tokens_seen": 1231749120
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003165596790371114,
      "loss": 2.8871,
      "theoretical_loss": 3.578859563772598,
      "tokens_seen": 1231814656
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003165496489468405,
      "loss": 2.7388,
      "theoretical_loss": 3.578842135682044,
      "tokens_seen": 1231880192
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031653961885656974,
      "loss": 3.0028,
      "theoretical_loss": 3.5788247087782326,
      "tokens_seen": 1231945728
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003165295887662989,
      "loss": 2.986,
      "theoretical_loss": 3.57880728306102,
      "tokens_seen": 1232011264
    },
    {
      "epoch": 0.37,
      "objective/train/advantage_avg": 0.4927855134010315,
      "objective/train/docs_used": 698724,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.9631714820861816,
      "objective/train/original_loss": 2.9631714820861816,
      "objective/train/theoretical_loss": 3.5787898585302615,
      "objective/train/tokens_used": 1252536800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24293221533298492,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050513505935669,
      "objective/train/weighted_lm_loss": 3.112781524658203,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0425469875335693,
      "theoretical_loss": 3.5787898585302615,
      "tokens_seen": 1232076800
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003165195586760281,
      "loss": 3.0768,
      "theoretical_loss": 3.5787898585302615,
      "tokens_seen": 1232076800
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003165095285857573,
      "loss": 2.881,
      "theoretical_loss": 3.5787724351858134,
      "tokens_seen": 1232142336
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031649949849548646,
      "loss": 2.853,
      "theoretical_loss": 3.578755013027533,
      "tokens_seen": 1232207872
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031648946840521564,
      "loss": 3.1344,
      "theoretical_loss": 3.5787375920552744,
      "tokens_seen": 1232273408
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003164794383149449,
      "loss": 2.9871,
      "theoretical_loss": 3.5787201722688957,
      "tokens_seen": 1232338944
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.000316469408224674,
      "loss": 2.7844,
      "theoretical_loss": 3.578702753668252,
      "tokens_seen": 1232404480
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031645937813440324,
      "loss": 3.0308,
      "theoretical_loss": 3.5786853362532005,
      "tokens_seen": 1232470016
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031644934804413237,
      "loss": 3.0044,
      "theoretical_loss": 3.5786679200235962,
      "tokens_seen": 1232535552
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003164393179538616,
      "loss": 2.8445,
      "theoretical_loss": 3.5786505049792963,
      "tokens_seen": 1232601088
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003164292878635908,
      "loss": 2.979,
      "theoretical_loss": 3.5786330911201567,
      "tokens_seen": 1232666624
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031641925777331996,
      "loss": 3.0555,
      "theoretical_loss": 3.5786156784460346,
      "tokens_seen": 1232732160
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031640922768304914,
      "loss": 2.8898,
      "theoretical_loss": 3.578598266956785,
      "tokens_seen": 1232797696
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003163991975927784,
      "loss": 2.8073,
      "theoretical_loss": 3.578580856652265,
      "tokens_seen": 1232863232
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003163891675025075,
      "loss": 3.0679,
      "theoretical_loss": 3.578563447532331,
      "tokens_seen": 1232928768
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031637913741223674,
      "loss": 2.7607,
      "theoretical_loss": 3.57854603959684,
      "tokens_seen": 1232994304
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031636910732196587,
      "loss": 2.9354,
      "theoretical_loss": 3.5785286328456474,
      "tokens_seen": 1233059840
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003163590772316951,
      "loss": 3.0129,
      "theoretical_loss": 3.57851122727861,
      "tokens_seen": 1233125376
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003163490471414243,
      "loss": 2.8704,
      "theoretical_loss": 3.5784938228955845,
      "tokens_seen": 1233190912
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031633901705115347,
      "loss": 2.9852,
      "theoretical_loss": 3.578476419696428,
      "tokens_seen": 1233256448
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031632898696088265,
      "loss": 3.0821,
      "theoretical_loss": 3.5784590176809963,
      "tokens_seen": 1233321984
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031631895687061183,
      "loss": 3.0144,
      "theoretical_loss": 3.578441616849146,
      "tokens_seen": 1233387520
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.000316308926780341,
      "loss": 3.0362,
      "theoretical_loss": 3.5784242172007343,
      "tokens_seen": 1233453056
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031629889669007024,
      "loss": 2.5537,
      "theoretical_loss": 3.5784068187356173,
      "tokens_seen": 1233518592
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031628886659979937,
      "loss": 3.0827,
      "theoretical_loss": 3.5783894214536516,
      "tokens_seen": 1233584128
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003162788365095286,
      "loss": 2.8845,
      "theoretical_loss": 3.5783720253546942,
      "tokens_seen": 1233649664
    },
    {
      "epoch": 0.37,
      "objective/train/advantage_avg": 0.49400797486305237,
      "objective/train/docs_used": 699861,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.741137981414795,
      "objective/train/original_loss": 2.741137981414795,
      "objective/train/theoretical_loss": 3.578354630438602,
      "objective/train/tokens_used": 1254175200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2441171407699585,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050641655921936,
      "objective/train/weighted_lm_loss": 2.8797056674957275,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0451053380966187,
      "theoretical_loss": 3.578354630438602,
      "tokens_seen": 1233715200
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031626880641925773,
      "loss": 2.7094,
      "theoretical_loss": 3.578354630438602,
      "tokens_seen": 1233715200
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031625877632898697,
      "loss": 2.9328,
      "theoretical_loss": 3.578337236705231,
      "tokens_seen": 1233780736
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031624874623871615,
      "loss": 3.0026,
      "theoretical_loss": 3.5783198441544393,
      "tokens_seen": 1233846272
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031623871614844533,
      "loss": 2.9888,
      "theoretical_loss": 3.578302452786082,
      "tokens_seen": 1233911808
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003162286860581745,
      "loss": 2.8038,
      "theoretical_loss": 3.578285062600017,
      "tokens_seen": 1233977344
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031621865596790375,
      "loss": 2.7906,
      "theoretical_loss": 3.578267673596101,
      "tokens_seen": 1234042880
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003162086258776329,
      "loss": 2.9223,
      "theoretical_loss": 3.5782502857741907,
      "tokens_seen": 1234108416
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003161985957873621,
      "loss": 2.8989,
      "theoretical_loss": 3.578232899134143,
      "tokens_seen": 1234173952
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031618856569709124,
      "loss": 2.7786,
      "theoretical_loss": 3.578215513675815,
      "tokens_seen": 1234239488
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031617853560682047,
      "loss": 2.9894,
      "theoretical_loss": 3.5781981293990635,
      "tokens_seen": 1234305024
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031616850551654965,
      "loss": 2.6909,
      "theoretical_loss": 3.5781807463037456,
      "tokens_seen": 1234370560
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031615847542627883,
      "loss": 2.8555,
      "theoretical_loss": 3.5781633643897175,
      "tokens_seen": 1234436096
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.000316148445336008,
      "loss": 2.8773,
      "theoretical_loss": 3.578145983656838,
      "tokens_seen": 1234501632
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003161384152457372,
      "loss": 2.6194,
      "theoretical_loss": 3.578128604104962,
      "tokens_seen": 1234567168
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003161283851554664,
      "loss": 2.9047,
      "theoretical_loss": 3.578111225733948,
      "tokens_seen": 1234632704
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003161183550651956,
      "loss": 3.0084,
      "theoretical_loss": 3.5780938485436526,
      "tokens_seen": 1234698240
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031610832497492474,
      "loss": 2.9133,
      "theoretical_loss": 3.578076472533933,
      "tokens_seen": 1234763776
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.000316098294884654,
      "loss": 3.1893,
      "theoretical_loss": 3.5780590977046467,
      "tokens_seen": 1234829312
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003160882647943831,
      "loss": 2.6933,
      "theoretical_loss": 3.57804172405565,
      "tokens_seen": 1234894848
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031607823470411234,
      "loss": 2.7528,
      "theoretical_loss": 3.5780243515868007,
      "tokens_seen": 1234960384
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003160682046138415,
      "loss": 2.7564,
      "theoretical_loss": 3.578006980297956,
      "tokens_seen": 1235025920
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003160581745235707,
      "loss": 2.6433,
      "theoretical_loss": 3.5779896101889737,
      "tokens_seen": 1235091456
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031604814443329994,
      "loss": 2.9666,
      "theoretical_loss": 3.5779722412597095,
      "tokens_seen": 1235156992
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003160381143430291,
      "loss": 2.7389,
      "theoretical_loss": 3.5779548735100217,
      "tokens_seen": 1235222528
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003160280842527583,
      "loss": 2.8373,
      "theoretical_loss": 3.577937506939768,
      "tokens_seen": 1235288064
    },
    {
      "epoch": 0.37,
      "objective/train/advantage_avg": 0.49250221252441406,
      "objective/train/docs_used": 700550,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.7739105224609375,
      "objective/train/original_loss": 2.7739107608795166,
      "objective/train/theoretical_loss": 3.577920141548805,
      "objective/train/tokens_used": 1255813600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.242705836892128,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0504839420318604,
      "objective/train/weighted_lm_loss": 2.9140114784240723,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0363596677780151,
      "theoretical_loss": 3.577920141548805,
      "tokens_seen": 1235353600
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003160180541624875,
      "loss": 2.9766,
      "theoretical_loss": 3.577920141548805,
      "tokens_seen": 1235353600
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031600802407221666,
      "loss": 3.0316,
      "theoretical_loss": 3.5779027773369902,
      "tokens_seen": 1235419136
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031599799398194584,
      "loss": 2.8066,
      "theoretical_loss": 3.577885414304181,
      "tokens_seen": 1235484672
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003159879638916751,
      "loss": 2.9929,
      "theoretical_loss": 3.5778680524502353,
      "tokens_seen": 1235550208
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003159779338014042,
      "loss": 3.0699,
      "theoretical_loss": 3.5778506917750095,
      "tokens_seen": 1235615744
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031596790371113344,
      "loss": 2.6845,
      "theoretical_loss": 3.577833332278362,
      "tokens_seen": 1235681280
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031595787362086257,
      "loss": 2.8027,
      "theoretical_loss": 3.5778159739601505,
      "tokens_seen": 1235746816
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003159478435305918,
      "loss": 3.0578,
      "theoretical_loss": 3.577798616820232,
      "tokens_seen": 1235812352
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.000315937813440321,
      "loss": 2.7769,
      "theoretical_loss": 3.5777812608584636,
      "tokens_seen": 1235877888
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031592778335005016,
      "loss": 2.9866,
      "theoretical_loss": 3.5777639060747033,
      "tokens_seen": 1235943424
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031591775325977934,
      "loss": 3.1458,
      "theoretical_loss": 3.577746552468809,
      "tokens_seen": 1236008960
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003159077231695086,
      "loss": 2.7576,
      "theoretical_loss": 3.577729200040638,
      "tokens_seen": 1236074496
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003158976930792377,
      "loss": 2.6795,
      "theoretical_loss": 3.577711848790048,
      "tokens_seen": 1236140032
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031588766298896694,
      "loss": 2.5866,
      "theoretical_loss": 3.5776944987168964,
      "tokens_seen": 1236205568
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031587763289869607,
      "loss": 3.0277,
      "theoretical_loss": 3.5776771498210413,
      "tokens_seen": 1236271104
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003158676028084253,
      "loss": 2.8654,
      "theoretical_loss": 3.57765980210234,
      "tokens_seen": 1236336640
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003158575727181545,
      "loss": 2.7295,
      "theoretical_loss": 3.5776424555606505,
      "tokens_seen": 1236402176
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031584754262788367,
      "loss": 2.8823,
      "theoretical_loss": 3.5776251101958305,
      "tokens_seen": 1236467712
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031583751253761285,
      "loss": 2.7871,
      "theoretical_loss": 3.5776077660077377,
      "tokens_seen": 1236533248
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031582748244734203,
      "loss": 3.1079,
      "theoretical_loss": 3.5775904229962303,
      "tokens_seen": 1236598784
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003158174523570712,
      "loss": 2.752,
      "theoretical_loss": 3.577573081161166,
      "tokens_seen": 1236664320
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031580742226680044,
      "loss": 2.9252,
      "theoretical_loss": 3.577555740502402,
      "tokens_seen": 1236729856
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031579739217652957,
      "loss": 2.9634,
      "theoretical_loss": 3.5775384010197966,
      "tokens_seen": 1236795392
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003157873620862588,
      "loss": 2.9666,
      "theoretical_loss": 3.5775210627132084,
      "tokens_seen": 1236860928
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031577733199598793,
      "loss": 2.7009,
      "theoretical_loss": 3.5775037255824946,
      "tokens_seen": 1236926464
    },
    {
      "epoch": 0.37,
      "objective/train/advantage_avg": 0.49409258365631104,
      "objective/train/docs_used": 701918,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.4823427200317383,
      "objective/train/original_loss": 2.4823431968688965,
      "objective/train/theoretical_loss": 3.577486389627513,
      "objective/train/tokens_used": 1257452000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2441953718662262,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506505966186523,
      "objective/train/weighted_lm_loss": 2.607887029647827,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 3.577486389627513,
      "tokens_seen": 1236992000
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031576730190571717,
      "loss": 2.9163,
      "theoretical_loss": 3.577486389627513,
      "tokens_seen": 1236992000
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031575727181544635,
      "loss": 3.0122,
      "theoretical_loss": 3.577469054848122,
      "tokens_seen": 1237057536
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031574724172517553,
      "loss": 2.7976,
      "theoretical_loss": 3.577451721244179,
      "tokens_seen": 1237123072
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003157372116349047,
      "loss": 2.9784,
      "theoretical_loss": 3.577434388815543,
      "tokens_seen": 1237188608
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031572718154463395,
      "loss": 2.944,
      "theoretical_loss": 3.577417057562071,
      "tokens_seen": 1237254144
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003157171514543631,
      "loss": 2.7871,
      "theoretical_loss": 3.5773997274836224,
      "tokens_seen": 1237319680
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.0003157071213640923,
      "loss": 2.9312,
      "theoretical_loss": 3.5773823985800544,
      "tokens_seen": 1237385216
    },
    {
      "epoch": 0.37,
      "learning_rate": 0.00031569709127382144,
      "loss": 2.9877,
      "theoretical_loss": 3.5773650708512252,
      "tokens_seen": 1237450752
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031568706118355067,
      "loss": 2.5812,
      "theoretical_loss": 3.577347744296993,
      "tokens_seen": 1237516288
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031567703109327985,
      "loss": 2.9835,
      "theoretical_loss": 3.577330418917216,
      "tokens_seen": 1237581824
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031566700100300903,
      "loss": 2.9908,
      "theoretical_loss": 3.5773130947117524,
      "tokens_seen": 1237647360
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003156569709127382,
      "loss": 3.0607,
      "theoretical_loss": 3.5772957716804603,
      "tokens_seen": 1237712896
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003156469408224674,
      "loss": 2.731,
      "theoretical_loss": 3.5772784498231984,
      "tokens_seen": 1237778432
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003156369107321966,
      "loss": 3.0383,
      "theoretical_loss": 3.577261129139824,
      "tokens_seen": 1237843968
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003156268806419258,
      "loss": 3.0025,
      "theoretical_loss": 3.577243809630197,
      "tokens_seen": 1237909504
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031561685055165494,
      "loss": 2.8536,
      "theoretical_loss": 3.5772264912941747,
      "tokens_seen": 1237975040
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003156068204613842,
      "loss": 2.8516,
      "theoretical_loss": 3.577209174131615,
      "tokens_seen": 1238040576
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003155967903711133,
      "loss": 2.8515,
      "theoretical_loss": 3.577191858142377,
      "tokens_seen": 1238106112
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031558676028084254,
      "loss": 2.8674,
      "theoretical_loss": 3.5771745433263193,
      "tokens_seen": 1238171648
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003155767301905717,
      "loss": 2.7867,
      "theoretical_loss": 3.5771572296832996,
      "tokens_seen": 1238237184
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003155667001003009,
      "loss": 2.6921,
      "theoretical_loss": 3.577139917213177,
      "tokens_seen": 1238302720
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003155566700100301,
      "loss": 2.8245,
      "theoretical_loss": 3.577122605915809,
      "tokens_seen": 1238368256
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003155466399197593,
      "loss": 2.8076,
      "theoretical_loss": 3.577105295791056,
      "tokens_seen": 1238433792
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031553660982948844,
      "loss": 2.8826,
      "theoretical_loss": 3.5770879868387744,
      "tokens_seen": 1238499328
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003155265797392177,
      "loss": 3.0974,
      "theoretical_loss": 3.577070679058824,
      "tokens_seen": 1238564864
    },
    {
      "epoch": 0.38,
      "objective/train/advantage_avg": 0.49552109837532043,
      "objective/train/docs_used": 702757,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.8145103454589844,
      "objective/train/original_loss": 2.8145105838775635,
      "objective/train/theoretical_loss": 3.5770533724510627,
      "objective/train/tokens_used": 1259090400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24558697640895844,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0508005619049072,
      "objective/train/weighted_lm_loss": 2.957505226135254,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.044721245765686,
      "theoretical_loss": 3.5770533724510627,
      "tokens_seen": 1238630400
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003155165496489468,
      "loss": 2.9244,
      "theoretical_loss": 3.5770533724510627,
      "tokens_seen": 1238630400
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031550651955867604,
      "loss": 2.8744,
      "theoretical_loss": 3.5770360670153494,
      "tokens_seen": 1238695936
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003154964894684052,
      "loss": 2.7925,
      "theoretical_loss": 3.5770187627515435,
      "tokens_seen": 1238761472
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003154864593781344,
      "loss": 2.8751,
      "theoretical_loss": 3.577001459659502,
      "tokens_seen": 1238827008
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003154764292878636,
      "loss": 2.707,
      "theoretical_loss": 3.5769841577390853,
      "tokens_seen": 1238892544
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031546639919759277,
      "loss": 2.562,
      "theoretical_loss": 3.5769668569901505,
      "tokens_seen": 1238958080
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031545636910732195,
      "loss": 2.7209,
      "theoretical_loss": 3.576949557412558,
      "tokens_seen": 1239023616
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003154463390170512,
      "loss": 2.6063,
      "theoretical_loss": 3.5769322590061647,
      "tokens_seen": 1239089152
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003154363089267803,
      "loss": 3.0007,
      "theoretical_loss": 3.5769149617708305,
      "tokens_seen": 1239154688
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031542627883650954,
      "loss": 2.8203,
      "theoretical_loss": 3.576897665706414,
      "tokens_seen": 1239220224
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003154162487462387,
      "loss": 2.8544,
      "theoretical_loss": 3.5768803708127743,
      "tokens_seen": 1239285760
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003154062186559679,
      "loss": 2.9202,
      "theoretical_loss": 3.57686307708977,
      "tokens_seen": 1239351296
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003153961885656971,
      "loss": 2.9297,
      "theoretical_loss": 3.5768457845372597,
      "tokens_seen": 1239416832
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031538615847542627,
      "loss": 3.0151,
      "theoretical_loss": 3.576828493155103,
      "tokens_seen": 1239482368
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031537612838515545,
      "loss": 3.0225,
      "theoretical_loss": 3.5768112029431576,
      "tokens_seen": 1239547904
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003153660982948847,
      "loss": 2.8591,
      "theoretical_loss": 3.576793913901284,
      "tokens_seen": 1239613440
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003153560682046138,
      "loss": 3.0485,
      "theoretical_loss": 3.5767766260293397,
      "tokens_seen": 1239678976
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031534603811434305,
      "loss": 2.84,
      "theoretical_loss": 3.576759339327184,
      "tokens_seen": 1239744512
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003153360080240722,
      "loss": 2.896,
      "theoretical_loss": 3.576742053794677,
      "tokens_seen": 1239810048
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003153259779338014,
      "loss": 2.7239,
      "theoretical_loss": 3.576724769431677,
      "tokens_seen": 1239875584
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003153159478435306,
      "loss": 2.9013,
      "theoretical_loss": 3.5767074862380426,
      "tokens_seen": 1239941120
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031530591775325977,
      "loss": 3.0079,
      "theoretical_loss": 3.5766902042136337,
      "tokens_seen": 1240006656
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.000315295887662989,
      "loss": 2.8865,
      "theoretical_loss": 3.5766729233583083,
      "tokens_seen": 1240072192
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031528585757271813,
      "loss": 2.9596,
      "theoretical_loss": 3.576655643671927,
      "tokens_seen": 1240137728
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031527582748244737,
      "loss": 2.9891,
      "theoretical_loss": 3.5766383651543485,
      "tokens_seen": 1240203264
    },
    {
      "epoch": 0.38,
      "objective/train/advantage_avg": 0.49207472801208496,
      "objective/train/docs_used": 703958,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.734250545501709,
      "objective/train/original_loss": 2.734250783920288,
      "objective/train/theoretical_loss": 3.5766210878054308,
      "objective/train/tokens_used": 1260728800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24227336049079895,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0504390001296997,
      "objective/train/weighted_lm_loss": 2.872213125228882,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.042362093925476,
      "theoretical_loss": 3.5766210878054308,
      "tokens_seen": 1240268800
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031526579739217655,
      "loss": 2.8659,
      "theoretical_loss": 3.5766210878054308,
      "tokens_seen": 1240268800
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031525576730190573,
      "loss": 2.9072,
      "theoretical_loss": 3.5766038116250347,
      "tokens_seen": 1240334336
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003152457372116349,
      "loss": 2.8574,
      "theoretical_loss": 3.5765865366130187,
      "tokens_seen": 1240399872
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031523570712136415,
      "loss": 2.8177,
      "theoretical_loss": 3.576569262769242,
      "tokens_seen": 1240465408
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003152256770310933,
      "loss": 2.9813,
      "theoretical_loss": 3.5765519900935643,
      "tokens_seen": 1240530944
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003152156469408225,
      "loss": 2.8572,
      "theoretical_loss": 3.5765347185858443,
      "tokens_seen": 1240596480
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031520561685055164,
      "loss": 2.9083,
      "theoretical_loss": 3.5765174482459416,
      "tokens_seen": 1240662016
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031519558676028087,
      "loss": 3.0298,
      "theoretical_loss": 3.576500179073716,
      "tokens_seen": 1240727552
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031518555667001005,
      "loss": 2.7634,
      "theoretical_loss": 3.576482911069026,
      "tokens_seen": 1240793088
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031517552657973924,
      "loss": 2.5147,
      "theoretical_loss": 3.5764656442317317,
      "tokens_seen": 1240858624
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003151654964894684,
      "loss": 2.8987,
      "theoretical_loss": 3.5764483785616923,
      "tokens_seen": 1240924160
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003151554663991976,
      "loss": 2.8712,
      "theoretical_loss": 3.5764311140587672,
      "tokens_seen": 1240989696
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003151454363089268,
      "loss": 2.8148,
      "theoretical_loss": 3.5764138507228163,
      "tokens_seen": 1241055232
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.000315135406218656,
      "loss": 2.7396,
      "theoretical_loss": 3.5763965885536986,
      "tokens_seen": 1241120768
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031512537612838514,
      "loss": 3.0359,
      "theoretical_loss": 3.5763793275512734,
      "tokens_seen": 1241186304
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003151153460381144,
      "loss": 2.7869,
      "theoretical_loss": 3.576362067715401,
      "tokens_seen": 1241251840
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003151053159478435,
      "loss": 3.0779,
      "theoretical_loss": 3.576344809045941,
      "tokens_seen": 1241317376
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031509528585757274,
      "loss": 2.9723,
      "theoretical_loss": 3.576327551542752,
      "tokens_seen": 1241382912
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003150852557673019,
      "loss": 2.8499,
      "theoretical_loss": 3.5763102952056944,
      "tokens_seen": 1241448448
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003150752256770311,
      "loss": 2.9549,
      "theoretical_loss": 3.576293040034628,
      "tokens_seen": 1241513984
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003150651955867603,
      "loss": 2.7953,
      "theoretical_loss": 3.576275786029411,
      "tokens_seen": 1241579520
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003150551654964895,
      "loss": 2.9368,
      "theoretical_loss": 3.5762585331899057,
      "tokens_seen": 1241645056
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031504513540621864,
      "loss": 2.8777,
      "theoretical_loss": 3.5762412815159696,
      "tokens_seen": 1241710592
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003150351053159479,
      "loss": 2.8192,
      "theoretical_loss": 3.576224031007463,
      "tokens_seen": 1241776128
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.000315025075225677,
      "loss": 2.6748,
      "theoretical_loss": 3.576206781664246,
      "tokens_seen": 1241841664
    },
    {
      "epoch": 0.38,
      "objective/train/advantage_avg": 0.4945969581604004,
      "objective/train/docs_used": 704662,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.002991199493408,
      "objective/train/original_loss": 3.002991199493408,
      "objective/train/theoretical_loss": 3.576189533486179,
      "objective/train/tokens_used": 1262367200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24467813968658447,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050703525543213,
      "objective/train/weighted_lm_loss": 3.1551296710968018,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0425469875335693,
      "theoretical_loss": 3.576189533486179,
      "tokens_seen": 1241907200
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031501504513540624,
      "loss": 2.832,
      "theoretical_loss": 3.576189533486179,
      "tokens_seen": 1241907200
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003150050150451354,
      "loss": 2.6546,
      "theoretical_loss": 3.5761722864731205,
      "tokens_seen": 1241972736
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003149949849548646,
      "loss": 2.9141,
      "theoretical_loss": 3.576155040624931,
      "tokens_seen": 1242038272
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003149849548645938,
      "loss": 2.8091,
      "theoretical_loss": 3.5761377959414706,
      "tokens_seen": 1242103808
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031497492477432297,
      "loss": 2.9032,
      "theoretical_loss": 3.5761205524225987,
      "tokens_seen": 1242169344
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031496489468405215,
      "loss": 3.0312,
      "theoretical_loss": 3.5761033100681754,
      "tokens_seen": 1242234880
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003149548645937814,
      "loss": 2.7868,
      "theoretical_loss": 3.5760860688780607,
      "tokens_seen": 1242300416
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003149448345035105,
      "loss": 2.5831,
      "theoretical_loss": 3.5760688288521143,
      "tokens_seen": 1242365952
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031493480441323974,
      "loss": 2.917,
      "theoretical_loss": 3.5760515899901963,
      "tokens_seen": 1242431488
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003149247743229689,
      "loss": 2.8043,
      "theoretical_loss": 3.576034352292167,
      "tokens_seen": 1242497024
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003149147442326981,
      "loss": 2.7463,
      "theoretical_loss": 3.576017115757886,
      "tokens_seen": 1242562560
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003149047141424273,
      "loss": 2.6445,
      "theoretical_loss": 3.575999880387214,
      "tokens_seen": 1242628096
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031489468405215647,
      "loss": 2.9148,
      "theoretical_loss": 3.5759826461800106,
      "tokens_seen": 1242693632
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031488465396188565,
      "loss": 2.7937,
      "theoretical_loss": 3.575965413136136,
      "tokens_seen": 1242759168
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003148746238716149,
      "loss": 2.9323,
      "theoretical_loss": 3.5759481812554506,
      "tokens_seen": 1242824704
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.000314864593781344,
      "loss": 2.8223,
      "theoretical_loss": 3.5759309505378134,
      "tokens_seen": 1242890240
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031485456369107325,
      "loss": 2.7824,
      "theoretical_loss": 3.5759137209830865,
      "tokens_seen": 1242955776
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003148445336008024,
      "loss": 2.8943,
      "theoretical_loss": 3.5758964925911285,
      "tokens_seen": 1243021312
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003148345035105316,
      "loss": 3.0639,
      "theoretical_loss": 3.5758792653618006,
      "tokens_seen": 1243086848
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003148244734202608,
      "loss": 3.0388,
      "theoretical_loss": 3.5758620392949623,
      "tokens_seen": 1243152384
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031481444332998997,
      "loss": 2.7685,
      "theoretical_loss": 3.575844814390474,
      "tokens_seen": 1243217920
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031480441323971915,
      "loss": 2.8905,
      "theoretical_loss": 3.5758275906481964,
      "tokens_seen": 1243283456
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031479438314944833,
      "loss": 3.0854,
      "theoretical_loss": 3.57581036806799,
      "tokens_seen": 1243348992
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003147843530591775,
      "loss": 3.0153,
      "theoretical_loss": 3.5757931466497146,
      "tokens_seen": 1243414528
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031477432296890675,
      "loss": 3.0236,
      "theoretical_loss": 3.575775926393231,
      "tokens_seen": 1243480064
    },
    {
      "epoch": 0.38,
      "objective/train/advantage_avg": 0.4951116442680359,
      "objective/train/docs_used": 705833,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.6487057209014893,
      "objective/train/original_loss": 2.64870548248291,
      "objective/train/theoretical_loss": 3.5757587072983994,
      "objective/train/tokens_used": 1264005600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24519138038158417,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050757646560669,
      "objective/train/weighted_lm_loss": 2.7829086780548096,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0396549701690674,
      "theoretical_loss": 3.5757587072983994,
      "tokens_seen": 1243545600
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003147642928786359,
      "loss": 2.7831,
      "theoretical_loss": 3.5757587072983994,
      "tokens_seen": 1243545600
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003147542627883651,
      "loss": 2.8205,
      "theoretical_loss": 3.57574148936508,
      "tokens_seen": 1243611136
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003147442326980943,
      "loss": 2.9601,
      "theoretical_loss": 3.5757242725931335,
      "tokens_seen": 1243676672
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003147342026078235,
      "loss": 3.1302,
      "theoretical_loss": 3.5757070569824205,
      "tokens_seen": 1243742208
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031472417251755266,
      "loss": 2.8879,
      "theoretical_loss": 3.575689842532801,
      "tokens_seen": 1243807744
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031471414242728184,
      "loss": 2.9459,
      "theoretical_loss": 3.575672629244136,
      "tokens_seen": 1243873280
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.000314704112337011,
      "loss": 3.0693,
      "theoretical_loss": 3.5756554171162858,
      "tokens_seen": 1243938816
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031469408224674025,
      "loss": 2.9812,
      "theoretical_loss": 3.575638206149111,
      "tokens_seen": 1244004352
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003146840521564694,
      "loss": 2.6627,
      "theoretical_loss": 3.5756209963424728,
      "tokens_seen": 1244069888
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003146740220661986,
      "loss": 2.7896,
      "theoretical_loss": 3.575603787696231,
      "tokens_seen": 1244135424
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031466399197592774,
      "loss": 2.9569,
      "theoretical_loss": 3.575586580210246,
      "tokens_seen": 1244200960
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.000314653961885657,
      "loss": 3.1268,
      "theoretical_loss": 3.5755693738843792,
      "tokens_seen": 1244266496
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031464393179538616,
      "loss": 2.9892,
      "theoretical_loss": 3.5755521687184917,
      "tokens_seen": 1244332032
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031463390170511534,
      "loss": 3.1536,
      "theoretical_loss": 3.5755349647124426,
      "tokens_seen": 1244397568
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003146238716148445,
      "loss": 2.9271,
      "theoretical_loss": 3.5755177618660943,
      "tokens_seen": 1244463104
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003146138415245737,
      "loss": 2.9594,
      "theoretical_loss": 3.5755005601793064,
      "tokens_seen": 1244528640
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003146038114343029,
      "loss": 3.0722,
      "theoretical_loss": 3.5754833596519404,
      "tokens_seen": 1244594176
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003145937813440321,
      "loss": 2.8839,
      "theoretical_loss": 3.575466160283857,
      "tokens_seen": 1244659712
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031458375125376125,
      "loss": 3.0156,
      "theoretical_loss": 3.5754489620749164,
      "tokens_seen": 1244725248
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003145737211634905,
      "loss": 3.0199,
      "theoretical_loss": 3.5754317650249803,
      "tokens_seen": 1244790784
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003145636910732197,
      "loss": 2.897,
      "theoretical_loss": 3.575414569133909,
      "tokens_seen": 1244856320
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031455366098294884,
      "loss": 3.0174,
      "theoretical_loss": 3.5753973744015637,
      "tokens_seen": 1244921856
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003145436308926781,
      "loss": 2.9331,
      "theoretical_loss": 3.5753801808278056,
      "tokens_seen": 1244987392
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003145336008024072,
      "loss": 3.11,
      "theoretical_loss": 3.575362988412495,
      "tokens_seen": 1245052928
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031452357071213644,
      "loss": 2.8007,
      "theoretical_loss": 3.575345797155493,
      "tokens_seen": 1245118464
    },
    {
      "epoch": 0.38,
      "objective/train/advantage_avg": 0.4933479428291321,
      "objective/train/docs_used": 706471,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.6740663051605225,
      "objective/train/original_loss": 2.6740665435791016,
      "objective/train/theoretical_loss": 3.5753286070566617,
      "objective/train/tokens_used": 1265644000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2434854656457901,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505725145339966,
      "objective/train/weighted_lm_loss": 2.8094258308410645,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0304546356201172,
      "theoretical_loss": 3.5753286070566617,
      "tokens_seen": 1245184000
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003145135406218656,
      "loss": 2.9028,
      "theoretical_loss": 3.5753286070566617,
      "tokens_seen": 1245184000
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003145035105315948,
      "loss": 2.9011,
      "theoretical_loss": 3.5753114181158603,
      "tokens_seen": 1245249536
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.000314493480441324,
      "loss": 2.8531,
      "theoretical_loss": 3.575294230332951,
      "tokens_seen": 1245315072
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031448345035105317,
      "loss": 2.96,
      "theoretical_loss": 3.5752770437077945,
      "tokens_seen": 1245380608
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031447342026078235,
      "loss": 3.1403,
      "theoretical_loss": 3.575259858240252,
      "tokens_seen": 1245446144
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003144633901705116,
      "loss": 2.6495,
      "theoretical_loss": 3.575242673930185,
      "tokens_seen": 1245511680
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003144533600802407,
      "loss": 2.972,
      "theoretical_loss": 3.575225490777454,
      "tokens_seen": 1245577216
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031444332998996994,
      "loss": 2.8523,
      "theoretical_loss": 3.575208308781921,
      "tokens_seen": 1245642752
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003144332998996991,
      "loss": 2.9024,
      "theoretical_loss": 3.575191127943446,
      "tokens_seen": 1245708288
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003144232698094283,
      "loss": 3.0296,
      "theoretical_loss": 3.5751739482618907,
      "tokens_seen": 1245773824
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003144132397191575,
      "loss": 2.725,
      "theoretical_loss": 3.5751567697371174,
      "tokens_seen": 1245839360
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031440320962888667,
      "loss": 2.9739,
      "theoretical_loss": 3.575139592368986,
      "tokens_seen": 1245904896
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031439317953861585,
      "loss": 2.9631,
      "theoretical_loss": 3.575122416157358,
      "tokens_seen": 1245970432
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003143831494483451,
      "loss": 2.7078,
      "theoretical_loss": 3.5751052411020954,
      "tokens_seen": 1246035968
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003143731193580742,
      "loss": 2.8021,
      "theoretical_loss": 3.5750880672030583,
      "tokens_seen": 1246101504
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031436308926780345,
      "loss": 2.9477,
      "theoretical_loss": 3.5750708944601097,
      "tokens_seen": 1246167040
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003143530591775326,
      "loss": 2.8066,
      "theoretical_loss": 3.57505372287311,
      "tokens_seen": 1246232576
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003143430290872618,
      "loss": 3.0847,
      "theoretical_loss": 3.5750365524419205,
      "tokens_seen": 1246298112
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.000314332998996991,
      "loss": 2.8006,
      "theoretical_loss": 3.5750193831664028,
      "tokens_seen": 1246363648
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031432296890672017,
      "loss": 2.9591,
      "theoretical_loss": 3.575002215046419,
      "tokens_seen": 1246429184
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031431293881644935,
      "loss": 2.6508,
      "theoretical_loss": 3.5749850480818295,
      "tokens_seen": 1246494720
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031430290872617853,
      "loss": 2.8901,
      "theoretical_loss": 3.574967882272497,
      "tokens_seen": 1246560256
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003142928786359077,
      "loss": 3.0736,
      "theoretical_loss": 3.5749507176182815,
      "tokens_seen": 1246625792
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031428284854563695,
      "loss": 3.0077,
      "theoretical_loss": 3.5749335541190455,
      "tokens_seen": 1246691328
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003142728184553661,
      "loss": 2.9682,
      "theoretical_loss": 3.574916391774651,
      "tokens_seen": 1246756864
    },
    {
      "epoch": 0.38,
      "objective/train/advantage_avg": 0.4944814145565033,
      "objective/train/docs_used": 707914,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.566805601119995,
      "objective/train/original_loss": 2.566804885864258,
      "objective/train/theoretical_loss": 3.574899230584959,
      "objective/train/tokens_used": 1267282400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24456030130386353,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506912469863892,
      "objective/train/weighted_lm_loss": 2.696798086166382,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0432155132293701,
      "theoretical_loss": 3.574899230584959,
      "tokens_seen": 1246822400
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003142627883650953,
      "loss": 2.9425,
      "theoretical_loss": 3.574899230584959,
      "tokens_seen": 1246822400
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003142527582748245,
      "loss": 3.079,
      "theoretical_loss": 3.574882070549831,
      "tokens_seen": 1246887936
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003142427281845537,
      "loss": 2.8801,
      "theoretical_loss": 3.574864911669129,
      "tokens_seen": 1246953472
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031423269809428286,
      "loss": 2.982,
      "theoretical_loss": 3.574847753942714,
      "tokens_seen": 1247019008
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031422266800401204,
      "loss": 2.8122,
      "theoretical_loss": 3.5748305973704486,
      "tokens_seen": 1247084544
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003142126379137412,
      "loss": 2.9983,
      "theoretical_loss": 3.5748134419521946,
      "tokens_seen": 1247150080
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031420260782347045,
      "loss": 2.9828,
      "theoretical_loss": 3.5747962876878123,
      "tokens_seen": 1247215616
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003141925777331996,
      "loss": 2.6568,
      "theoretical_loss": 3.5747791345771653,
      "tokens_seen": 1247281152
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003141825476429288,
      "loss": 3.0608,
      "theoretical_loss": 3.5747619826201142,
      "tokens_seen": 1247346688
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031417251755265794,
      "loss": 2.8208,
      "theoretical_loss": 3.574744831816521,
      "tokens_seen": 1247412224
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003141624874623872,
      "loss": 2.9101,
      "theoretical_loss": 3.574727682166248,
      "tokens_seen": 1247477760
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031415245737211636,
      "loss": 2.8039,
      "theoretical_loss": 3.5747105336691565,
      "tokens_seen": 1247543296
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031414242728184554,
      "loss": 2.9258,
      "theoretical_loss": 3.5746933863251087,
      "tokens_seen": 1247608832
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003141323971915747,
      "loss": 2.8799,
      "theoretical_loss": 3.5746762401339662,
      "tokens_seen": 1247674368
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003141223671013039,
      "loss": 3.1096,
      "theoretical_loss": 3.5746590950955914,
      "tokens_seen": 1247739904
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003141123370110331,
      "loss": 2.8547,
      "theoretical_loss": 3.5746419512098457,
      "tokens_seen": 1247805440
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003141023069207623,
      "loss": 2.7317,
      "theoretical_loss": 3.5746248084765915,
      "tokens_seen": 1247870976
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031409227683049145,
      "loss": 2.7968,
      "theoretical_loss": 3.574607666895691,
      "tokens_seen": 1247936512
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003140822467402207,
      "loss": 2.8796,
      "theoretical_loss": 3.5745905264670057,
      "tokens_seen": 1248002048
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031407221664994986,
      "loss": 2.9125,
      "theoretical_loss": 3.574573387190398,
      "tokens_seen": 1248067584
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031406218655967904,
      "loss": 2.6593,
      "theoretical_loss": 3.57455624906573,
      "tokens_seen": 1248133120
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003140521564694082,
      "loss": 2.8751,
      "theoretical_loss": 3.574539112092863,
      "tokens_seen": 1248198656
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003140421263791374,
      "loss": 2.9764,
      "theoretical_loss": 3.5745219762716607,
      "tokens_seen": 1248264192
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003140320962888666,
      "loss": 2.9721,
      "theoretical_loss": 3.574504841601984,
      "tokens_seen": 1248329728
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003140220661985958,
      "loss": 2.6219,
      "theoretical_loss": 3.5744877080836948,
      "tokens_seen": 1248395264
    },
    {
      "epoch": 0.38,
      "objective/train/advantage_avg": 0.49538493156433105,
      "objective/train/docs_used": 708590,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.6781697273254395,
      "objective/train/original_loss": 2.6781697273254395,
      "objective/train/theoretical_loss": 3.5744705757166564,
      "objective/train/tokens_used": 1268920800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24547182023525238,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507863759994507,
      "objective/train/weighted_lm_loss": 2.8140461444854736,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 3.5744705757166564,
      "tokens_seen": 1248460800
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031401203610832495,
      "loss": 2.7489,
      "theoretical_loss": 3.5744705757166564,
      "tokens_seen": 1248460800
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003140020060180542,
      "loss": 2.9309,
      "theoretical_loss": 3.5744534445007305,
      "tokens_seen": 1248526336
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003139919759277833,
      "loss": 2.8515,
      "theoretical_loss": 3.574436314435779,
      "tokens_seen": 1248591872
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031398194583751255,
      "loss": 2.7471,
      "theoretical_loss": 3.5744191855216645,
      "tokens_seen": 1248657408
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031397191574724173,
      "loss": 2.7455,
      "theoretical_loss": 3.57440205775825,
      "tokens_seen": 1248722944
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003139618856569709,
      "loss": 2.7922,
      "theoretical_loss": 3.5743849311453966,
      "tokens_seen": 1248788480
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003139518555667001,
      "loss": 2.8256,
      "theoretical_loss": 3.574367805682967,
      "tokens_seen": 1248854016
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003139418254764293,
      "loss": 2.7964,
      "theoretical_loss": 3.5743506813708237,
      "tokens_seen": 1248919552
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031393179538615845,
      "loss": 2.987,
      "theoretical_loss": 3.5743335582088296,
      "tokens_seen": 1248985088
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003139217652958877,
      "loss": 2.8293,
      "theoretical_loss": 3.5743164361968462,
      "tokens_seen": 1249050624
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003139117352056168,
      "loss": 2.7675,
      "theoretical_loss": 3.574299315334736,
      "tokens_seen": 1249116160
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031390170511534605,
      "loss": 2.8401,
      "theoretical_loss": 3.5742821956223625,
      "tokens_seen": 1249181696
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031389167502507523,
      "loss": 2.7114,
      "theoretical_loss": 3.5742650770595867,
      "tokens_seen": 1249247232
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003138816449348044,
      "loss": 2.9537,
      "theoretical_loss": 3.5742479596462724,
      "tokens_seen": 1249312768
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003138716148445336,
      "loss": 2.5631,
      "theoretical_loss": 3.5742308433822814,
      "tokens_seen": 1249378304
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003138615847542628,
      "loss": 2.9095,
      "theoretical_loss": 3.574213728267477,
      "tokens_seen": 1249443840
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031385155466399196,
      "loss": 2.6227,
      "theoretical_loss": 3.5741966143017203,
      "tokens_seen": 1249509376
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003138415245737212,
      "loss": 2.8793,
      "theoretical_loss": 3.574179501484875,
      "tokens_seen": 1249574912
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003138314944834503,
      "loss": 2.8463,
      "theoretical_loss": 3.574162389816804,
      "tokens_seen": 1249640448
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031382146439317955,
      "loss": 2.745,
      "theoretical_loss": 3.574145279297369,
      "tokens_seen": 1249705984
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031381143430290873,
      "loss": 2.8956,
      "theoretical_loss": 3.574128169926433,
      "tokens_seen": 1249771520
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003138014042126379,
      "loss": 3.0791,
      "theoretical_loss": 3.574111061703859,
      "tokens_seen": 1249837056
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031379137412236715,
      "loss": 2.7593,
      "theoretical_loss": 3.57409395462951,
      "tokens_seen": 1249902592
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003137813440320963,
      "loss": 2.8053,
      "theoretical_loss": 3.574076848703248,
      "tokens_seen": 1249968128
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003137713139418255,
      "loss": 2.8792,
      "theoretical_loss": 3.574059743924936,
      "tokens_seen": 1250033664
    },
    {
      "epoch": 0.38,
      "objective/train/advantage_avg": 0.4931217432022095,
      "objective/train/docs_used": 709757,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.7611193656921387,
      "objective/train/original_loss": 2.7611191272735596,
      "objective/train/theoretical_loss": 3.5740426402944365,
      "objective/train/tokens_used": 1270559200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2432781159877777,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050548791885376,
      "objective/train/weighted_lm_loss": 2.9005632400512695,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 3.5740426402944365,
      "tokens_seen": 1250099200
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003137612838515547,
      "loss": 2.7106,
      "theoretical_loss": 3.5740426402944365,
      "tokens_seen": 1250099200
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003137512537612839,
      "loss": 3.0723,
      "theoretical_loss": 3.574025537811613,
      "tokens_seen": 1250164736
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031374122367101306,
      "loss": 2.8848,
      "theoretical_loss": 3.5740084364763276,
      "tokens_seen": 1250230272
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031373119358074224,
      "loss": 2.9039,
      "theoretical_loss": 3.5739913362884437,
      "tokens_seen": 1250295808
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003137211634904714,
      "loss": 2.7147,
      "theoretical_loss": 3.573974237247824,
      "tokens_seen": 1250361344
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031371113340020065,
      "loss": 3.0144,
      "theoretical_loss": 3.5739571393543317,
      "tokens_seen": 1250426880
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003137011033099298,
      "loss": 2.7618,
      "theoretical_loss": 3.5739400426078287,
      "tokens_seen": 1250492416
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.000313691073219659,
      "loss": 2.7866,
      "theoretical_loss": 3.573922947008179,
      "tokens_seen": 1250557952
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031368104312938814,
      "loss": 2.9571,
      "theoretical_loss": 3.5739058525552454,
      "tokens_seen": 1250623488
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003136710130391174,
      "loss": 2.8559,
      "theoretical_loss": 3.5738887592488906,
      "tokens_seen": 1250689024
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031366098294884656,
      "loss": 2.9415,
      "theoretical_loss": 3.5738716670889783,
      "tokens_seen": 1250754560
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031365095285857574,
      "loss": 2.7757,
      "theoretical_loss": 3.5738545760753704,
      "tokens_seen": 1250820096
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003136409227683049,
      "loss": 2.8994,
      "theoretical_loss": 3.573837486207931,
      "tokens_seen": 1250885632
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003136308926780341,
      "loss": 2.6161,
      "theoretical_loss": 3.5738203974865224,
      "tokens_seen": 1250951168
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003136208625877633,
      "loss": 2.9783,
      "theoretical_loss": 3.573803309911008,
      "tokens_seen": 1251016704
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003136108324974925,
      "loss": 2.8746,
      "theoretical_loss": 3.5737862234812514,
      "tokens_seen": 1251082240
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031360080240722165,
      "loss": 2.7726,
      "theoretical_loss": 3.5737691381971146,
      "tokens_seen": 1251147776
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003135907723169509,
      "loss": 2.7273,
      "theoretical_loss": 3.5737520540584624,
      "tokens_seen": 1251213312
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031358074222668006,
      "loss": 2.5165,
      "theoretical_loss": 3.573734971065157,
      "tokens_seen": 1251278848
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031357071213640924,
      "loss": 2.7252,
      "theoretical_loss": 3.573717889217062,
      "tokens_seen": 1251344384
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003135606820461384,
      "loss": 2.7005,
      "theoretical_loss": 3.57370080851404,
      "tokens_seen": 1251409920
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003135506519558676,
      "loss": 2.8611,
      "theoretical_loss": 3.573683728955955,
      "tokens_seen": 1251475456
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003135406218655968,
      "loss": 2.9479,
      "theoretical_loss": 3.5736666505426697,
      "tokens_seen": 1251540992
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.000313530591775326,
      "loss": 2.7882,
      "theoretical_loss": 3.5736495732740483,
      "tokens_seen": 1251606528
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031352056168505515,
      "loss": 2.734,
      "theoretical_loss": 3.573632497149953,
      "tokens_seen": 1251672064
    },
    {
      "epoch": 0.38,
      "objective/train/advantage_avg": 0.4951713979244232,
      "objective/train/docs_used": 710221,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.220667600631714,
      "objective/train/original_loss": 2.2206671237945557,
      "objective/train/theoretical_loss": 3.5736154221702483,
      "objective/train/tokens_used": 1272197600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2452508956193924,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507638454437256,
      "objective/train/weighted_lm_loss": 2.333355665206909,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 3.5736154221702483,
      "tokens_seen": 1251737600
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003135105315947844,
      "loss": 2.6277,
      "theoretical_loss": 3.5736154221702483,
      "tokens_seen": 1251737600
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003135005015045135,
      "loss": 2.9444,
      "theoretical_loss": 3.573598348334797,
      "tokens_seen": 1251803136
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031349047141424275,
      "loss": 2.8131,
      "theoretical_loss": 3.5735812756434626,
      "tokens_seen": 1251868672
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031348044132397193,
      "loss": 2.7824,
      "theoretical_loss": 3.5735642040961086,
      "tokens_seen": 1251934208
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003134704112337011,
      "loss": 2.873,
      "theoretical_loss": 3.5735471336925984,
      "tokens_seen": 1251999744
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003134603811434303,
      "loss": 2.882,
      "theoretical_loss": 3.5735300644327954,
      "tokens_seen": 1252065280
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003134503510531595,
      "loss": 3.0029,
      "theoretical_loss": 3.5735129963165635,
      "tokens_seen": 1252130816
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031344032096288865,
      "loss": 2.8839,
      "theoretical_loss": 3.573495929343766,
      "tokens_seen": 1252196352
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003134302908726179,
      "loss": 2.7223,
      "theoretical_loss": 3.573478863514266,
      "tokens_seen": 1252261888
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.000313420260782347,
      "loss": 2.8352,
      "theoretical_loss": 3.5734617988279274,
      "tokens_seen": 1252327424
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031341023069207625,
      "loss": 3.0027,
      "theoretical_loss": 3.5734447352846144,
      "tokens_seen": 1252392960
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031340020060180543,
      "loss": 2.7583,
      "theoretical_loss": 3.57342767288419,
      "tokens_seen": 1252458496
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003133901705115346,
      "loss": 2.8993,
      "theoretical_loss": 3.573410611626518,
      "tokens_seen": 1252524032
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003133801404212638,
      "loss": 2.9811,
      "theoretical_loss": 3.5733935515114617,
      "tokens_seen": 1252589568
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.000313370110330993,
      "loss": 2.8055,
      "theoretical_loss": 3.573376492538886,
      "tokens_seen": 1252655104
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031336008024072216,
      "loss": 2.6791,
      "theoretical_loss": 3.573359434708653,
      "tokens_seen": 1252720640
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003133500501504514,
      "loss": 2.7694,
      "theoretical_loss": 3.5733423780206275,
      "tokens_seen": 1252786176
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003133400200601805,
      "loss": 2.6889,
      "theoretical_loss": 3.5733253224746733,
      "tokens_seen": 1252851712
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031332998996990975,
      "loss": 2.7923,
      "theoretical_loss": 3.5733082680706536,
      "tokens_seen": 1252917248
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003133199598796389,
      "loss": 3.0411,
      "theoretical_loss": 3.5732912148084326,
      "tokens_seen": 1252982784
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003133099297893681,
      "loss": 2.7344,
      "theoretical_loss": 3.5732741626878743,
      "tokens_seen": 1253048320
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003132998996990973,
      "loss": 2.8688,
      "theoretical_loss": 3.573257111708842,
      "tokens_seen": 1253113856
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003132898696088265,
      "loss": 2.9301,
      "theoretical_loss": 3.5732400618712004,
      "tokens_seen": 1253179392
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031327983951855566,
      "loss": 3.1512,
      "theoretical_loss": 3.573223013174813,
      "tokens_seen": 1253244928
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003132698094282849,
      "loss": 2.9353,
      "theoretical_loss": 3.573205965619543,
      "tokens_seen": 1253310464
    },
    {
      "epoch": 0.38,
      "objective/train/advantage_avg": 0.49510034918785095,
      "objective/train/docs_used": 710893,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.52445650100708,
      "objective/train/original_loss": 2.52445650100708,
      "objective/train/theoretical_loss": 3.573188919205256,
      "objective/train/tokens_used": 1273836000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24518117308616638,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507564544677734,
      "objective/train/weighted_lm_loss": 2.652557611465454,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0338950157165527,
      "theoretical_loss": 3.573188919205256,
      "tokens_seen": 1253376000
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.000313259779338014,
      "loss": 2.9641,
      "theoretical_loss": 3.573188919205256,
      "tokens_seen": 1253376000
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031324974924774326,
      "loss": 3.0904,
      "theoretical_loss": 3.5731718739318143,
      "tokens_seen": 1253441536
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003132397191574724,
      "loss": 2.8201,
      "theoretical_loss": 3.5731548297990825,
      "tokens_seen": 1253507072
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003132296890672016,
      "loss": 2.4647,
      "theoretical_loss": 3.573137786806925,
      "tokens_seen": 1253572608
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003132196589769308,
      "loss": 2.8636,
      "theoretical_loss": 3.573120744955206,
      "tokens_seen": 1253638144
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031320962888666,
      "loss": 2.8838,
      "theoretical_loss": 3.5731037042437883,
      "tokens_seen": 1253703680
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031319959879638916,
      "loss": 2.7981,
      "theoretical_loss": 3.5730866646725374,
      "tokens_seen": 1253769216
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031318956870611834,
      "loss": 2.9229,
      "theoretical_loss": 3.573069626241317,
      "tokens_seen": 1253834752
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003131795386158475,
      "loss": 2.7144,
      "theoretical_loss": 3.573052588949991,
      "tokens_seen": 1253900288
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031316950852557676,
      "loss": 2.8886,
      "theoretical_loss": 3.573035552798424,
      "tokens_seen": 1253965824
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003131594784353059,
      "loss": 2.8057,
      "theoretical_loss": 3.5730185177864797,
      "tokens_seen": 1254031360
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003131494483450351,
      "loss": 2.9277,
      "theoretical_loss": 3.5730014839140223,
      "tokens_seen": 1254096896
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031313941825476425,
      "loss": 2.6729,
      "theoretical_loss": 3.5729844511809166,
      "tokens_seen": 1254162432
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003131293881644935,
      "loss": 2.6703,
      "theoretical_loss": 3.5729674195870262,
      "tokens_seen": 1254227968
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031311935807422267,
      "loss": 2.7554,
      "theoretical_loss": 3.572950389132216,
      "tokens_seen": 1254293504
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031310932798395185,
      "loss": 2.933,
      "theoretical_loss": 3.5729333598163495,
      "tokens_seen": 1254359040
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031309929789368103,
      "loss": 3.0623,
      "theoretical_loss": 3.5729163316392922,
      "tokens_seen": 1254424576
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031308926780341026,
      "loss": 2.8011,
      "theoretical_loss": 3.5728993046009077,
      "tokens_seen": 1254490112
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003130792377131394,
      "loss": 2.9564,
      "theoretical_loss": 3.57288227870106,
      "tokens_seen": 1254555648
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003130692076228686,
      "loss": 2.735,
      "theoretical_loss": 3.5728652539396144,
      "tokens_seen": 1254621184
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003130591775325978,
      "loss": 3.0662,
      "theoretical_loss": 3.572848230316435,
      "tokens_seen": 1254686720
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.000313049147442327,
      "loss": 2.7335,
      "theoretical_loss": 3.5728312078313857,
      "tokens_seen": 1254752256
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003130391173520562,
      "loss": 2.7106,
      "theoretical_loss": 3.5728141864843312,
      "tokens_seen": 1254817792
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031302908726178535,
      "loss": 3.0026,
      "theoretical_loss": 3.5727971662751363,
      "tokens_seen": 1254883328
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003130190571715146,
      "loss": 3.0534,
      "theoretical_loss": 3.572780147203666,
      "tokens_seen": 1254948864
    },
    {
      "epoch": 0.38,
      "objective/train/advantage_avg": 0.4953668713569641,
      "objective/train/docs_used": 712142,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.0440917015075684,
      "objective/train/original_loss": 2.0440914630889893,
      "objective/train/theoretical_loss": 3.5727631292697843,
      "objective/train/tokens_used": 1275474400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24545729160308838,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507844686508179,
      "objective/train/weighted_lm_loss": 2.147763967514038,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0420360565185547,
      "theoretical_loss": 3.5727631292697843,
      "tokens_seen": 1255014400
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003130090270812437,
      "loss": 2.7588,
      "theoretical_loss": 3.5727631292697843,
      "tokens_seen": 1255014400
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031299899699097295,
      "loss": 2.8435,
      "theoretical_loss": 3.5727461124733555,
      "tokens_seen": 1255079936
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031298896690070213,
      "loss": 2.8659,
      "theoretical_loss": 3.5727290968142444,
      "tokens_seen": 1255145472
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003129789368104313,
      "loss": 2.7326,
      "theoretical_loss": 3.5727120822923153,
      "tokens_seen": 1255211008
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003129689067201605,
      "loss": 2.9143,
      "theoretical_loss": 3.5726950689074335,
      "tokens_seen": 1255276544
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003129588766298897,
      "loss": 2.8519,
      "theoretical_loss": 3.572678056659463,
      "tokens_seen": 1255342080
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031294884653961885,
      "loss": 2.8936,
      "theoretical_loss": 3.5726610455482692,
      "tokens_seen": 1255407616
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003129388164493481,
      "loss": 2.8916,
      "theoretical_loss": 3.572644035573716,
      "tokens_seen": 1255473152
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003129287863590772,
      "loss": 2.894,
      "theoretical_loss": 3.5726270267356686,
      "tokens_seen": 1255538688
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031291875626880645,
      "loss": 2.8342,
      "theoretical_loss": 3.5726100190339922,
      "tokens_seen": 1255604224
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031290872617853563,
      "loss": 3.066,
      "theoretical_loss": 3.5725930124685505,
      "tokens_seen": 1255669760
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003128986960882648,
      "loss": 3.0359,
      "theoretical_loss": 3.572576007039209,
      "tokens_seen": 1255735296
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.000312888665997994,
      "loss": 2.7658,
      "theoretical_loss": 3.5725590027458325,
      "tokens_seen": 1255800832
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003128786359077232,
      "loss": 3.0551,
      "theoretical_loss": 3.572541999588285,
      "tokens_seen": 1255866368
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031286860581745236,
      "loss": 2.7406,
      "theoretical_loss": 3.5725249975664326,
      "tokens_seen": 1255931904
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003128585757271816,
      "loss": 2.5798,
      "theoretical_loss": 3.57250799668014,
      "tokens_seen": 1255997440
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003128485456369107,
      "loss": 2.9883,
      "theoretical_loss": 3.5724909969292713,
      "tokens_seen": 1256062976
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031283851554663995,
      "loss": 2.5961,
      "theoretical_loss": 3.572473998313692,
      "tokens_seen": 1256128512
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003128284854563691,
      "loss": 2.9322,
      "theoretical_loss": 3.572457000833267,
      "tokens_seen": 1256194048
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003128184553660983,
      "loss": 2.9936,
      "theoretical_loss": 3.572440004487861,
      "tokens_seen": 1256259584
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003128084252758275,
      "loss": 2.7102,
      "theoretical_loss": 3.5724230092773395,
      "tokens_seen": 1256325120
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003127983951855567,
      "loss": 2.8729,
      "theoretical_loss": 3.572406015201567,
      "tokens_seen": 1256390656
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031278836509528586,
      "loss": 2.6784,
      "theoretical_loss": 3.5723890222604093,
      "tokens_seen": 1256456192
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003127783350050151,
      "loss": 2.8013,
      "theoretical_loss": 3.5723720304537308,
      "tokens_seen": 1256521728
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003127683049147442,
      "loss": 2.9814,
      "theoretical_loss": 3.5723550397813963,
      "tokens_seen": 1256587264
    },
    {
      "epoch": 0.38,
      "objective/train/advantage_avg": 0.49356749653816223,
      "objective/train/docs_used": 712613,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.808419942855835,
      "objective/train/original_loss": 2.808420181274414,
      "objective/train/theoretical_loss": 3.572338050243272,
      "objective/train/tokens_used": 1277112800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24370355904102325,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505954027175903,
      "objective/train/weighted_lm_loss": 2.950486183166504,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.038975715637207,
      "theoretical_loss": 3.572338050243272,
      "tokens_seen": 1256652800
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031275827482447346,
      "loss": 2.7479,
      "theoretical_loss": 3.572338050243272,
      "tokens_seen": 1256652800
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003127482447342026,
      "loss": 3.0532,
      "theoretical_loss": 3.572321061839222,
      "tokens_seen": 1256718336
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003127382146439318,
      "loss": 3.053,
      "theoretical_loss": 3.5723040745691117,
      "tokens_seen": 1256783872
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.000312728184553661,
      "loss": 2.7939,
      "theoretical_loss": 3.572287088432807,
      "tokens_seen": 1256849408
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003127181544633902,
      "loss": 3.0084,
      "theoretical_loss": 3.5722701034301725,
      "tokens_seen": 1256914944
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031270812437311936,
      "loss": 2.866,
      "theoretical_loss": 3.572253119561073,
      "tokens_seen": 1256980480
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031269809428284854,
      "loss": 3.2134,
      "theoretical_loss": 3.572236136825375,
      "tokens_seen": 1257046016
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003126880641925777,
      "loss": 2.9007,
      "theoretical_loss": 3.5722191552229425,
      "tokens_seen": 1257111552
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031267803410230696,
      "loss": 2.993,
      "theoretical_loss": 3.5722021747536417,
      "tokens_seen": 1257177088
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003126680040120361,
      "loss": 2.7889,
      "theoretical_loss": 3.5721851954173376,
      "tokens_seen": 1257242624
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003126579739217653,
      "loss": 2.797,
      "theoretical_loss": 3.5721682172138953,
      "tokens_seen": 1257308160
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031264794383149445,
      "loss": 2.7099,
      "theoretical_loss": 3.5721512401431808,
      "tokens_seen": 1257373696
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003126379137412237,
      "loss": 2.7829,
      "theoretical_loss": 3.572134264205059,
      "tokens_seen": 1257439232
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031262788365095287,
      "loss": 3.0401,
      "theoretical_loss": 3.5721172893993955,
      "tokens_seen": 1257504768
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031261785356068205,
      "loss": 2.7799,
      "theoretical_loss": 3.5721003157260554,
      "tokens_seen": 1257570304
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031260782347041123,
      "loss": 2.9754,
      "theoretical_loss": 3.5720833431849046,
      "tokens_seen": 1257635840
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031259779338014046,
      "loss": 2.7708,
      "theoretical_loss": 3.5720663717758083,
      "tokens_seen": 1257701376
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003125877632898696,
      "loss": 2.7122,
      "theoretical_loss": 3.5720494014986324,
      "tokens_seen": 1257766912
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003125777331995988,
      "loss": 2.8507,
      "theoretical_loss": 3.5720324323532417,
      "tokens_seen": 1257832448
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031256770310932795,
      "loss": 2.8145,
      "theoretical_loss": 3.5720154643395023,
      "tokens_seen": 1257897984
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003125576730190572,
      "loss": 2.931,
      "theoretical_loss": 3.57199849745728,
      "tokens_seen": 1257963520
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031254764292878637,
      "loss": 2.8346,
      "theoretical_loss": 3.57198153170644,
      "tokens_seen": 1258029056
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031253761283851555,
      "loss": 2.8035,
      "theoretical_loss": 3.571964567086848,
      "tokens_seen": 1258094592
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031252758274824473,
      "loss": 2.8368,
      "theoretical_loss": 3.57194760359837,
      "tokens_seen": 1258160128
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003125175526579739,
      "loss": 2.6081,
      "theoretical_loss": 3.5719306412408707,
      "tokens_seen": 1258225664
    },
    {
      "epoch": 0.38,
      "objective/train/advantage_avg": 0.4941180348396301,
      "objective/train/docs_used": 713934,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.919320583343506,
      "objective/train/original_loss": 2.919320583343506,
      "objective/train/theoretical_loss": 3.571913680014217,
      "objective/train/tokens_used": 1278751200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24423937499523163,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050653338432312,
      "objective/train/weighted_lm_loss": 3.067134141921997,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0363596677780151,
      "theoretical_loss": 3.571913680014217,
      "tokens_seen": 1258291200
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003125075225677031,
      "loss": 2.9691,
      "theoretical_loss": 3.571913680014217,
      "tokens_seen": 1258291200
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031249749247743233,
      "loss": 2.988,
      "theoretical_loss": 3.5718967199182736,
      "tokens_seen": 1258356736
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031248746238716146,
      "loss": 2.721,
      "theoretical_loss": 3.571879760952907,
      "tokens_seen": 1258422272
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003124774322968907,
      "loss": 2.7927,
      "theoretical_loss": 3.5718628031179827,
      "tokens_seen": 1258487808
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003124674022066198,
      "loss": 2.9369,
      "theoretical_loss": 3.571845846413366,
      "tokens_seen": 1258553344
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031245737211634905,
      "loss": 3.0857,
      "theoretical_loss": 3.5718288908389235,
      "tokens_seen": 1258618880
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031244734202607823,
      "loss": 2.9611,
      "theoretical_loss": 3.571811936394521,
      "tokens_seen": 1258684416
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003124373119358074,
      "loss": 2.8128,
      "theoretical_loss": 3.5717949830800233,
      "tokens_seen": 1258749952
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003124272818455366,
      "loss": 3.0298,
      "theoretical_loss": 3.5717780308952976,
      "tokens_seen": 1258815488
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031241725175526583,
      "loss": 2.8077,
      "theoretical_loss": 3.5717610798402086,
      "tokens_seen": 1258881024
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031240722166499496,
      "loss": 2.7772,
      "theoretical_loss": 3.5717441299146238,
      "tokens_seen": 1258946560
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003123971915747242,
      "loss": 3.0587,
      "theoretical_loss": 3.5717271811184075,
      "tokens_seen": 1259012096
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003123871614844533,
      "loss": 2.7281,
      "theoretical_loss": 3.571710233451426,
      "tokens_seen": 1259077632
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031237713139418256,
      "loss": 2.8488,
      "theoretical_loss": 3.571693286913546,
      "tokens_seen": 1259143168
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031236710130391174,
      "loss": 2.8448,
      "theoretical_loss": 3.5716763415046335,
      "tokens_seen": 1259208704
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003123570712136409,
      "loss": 2.8621,
      "theoretical_loss": 3.571659397224554,
      "tokens_seen": 1259274240
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003123470411233701,
      "loss": 3.0268,
      "theoretical_loss": 3.5716424540731735,
      "tokens_seen": 1259339776
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003123370110330993,
      "loss": 2.9832,
      "theoretical_loss": 3.5716255120503586,
      "tokens_seen": 1259405312
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031232698094282846,
      "loss": 2.9062,
      "theoretical_loss": 3.5716085711559753,
      "tokens_seen": 1259470848
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003123169508525577,
      "loss": 2.8484,
      "theoretical_loss": 3.5715916313898894,
      "tokens_seen": 1259536384
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003123069207622869,
      "loss": 2.828,
      "theoretical_loss": 3.571574692751967,
      "tokens_seen": 1259601920
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031229689067201606,
      "loss": 2.8344,
      "theoretical_loss": 3.5715577552420745,
      "tokens_seen": 1259667456
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003122868605817453,
      "loss": 2.8887,
      "theoretical_loss": 3.571540818860078,
      "tokens_seen": 1259732992
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003122768304914744,
      "loss": 2.9658,
      "theoretical_loss": 3.5715238836058445,
      "tokens_seen": 1259798528
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031226680040120366,
      "loss": 3.123,
      "theoretical_loss": 3.5715069494792395,
      "tokens_seen": 1259864064
    },
    {
      "epoch": 0.38,
      "objective/train/advantage_avg": 0.4930480420589447,
      "objective/train/docs_used": 714588,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.8334994316101074,
      "objective/train/original_loss": 2.8334994316101074,
      "objective/train/theoretical_loss": 3.5714900164801286,
      "objective/train/tokens_used": 1280389600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24322102963924408,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505411624908447,
      "objective/train/weighted_lm_loss": 2.976574420928955,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0338950157165527,
      "theoretical_loss": 3.5714900164801286,
      "tokens_seen": 1259929600
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003122567703109328,
      "loss": 3.0636,
      "theoretical_loss": 3.5714900164801286,
      "tokens_seen": 1259929600
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.000312246740220662,
      "loss": 2.9053,
      "theoretical_loss": 3.571473084608379,
      "tokens_seen": 1259995136
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003122367101303912,
      "loss": 2.9058,
      "theoretical_loss": 3.571456153863857,
      "tokens_seen": 1260060672
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003122266800401204,
      "loss": 2.9542,
      "theoretical_loss": 3.571439224246429,
      "tokens_seen": 1260126208
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031221664994984956,
      "loss": 2.9607,
      "theoretical_loss": 3.5714222957559607,
      "tokens_seen": 1260191744
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031220661985957874,
      "loss": 2.9409,
      "theoretical_loss": 3.571405368392319,
      "tokens_seen": 1260257280
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003121965897693079,
      "loss": 2.7864,
      "theoretical_loss": 3.5713884421553703,
      "tokens_seen": 1260322816
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031218655967903716,
      "loss": 3.1097,
      "theoretical_loss": 3.571371517044981,
      "tokens_seen": 1260388352
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003121765295887663,
      "loss": 2.7814,
      "theoretical_loss": 3.571354593061017,
      "tokens_seen": 1260453888
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003121664994984955,
      "loss": 2.8528,
      "theoretical_loss": 3.571337670203346,
      "tokens_seen": 1260519424
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031215646940822465,
      "loss": 2.9672,
      "theoretical_loss": 3.571320748471833,
      "tokens_seen": 1260584960
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003121464393179539,
      "loss": 2.7523,
      "theoretical_loss": 3.5713038278663456,
      "tokens_seen": 1260650496
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031213640922768307,
      "loss": 2.8506,
      "theoretical_loss": 3.5712869083867496,
      "tokens_seen": 1260716032
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031212637913741225,
      "loss": 2.5679,
      "theoretical_loss": 3.5712699900329126,
      "tokens_seen": 1260781568
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031211634904714143,
      "loss": 2.9452,
      "theoretical_loss": 3.5712530728047005,
      "tokens_seen": 1260847104
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031210631895687066,
      "loss": 2.8258,
      "theoretical_loss": 3.5712361567019792,
      "tokens_seen": 1260912640
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003120962888665998,
      "loss": 2.6653,
      "theoretical_loss": 3.571219241724617,
      "tokens_seen": 1260978176
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.000312086258776329,
      "loss": 2.9884,
      "theoretical_loss": 3.571202327872479,
      "tokens_seen": 1261043712
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031207622868605815,
      "loss": 2.7076,
      "theoretical_loss": 3.5711854151454325,
      "tokens_seen": 1261109248
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003120661985957874,
      "loss": 2.9212,
      "theoretical_loss": 3.5711685035433445,
      "tokens_seen": 1261174784
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031205616850551657,
      "loss": 2.9366,
      "theoretical_loss": 3.571151593066081,
      "tokens_seen": 1261240320
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031204613841524575,
      "loss": 2.9766,
      "theoretical_loss": 3.5711346837135096,
      "tokens_seen": 1261305856
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031203610832497493,
      "loss": 3.0347,
      "theoretical_loss": 3.5711177754854964,
      "tokens_seen": 1261371392
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003120260782347041,
      "loss": 2.9163,
      "theoretical_loss": 3.571100868381909,
      "tokens_seen": 1261436928
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003120160481444333,
      "loss": 3.0784,
      "theoretical_loss": 3.571083962402613,
      "tokens_seen": 1261502464
    },
    {
      "epoch": 0.38,
      "objective/train/advantage_avg": 0.4942036271095276,
      "objective/train/docs_used": 715954,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.376552104949951,
      "objective/train/original_loss": 2.376551866531372,
      "objective/train/theoretical_loss": 3.5710670575474763,
      "objective/train/tokens_used": 1282028000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.244283065199852,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506621599197388,
      "objective/train/weighted_lm_loss": 2.4968507289886475,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0432155132293701,
      "theoretical_loss": 3.5710670575474763,
      "tokens_seen": 1261568000
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031200601805416253,
      "loss": 2.8443,
      "theoretical_loss": 3.5710670575474763,
      "tokens_seen": 1261568000
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031199598796389166,
      "loss": 2.925,
      "theoretical_loss": 3.5710501538163646,
      "tokens_seen": 1261633536
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003119859578736209,
      "loss": 3.1952,
      "theoretical_loss": 3.571033251209146,
      "tokens_seen": 1261699072
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031197592778335,
      "loss": 2.6833,
      "theoretical_loss": 3.571016349725687,
      "tokens_seen": 1261764608
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031196589769307925,
      "loss": 2.5491,
      "theoretical_loss": 3.5709994493658543,
      "tokens_seen": 1261830144
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031195586760280843,
      "loss": 2.9258,
      "theoretical_loss": 3.570982550129515,
      "tokens_seen": 1261895680
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003119458375125376,
      "loss": 3.0121,
      "theoretical_loss": 3.570965652016536,
      "tokens_seen": 1261961216
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003119358074222668,
      "loss": 2.956,
      "theoretical_loss": 3.5709487550267847,
      "tokens_seen": 1262026752
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031192577733199603,
      "loss": 3.0047,
      "theoretical_loss": 3.570931859160127,
      "tokens_seen": 1262092288
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031191574724172516,
      "loss": 3.0177,
      "theoretical_loss": 3.5709149644164313,
      "tokens_seen": 1262157824
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003119057171514544,
      "loss": 3.0387,
      "theoretical_loss": 3.5708980707955638,
      "tokens_seen": 1262223360
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003118956870611835,
      "loss": 2.772,
      "theoretical_loss": 3.570881178297392,
      "tokens_seen": 1262288896
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031188565697091276,
      "loss": 2.8451,
      "theoretical_loss": 3.5708642869217826,
      "tokens_seen": 1262354432
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031187562688064194,
      "loss": 2.8488,
      "theoretical_loss": 3.5708473966686034,
      "tokens_seen": 1262419968
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003118655967903711,
      "loss": 2.8169,
      "theoretical_loss": 3.5708305075377207,
      "tokens_seen": 1262485504
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003118555667001003,
      "loss": 2.8455,
      "theoretical_loss": 3.5708136195290026,
      "tokens_seen": 1262551040
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003118455366098295,
      "loss": 2.8855,
      "theoretical_loss": 3.5707967326423153,
      "tokens_seen": 1262616576
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031183550651955866,
      "loss": 3.0759,
      "theoretical_loss": 3.5707798468775267,
      "tokens_seen": 1262682112
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003118254764292879,
      "loss": 2.9268,
      "theoretical_loss": 3.570762962234504,
      "tokens_seen": 1262747648
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.000311815446339017,
      "loss": 3.0392,
      "theoretical_loss": 3.570746078713114,
      "tokens_seen": 1262813184
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031180541624874626,
      "loss": 2.5628,
      "theoretical_loss": 3.5707291963132244,
      "tokens_seen": 1262878720
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003117953861584754,
      "loss": 2.7707,
      "theoretical_loss": 3.5707123150347027,
      "tokens_seen": 1262944256
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003117853560682046,
      "loss": 2.8046,
      "theoretical_loss": 3.5706954348774156,
      "tokens_seen": 1263009792
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003117753259779338,
      "loss": 2.7808,
      "theoretical_loss": 3.5706785558412313,
      "tokens_seen": 1263075328
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.000311765295887663,
      "loss": 2.9944,
      "theoretical_loss": 3.5706616779260165,
      "tokens_seen": 1263140864
    },
    {
      "epoch": 0.38,
      "objective/train/advantage_avg": 0.495075523853302,
      "objective/train/docs_used": 716702,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.6087265014648438,
      "objective/train/original_loss": 2.6087265014648438,
      "objective/train/theoretical_loss": 3.570644801131639,
      "objective/train/tokens_used": 1283666400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24516506493091583,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507538318634033,
      "objective/train/weighted_lm_loss": 2.7411081790924072,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0417574644088745,
      "theoretical_loss": 3.570644801131639,
      "tokens_seen": 1263206400
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031175526579739217,
      "loss": 2.858,
      "theoretical_loss": 3.570644801131639,
      "tokens_seen": 1263206400
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003117452357071214,
      "loss": 2.9511,
      "theoretical_loss": 3.5706279254579654,
      "tokens_seen": 1263271936
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031173520561685053,
      "loss": 2.7253,
      "theoretical_loss": 3.570611050904864,
      "tokens_seen": 1263337472
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031172517552657976,
      "loss": 2.827,
      "theoretical_loss": 3.570594177472202,
      "tokens_seen": 1263403008
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003117151454363089,
      "loss": 2.6892,
      "theoretical_loss": 3.5705773051598477,
      "tokens_seen": 1263468544
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003117051153460381,
      "loss": 2.7773,
      "theoretical_loss": 3.5705604339676666,
      "tokens_seen": 1263534080
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003116950852557673,
      "loss": 2.9794,
      "theoretical_loss": 3.5705435638955283,
      "tokens_seen": 1263599616
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003116850551654965,
      "loss": 2.6657,
      "theoretical_loss": 3.5705266949432994,
      "tokens_seen": 1263665152
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031167502507522567,
      "loss": 2.8237,
      "theoretical_loss": 3.5705098271108477,
      "tokens_seen": 1263730688
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031166499498495485,
      "loss": 3.1222,
      "theoretical_loss": 3.57049296039804,
      "tokens_seen": 1263796224
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031165496489468403,
      "loss": 2.7564,
      "theoretical_loss": 3.5704760948047456,
      "tokens_seen": 1263861760
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031164493480441327,
      "loss": 2.9864,
      "theoretical_loss": 3.5704592303308305,
      "tokens_seen": 1263927296
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003116349047141424,
      "loss": 2.8872,
      "theoretical_loss": 3.570442366976163,
      "tokens_seen": 1263992832
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031162487462387163,
      "loss": 2.8592,
      "theoretical_loss": 3.5704255047406113,
      "tokens_seen": 1264058368
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003116148445336008,
      "loss": 2.7266,
      "theoretical_loss": 3.5704086436240425,
      "tokens_seen": 1264123904
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031160481444333,
      "loss": 2.7774,
      "theoretical_loss": 3.5703917836263246,
      "tokens_seen": 1264189440
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031159478435305917,
      "loss": 2.7166,
      "theoretical_loss": 3.570374924747325,
      "tokens_seen": 1264254976
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031158475426278835,
      "loss": 2.9302,
      "theoretical_loss": 3.5703580669869126,
      "tokens_seen": 1264320512
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031157472417251753,
      "loss": 3.0086,
      "theoretical_loss": 3.5703412103449534,
      "tokens_seen": 1264386048
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031156469408224677,
      "loss": 2.8372,
      "theoretical_loss": 3.5703243548213166,
      "tokens_seen": 1264451584
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031155466399197595,
      "loss": 2.7976,
      "theoretical_loss": 3.5703075004158693,
      "tokens_seen": 1264517120
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031154463390170513,
      "loss": 2.7856,
      "theoretical_loss": 3.57029064712848,
      "tokens_seen": 1264582656
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003115346038114343,
      "loss": 2.5871,
      "theoretical_loss": 3.570273794959016,
      "tokens_seen": 1264648192
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003115245737211635,
      "loss": 2.8109,
      "theoretical_loss": 3.5702569439073457,
      "tokens_seen": 1264713728
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031151454363089273,
      "loss": 2.7584,
      "theoretical_loss": 3.5702400939733367,
      "tokens_seen": 1264779264
    },
    {
      "epoch": 0.38,
      "objective/train/advantage_avg": 0.4946286082267761,
      "objective/train/docs_used": 717941,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.485410690307617,
      "objective/train/original_loss": 2.485410451889038,
      "objective/train/theoretical_loss": 3.570223245156858,
      "objective/train/tokens_used": 1285304800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24471932649612427,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507067441940308,
      "objective/train/weighted_lm_loss": 2.611320734024048,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0425469875335693,
      "theoretical_loss": 3.570223245156858,
      "tokens_seen": 1264844800
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031150451354062186,
      "loss": 2.7449,
      "theoretical_loss": 3.570223245156858,
      "tokens_seen": 1264844800
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003114944834503511,
      "loss": 2.8929,
      "theoretical_loss": 3.570206397457776,
      "tokens_seen": 1264910336
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003114844533600802,
      "loss": 2.8697,
      "theoretical_loss": 3.570189550875959,
      "tokens_seen": 1264975872
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031147442326980945,
      "loss": 2.7844,
      "theoretical_loss": 3.570172705411276,
      "tokens_seen": 1265041408
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031146439317953863,
      "loss": 2.9807,
      "theoretical_loss": 3.5701558610635944,
      "tokens_seen": 1265106944
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003114543630892678,
      "loss": 2.8701,
      "theoretical_loss": 3.5701390178327825,
      "tokens_seen": 1265172480
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.000311444332998997,
      "loss": 2.8957,
      "theoretical_loss": 3.5701221757187076,
      "tokens_seen": 1265238016
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031143430290872623,
      "loss": 2.99,
      "theoretical_loss": 3.570105334721239,
      "tokens_seen": 1265303552
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031142427281845536,
      "loss": 2.9665,
      "theoretical_loss": 3.5700884948402445,
      "tokens_seen": 1265369088
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003114142427281846,
      "loss": 2.6164,
      "theoretical_loss": 3.570071656075592,
      "tokens_seen": 1265434624
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003114042126379137,
      "loss": 3.0097,
      "theoretical_loss": 3.57005481842715,
      "tokens_seen": 1265500160
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031139418254764296,
      "loss": 2.8726,
      "theoretical_loss": 3.570037981894786,
      "tokens_seen": 1265565696
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031138415245737214,
      "loss": 2.6565,
      "theoretical_loss": 3.5700211464783687,
      "tokens_seen": 1265631232
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003113741223671013,
      "loss": 2.7891,
      "theoretical_loss": 3.570004312177767,
      "tokens_seen": 1265696768
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003113640922768305,
      "loss": 3.1325,
      "theoretical_loss": 3.5699874789928483,
      "tokens_seen": 1265762304
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003113540621865597,
      "loss": 2.8623,
      "theoretical_loss": 3.5699706469234807,
      "tokens_seen": 1265827840
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031134403209628886,
      "loss": 2.7535,
      "theoretical_loss": 3.5699538159695337,
      "tokens_seen": 1265893376
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003113340020060181,
      "loss": 2.7994,
      "theoretical_loss": 3.569936986130874,
      "tokens_seen": 1265958912
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003113239719157472,
      "loss": 2.9512,
      "theoretical_loss": 3.5699201574073713,
      "tokens_seen": 1266024448
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031131394182547646,
      "loss": 3.0598,
      "theoretical_loss": 3.569903329798894,
      "tokens_seen": 1266089984
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003113039117352056,
      "loss": 2.9079,
      "theoretical_loss": 3.5698865033053098,
      "tokens_seen": 1266155520
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003112938816449348,
      "loss": 2.9466,
      "theoretical_loss": 3.569869677926487,
      "tokens_seen": 1266221056
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.000311283851554664,
      "loss": 2.7371,
      "theoretical_loss": 3.569852853662295,
      "tokens_seen": 1266286592
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003112738214643932,
      "loss": 2.9644,
      "theoretical_loss": 3.569836030512602,
      "tokens_seen": 1266352128
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031126379137412237,
      "loss": 2.9386,
      "theoretical_loss": 3.5698192084772757,
      "tokens_seen": 1266417664
    },
    {
      "epoch": 0.38,
      "objective/train/advantage_avg": 0.4942265450954437,
      "objective/train/docs_used": 718481,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.6768369674682617,
      "objective/train/original_loss": 2.676837205886841,
      "objective/train/theoretical_loss": 3.5698023875561855,
      "objective/train/tokens_used": 1286943200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2443251609802246,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506646633148193,
      "objective/train/weighted_lm_loss": 2.812364339828491,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0428944826126099,
      "theoretical_loss": 3.5698023875561855,
      "tokens_seen": 1266483200
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003112537612838516,
      "loss": 2.7366,
      "theoretical_loss": 3.5698023875561855,
      "tokens_seen": 1266483200
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031124373119358073,
      "loss": 2.7618,
      "theoretical_loss": 3.569785567749199,
      "tokens_seen": 1266548736
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031123370110330996,
      "loss": 2.7569,
      "theoretical_loss": 3.5697687490561862,
      "tokens_seen": 1266614272
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003112236710130391,
      "loss": 2.8808,
      "theoretical_loss": 3.5697519314770148,
      "tokens_seen": 1266679808
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003112136409227683,
      "loss": 2.8109,
      "theoretical_loss": 3.569735115011553,
      "tokens_seen": 1266745344
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003112036108324975,
      "loss": 3.0129,
      "theoretical_loss": 3.5697182996596704,
      "tokens_seen": 1266810880
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003111935807422267,
      "loss": 2.6286,
      "theoretical_loss": 3.5697014854212346,
      "tokens_seen": 1266876416
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031118355065195587,
      "loss": 3.077,
      "theoretical_loss": 3.5696846722961153,
      "tokens_seen": 1266941952
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031117352056168505,
      "loss": 2.7719,
      "theoretical_loss": 3.5696678602841807,
      "tokens_seen": 1267007488
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031116349047141423,
      "loss": 2.9518,
      "theoretical_loss": 3.5696510493852998,
      "tokens_seen": 1267073024
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031115346038114347,
      "loss": 2.9884,
      "theoretical_loss": 3.5696342395993406,
      "tokens_seen": 1267138560
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003111434302908726,
      "loss": 2.9187,
      "theoretical_loss": 3.5696174309261726,
      "tokens_seen": 1267204096
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031113340020060183,
      "loss": 3.0597,
      "theoretical_loss": 3.5696006233656643,
      "tokens_seen": 1267269632
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.000311123370110331,
      "loss": 2.9926,
      "theoretical_loss": 3.569583816917685,
      "tokens_seen": 1267335168
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003111133400200602,
      "loss": 2.8151,
      "theoretical_loss": 3.5695670115821025,
      "tokens_seen": 1267400704
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031110330992978937,
      "loss": 2.6472,
      "theoretical_loss": 3.5695502073587866,
      "tokens_seen": 1267466240
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031109327983951855,
      "loss": 2.9906,
      "theoretical_loss": 3.569533404247606,
      "tokens_seen": 1267531776
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031108324974924773,
      "loss": 3.003,
      "theoretical_loss": 3.569516602248429,
      "tokens_seen": 1267597312
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031107321965897697,
      "loss": 2.9844,
      "theoretical_loss": 3.5694998013611254,
      "tokens_seen": 1267662848
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003110631895687061,
      "loss": 3.1427,
      "theoretical_loss": 3.5694830015855636,
      "tokens_seen": 1267728384
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031105315947843533,
      "loss": 2.9126,
      "theoretical_loss": 3.569466202921613,
      "tokens_seen": 1267793920
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031104312938816446,
      "loss": 3.0113,
      "theoretical_loss": 3.5694494053691423,
      "tokens_seen": 1267859456
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003110330992978937,
      "loss": 2.9259,
      "theoretical_loss": 3.56943260892802,
      "tokens_seen": 1267924992
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003110230692076229,
      "loss": 2.9162,
      "theoretical_loss": 3.569415813598116,
      "tokens_seen": 1267990528
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031101303911735206,
      "loss": 2.6831,
      "theoretical_loss": 3.569399019379299,
      "tokens_seen": 1268056064
    },
    {
      "epoch": 0.38,
      "objective/train/advantage_avg": 0.49419301748275757,
      "objective/train/docs_used": 719716,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.2960004806518555,
      "objective/train/original_loss": 2.2960007190704346,
      "objective/train/theoretical_loss": 3.569382226271438,
      "objective/train/tokens_used": 1288581600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24429266154766083,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506612062454224,
      "objective/train/weighted_lm_loss": 2.4120547771453857,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0430042743682861,
      "theoretical_loss": 3.569382226271438,
      "tokens_seen": 1268121600
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031100300902708124,
      "loss": 2.8731,
      "theoretical_loss": 3.569382226271438,
      "tokens_seen": 1268121600
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003109929789368104,
      "loss": 3.2188,
      "theoretical_loss": 3.569365434274402,
      "tokens_seen": 1268187136
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003109829488465396,
      "loss": 2.7619,
      "theoretical_loss": 3.569348643388061,
      "tokens_seen": 1268252672
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031097291875626883,
      "loss": 2.9689,
      "theoretical_loss": 3.5693318536122822,
      "tokens_seen": 1268318208
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031096288866599796,
      "loss": 2.9326,
      "theoretical_loss": 3.569315064946937,
      "tokens_seen": 1268383744
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003109528585757272,
      "loss": 2.7935,
      "theoretical_loss": 3.5692982773918933,
      "tokens_seen": 1268449280
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003109428284854564,
      "loss": 2.8607,
      "theoretical_loss": 3.569281490947021,
      "tokens_seen": 1268514816
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031093279839518556,
      "loss": 2.7909,
      "theoretical_loss": 3.5692647056121882,
      "tokens_seen": 1268580352
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031092276830491474,
      "loss": 3.1543,
      "theoretical_loss": 3.5692479213872654,
      "tokens_seen": 1268645888
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003109127382146439,
      "loss": 2.6878,
      "theoretical_loss": 3.569231138272121,
      "tokens_seen": 1268711424
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003109027081243731,
      "loss": 3.0724,
      "theoretical_loss": 3.569214356266625,
      "tokens_seen": 1268776960
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031089267803410234,
      "loss": 3.0642,
      "theoretical_loss": 3.569197575370646,
      "tokens_seen": 1268842496
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031088264794383147,
      "loss": 2.8979,
      "theoretical_loss": 3.569180795584055,
      "tokens_seen": 1268908032
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003108726178535607,
      "loss": 2.7358,
      "theoretical_loss": 3.569164016906719,
      "tokens_seen": 1268973568
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031086258776328983,
      "loss": 2.6792,
      "theoretical_loss": 3.5691472393385086,
      "tokens_seen": 1269039104
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031085255767301906,
      "loss": 2.8745,
      "theoretical_loss": 3.5691304628792935,
      "tokens_seen": 1269104640
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031084252758274824,
      "loss": 2.9367,
      "theoretical_loss": 3.569113687528942,
      "tokens_seen": 1269170176
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003108324974924774,
      "loss": 3.1219,
      "theoretical_loss": 3.569096913287325,
      "tokens_seen": 1269235712
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003108224674022066,
      "loss": 3.0804,
      "theoretical_loss": 3.569080140154311,
      "tokens_seen": 1269301248
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003108124373119358,
      "loss": 2.8366,
      "theoretical_loss": 3.56906336812977,
      "tokens_seen": 1269366784
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.000310802407221665,
      "loss": 2.9743,
      "theoretical_loss": 3.569046597213571,
      "tokens_seen": 1269432320
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003107923771313942,
      "loss": 2.8385,
      "theoretical_loss": 3.569029827405584,
      "tokens_seen": 1269497856
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003107823470411234,
      "loss": 2.902,
      "theoretical_loss": 3.5690130587056785,
      "tokens_seen": 1269563392
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031077231695085257,
      "loss": 2.9101,
      "theoretical_loss": 3.5689962911137236,
      "tokens_seen": 1269628928
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003107622868605818,
      "loss": 2.8642,
      "theoretical_loss": 3.56897952462959,
      "tokens_seen": 1269694464
    },
    {
      "epoch": 0.38,
      "objective/train/advantage_avg": 0.49321210384368896,
      "objective/train/docs_used": 720440,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.870412588119507,
      "objective/train/original_loss": 2.870412826538086,
      "objective/train/theoretical_loss": 3.5689627592531465,
      "objective/train/tokens_used": 1290220000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24336855113506317,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.05055832862854,
      "objective/train/weighted_lm_loss": 3.015561103820801,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0363596677780151,
      "theoretical_loss": 3.5689627592531465,
      "tokens_seen": 1269760000
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031075225677031093,
      "loss": 2.9513,
      "theoretical_loss": 3.5689627592531465,
      "tokens_seen": 1269760000
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031074222668004016,
      "loss": 2.8588,
      "theoretical_loss": 3.5689459949842623,
      "tokens_seen": 1269825536
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003107321965897693,
      "loss": 2.953,
      "theoretical_loss": 3.568929231822808,
      "tokens_seen": 1269891072
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003107221664994985,
      "loss": 2.7424,
      "theoretical_loss": 3.5689124697686534,
      "tokens_seen": 1269956608
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003107121364092277,
      "loss": 3.0192,
      "theoretical_loss": 3.568895708821667,
      "tokens_seen": 1270022144
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003107021063189569,
      "loss": 2.9633,
      "theoretical_loss": 3.5688789489817196,
      "tokens_seen": 1270087680
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031069207622868607,
      "loss": 2.6972,
      "theoretical_loss": 3.568862190248681,
      "tokens_seen": 1270153216
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031068204613841525,
      "loss": 2.906,
      "theoretical_loss": 3.568845432622421,
      "tokens_seen": 1270218752
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031067201604814443,
      "loss": 3.0168,
      "theoretical_loss": 3.5688286761028083,
      "tokens_seen": 1270284288
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031066198595787367,
      "loss": 2.838,
      "theoretical_loss": 3.568811920689714,
      "tokens_seen": 1270349824
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.0003106519558676028,
      "loss": 2.8193,
      "theoretical_loss": 3.5687951663830075,
      "tokens_seen": 1270415360
    },
    {
      "epoch": 0.38,
      "learning_rate": 0.00031064192577733203,
      "loss": 3.1984,
      "theoretical_loss": 3.5687784131825584,
      "tokens_seen": 1270480896
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003106318956870612,
      "loss": 2.5355,
      "theoretical_loss": 3.568761661088237,
      "tokens_seen": 1270546432
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003106218655967904,
      "loss": 2.6218,
      "theoretical_loss": 3.568744910099913,
      "tokens_seen": 1270611968
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00031061183550651957,
      "loss": 3.0687,
      "theoretical_loss": 3.5687281602174563,
      "tokens_seen": 1270677504
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00031060180541624875,
      "loss": 2.7501,
      "theoretical_loss": 3.568711411440737,
      "tokens_seen": 1270743040
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00031059177532597793,
      "loss": 2.8891,
      "theoretical_loss": 3.568694663769625,
      "tokens_seen": 1270808576
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00031058174523570717,
      "loss": 3.2266,
      "theoretical_loss": 3.5686779172039906,
      "tokens_seen": 1270874112
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003105717151454363,
      "loss": 2.8642,
      "theoretical_loss": 3.5686611717437033,
      "tokens_seen": 1270939648
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00031056168505516553,
      "loss": 2.8917,
      "theoretical_loss": 3.5686444273886337,
      "tokens_seen": 1271005184
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00031055165496489466,
      "loss": 2.719,
      "theoretical_loss": 3.5686276841386517,
      "tokens_seen": 1271070720
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003105416248746239,
      "loss": 2.8716,
      "theoretical_loss": 3.5686109419936267,
      "tokens_seen": 1271136256
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003105315947843531,
      "loss": 3.0157,
      "theoretical_loss": 3.56859420095343,
      "tokens_seen": 1271201792
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00031052156469408226,
      "loss": 2.8712,
      "theoretical_loss": 3.5685774610179313,
      "tokens_seen": 1271267328
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00031051153460381144,
      "loss": 2.9225,
      "theoretical_loss": 3.5685607221870006,
      "tokens_seen": 1271332864
    },
    {
      "epoch": 0.39,
      "objective/train/advantage_avg": 0.49406465888023376,
      "objective/train/docs_used": 721295,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.7073934078216553,
      "objective/train/original_loss": 2.7073934078216553,
      "objective/train/theoretical_loss": 3.568543984460508,
      "objective/train/tokens_used": 1291858400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24415379762649536,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506476163864136,
      "objective/train/weighted_lm_loss": 2.844487428665161,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 3.568543984460508,
      "tokens_seen": 1271398400
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003105015045135406,
      "loss": 2.9733,
      "theoretical_loss": 3.568543984460508,
      "tokens_seen": 1271398400
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003104914744232698,
      "loss": 2.9704,
      "theoretical_loss": 3.568527247838323,
      "tokens_seen": 1271463936
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00031048144433299904,
      "loss": 2.9909,
      "theoretical_loss": 3.5685105123203176,
      "tokens_seen": 1271529472
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00031047141424272816,
      "loss": 2.7514,
      "theoretical_loss": 3.568493777906361,
      "tokens_seen": 1271595008
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003104613841524574,
      "loss": 2.7869,
      "theoretical_loss": 3.568477044596323,
      "tokens_seen": 1271660544
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003104513540621866,
      "loss": 3.0026,
      "theoretical_loss": 3.568460312390075,
      "tokens_seen": 1271726080
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00031044132397191576,
      "loss": 2.8416,
      "theoretical_loss": 3.568443581287487,
      "tokens_seen": 1271791616
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00031043129388164494,
      "loss": 2.8591,
      "theoretical_loss": 3.5684268512884283,
      "tokens_seen": 1271857152
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003104212637913741,
      "loss": 2.884,
      "theoretical_loss": 3.5684101223927702,
      "tokens_seen": 1271922688
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003104112337011033,
      "loss": 2.8351,
      "theoretical_loss": 3.568393394600383,
      "tokens_seen": 1271988224
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00031040120361083254,
      "loss": 2.9023,
      "theoretical_loss": 3.5683766679111377,
      "tokens_seen": 1272053760
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00031039117352056167,
      "loss": 2.9785,
      "theoretical_loss": 3.5683599423249035,
      "tokens_seen": 1272119296
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003103811434302909,
      "loss": 2.8254,
      "theoretical_loss": 3.5683432178415515,
      "tokens_seen": 1272184832
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00031037111334002003,
      "loss": 2.9074,
      "theoretical_loss": 3.5683264944609516,
      "tokens_seen": 1272250368
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00031036108324974926,
      "loss": 2.8207,
      "theoretical_loss": 3.5683097721829755,
      "tokens_seen": 1272315904
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00031035105315947844,
      "loss": 3.0045,
      "theoretical_loss": 3.5682930510074926,
      "tokens_seen": 1272381440
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003103410230692076,
      "loss": 2.908,
      "theoretical_loss": 3.5682763309343737,
      "tokens_seen": 1272446976
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003103309929789368,
      "loss": 2.8531,
      "theoretical_loss": 3.56825961196349,
      "tokens_seen": 1272512512
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.000310320962888666,
      "loss": 2.9224,
      "theoretical_loss": 3.568242894094711,
      "tokens_seen": 1272578048
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00031031093279839517,
      "loss": 2.7947,
      "theoretical_loss": 3.568226177327908,
      "tokens_seen": 1272643584
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003103009027081244,
      "loss": 2.8427,
      "theoretical_loss": 3.568209461662951,
      "tokens_seen": 1272709120
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00031029087261785353,
      "loss": 3.1065,
      "theoretical_loss": 3.5681927470997117,
      "tokens_seen": 1272774656
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00031028084252758277,
      "loss": 2.8253,
      "theoretical_loss": 3.5681760336380606,
      "tokens_seen": 1272840192
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00031027081243731195,
      "loss": 2.8081,
      "theoretical_loss": 3.568159321277867,
      "tokens_seen": 1272905728
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00031026078234704113,
      "loss": 3.0956,
      "theoretical_loss": 3.568142610019003,
      "tokens_seen": 1272971264
    },
    {
      "epoch": 0.39,
      "objective/train/advantage_avg": 0.49295222759246826,
      "objective/train/docs_used": 722001,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.9332616329193115,
      "objective/train/original_loss": 2.9332613945007324,
      "objective/train/theoretical_loss": 3.5681258998613385,
      "objective/train/tokens_used": 1293496800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24313989281654358,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505311489105225,
      "objective/train/weighted_lm_loss": 3.081697702407837,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0363596677780151,
      "theoretical_loss": 3.5681258998613385,
      "tokens_seen": 1273036800
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003102507522567703,
      "loss": 2.8027,
      "theoretical_loss": 3.5681258998613385,
      "tokens_seen": 1273036800
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003102407221664995,
      "loss": 2.9918,
      "theoretical_loss": 3.568109190804745,
      "tokens_seen": 1273102336
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00031023069207622867,
      "loss": 2.9142,
      "theoretical_loss": 3.568092482849093,
      "tokens_seen": 1273167872
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003102206619859579,
      "loss": 2.9534,
      "theoretical_loss": 3.568075775994253,
      "tokens_seen": 1273233408
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00031021063189568703,
      "loss": 2.8779,
      "theoretical_loss": 3.5680590702400963,
      "tokens_seen": 1273298944
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00031020060180541627,
      "loss": 2.7205,
      "theoretical_loss": 3.568042365586493,
      "tokens_seen": 1273364480
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003101905717151454,
      "loss": 2.8457,
      "theoretical_loss": 3.568025662033315,
      "tokens_seen": 1273430016
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00031018054162487463,
      "loss": 3.0504,
      "theoretical_loss": 3.568008959580432,
      "tokens_seen": 1273495552
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003101705115346038,
      "loss": 2.8411,
      "theoretical_loss": 3.567992258227716,
      "tokens_seen": 1273561088
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.000310160481444333,
      "loss": 3.1603,
      "theoretical_loss": 3.567975557975037,
      "tokens_seen": 1273626624
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003101504513540622,
      "loss": 2.781,
      "theoretical_loss": 3.5679588588222666,
      "tokens_seen": 1273692160
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003101404212637914,
      "loss": 2.7521,
      "theoretical_loss": 3.5679421607692756,
      "tokens_seen": 1273757696
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00031013039117352054,
      "loss": 2.8749,
      "theoretical_loss": 3.567925463815935,
      "tokens_seen": 1273823232
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00031012036108324977,
      "loss": 2.8843,
      "theoretical_loss": 3.5679087679621153,
      "tokens_seen": 1273888768
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003101103309929789,
      "loss": 2.6153,
      "theoretical_loss": 3.5678920732076884,
      "tokens_seen": 1273954304
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00031010030090270813,
      "loss": 2.8753,
      "theoretical_loss": 3.567875379552525,
      "tokens_seen": 1274019840
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003100902708124373,
      "loss": 2.8538,
      "theoretical_loss": 3.567858686996496,
      "tokens_seen": 1274085376
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003100802407221665,
      "loss": 2.8827,
      "theoretical_loss": 3.5678419955394727,
      "tokens_seen": 1274150912
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003100702106318957,
      "loss": 3.129,
      "theoretical_loss": 3.567825305181326,
      "tokens_seen": 1274216448
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00031006018054162486,
      "loss": 3.2265,
      "theoretical_loss": 3.567808615921927,
      "tokens_seen": 1274281984
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003100501504513541,
      "loss": 2.9649,
      "theoretical_loss": 3.5677919277611476,
      "tokens_seen": 1274347520
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003100401203610833,
      "loss": 2.9267,
      "theoretical_loss": 3.5677752406988583,
      "tokens_seen": 1274413056
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00031003009027081246,
      "loss": 2.974,
      "theoretical_loss": 3.5677585547349304,
      "tokens_seen": 1274478592
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00031002006018054164,
      "loss": 2.9709,
      "theoretical_loss": 3.5677418698692347,
      "tokens_seen": 1274544128
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003100100300902708,
      "loss": 2.7151,
      "theoretical_loss": 3.5677251861016437,
      "tokens_seen": 1274609664
    },
    {
      "epoch": 0.39,
      "objective/train/advantage_avg": 0.49394381046295166,
      "objective/train/docs_used": 723183,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.746912717819214,
      "objective/train/original_loss": 2.746912717819214,
      "objective/train/theoretical_loss": 3.5677085034320273,
      "objective/train/tokens_used": 1295135200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24406728148460388,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506349802017212,
      "objective/train/weighted_lm_loss": 2.886054277420044,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 3.5677085034320273,
      "tokens_seen": 1274675200
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00031,
      "loss": 2.9488,
      "theoretical_loss": 3.5677085034320273,
      "tokens_seen": 1274675200
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030998996990972924,
      "loss": 2.9455,
      "theoretical_loss": 3.5676918218602576,
      "tokens_seen": 1274740736
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030997993981945836,
      "loss": 2.8454,
      "theoretical_loss": 3.567675141386206,
      "tokens_seen": 1274806272
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003099699097291876,
      "loss": 2.9025,
      "theoretical_loss": 3.567658462009743,
      "tokens_seen": 1274871808
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003099598796389168,
      "loss": 2.9988,
      "theoretical_loss": 3.567641783730741,
      "tokens_seen": 1274937344
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030994984954864596,
      "loss": 3.0028,
      "theoretical_loss": 3.5676251065490705,
      "tokens_seen": 1275002880
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030993981945837514,
      "loss": 3.0755,
      "theoretical_loss": 3.567608430464604,
      "tokens_seen": 1275068416
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003099297893681043,
      "loss": 3.1225,
      "theoretical_loss": 3.5675917554772116,
      "tokens_seen": 1275133952
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003099197592778335,
      "loss": 2.8731,
      "theoretical_loss": 3.5675750815867655,
      "tokens_seen": 1275199488
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030990972918756274,
      "loss": 3.0105,
      "theoretical_loss": 3.567558408793137,
      "tokens_seen": 1275265024
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030989969909729187,
      "loss": 3.0424,
      "theoretical_loss": 3.5675417370961977,
      "tokens_seen": 1275330560
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003098896690070211,
      "loss": 2.8162,
      "theoretical_loss": 3.5675250664958194,
      "tokens_seen": 1275396096
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030987963891675023,
      "loss": 2.9497,
      "theoretical_loss": 3.567508396991873,
      "tokens_seen": 1275461632
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030986960882647946,
      "loss": 2.9064,
      "theoretical_loss": 3.5674917285842302,
      "tokens_seen": 1275527168
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030985957873620864,
      "loss": 2.7193,
      "theoretical_loss": 3.567475061272763,
      "tokens_seen": 1275592704
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003098495486459378,
      "loss": 2.8189,
      "theoretical_loss": 3.5674583950573426,
      "tokens_seen": 1275658240
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.000309839518555667,
      "loss": 3.0178,
      "theoretical_loss": 3.5674417299378405,
      "tokens_seen": 1275723776
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003098294884653962,
      "loss": 2.9335,
      "theoretical_loss": 3.567425065914129,
      "tokens_seen": 1275789312
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030981945837512537,
      "loss": 3.0519,
      "theoretical_loss": 3.567408402986079,
      "tokens_seen": 1275854848
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003098094282848546,
      "loss": 2.784,
      "theoretical_loss": 3.5673917411535627,
      "tokens_seen": 1275920384
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030979939819458373,
      "loss": 3.0043,
      "theoretical_loss": 3.5673750804164515,
      "tokens_seen": 1275985920
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030978936810431297,
      "loss": 2.9188,
      "theoretical_loss": 3.5673584207746174,
      "tokens_seen": 1276051456
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030977933801404215,
      "loss": 2.8733,
      "theoretical_loss": 3.567341762227932,
      "tokens_seen": 1276116992
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030976930792377133,
      "loss": 2.8359,
      "theoretical_loss": 3.567325104776267,
      "tokens_seen": 1276182528
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003097592778335005,
      "loss": 2.8365,
      "theoretical_loss": 3.5673084484194946,
      "tokens_seen": 1276248064
    },
    {
      "epoch": 0.39,
      "objective/train/advantage_avg": 0.4931427538394928,
      "objective/train/docs_used": 723876,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.5729117393493652,
      "objective/train/original_loss": 2.5729119777679443,
      "objective/train/theoretical_loss": 3.567291793157486,
      "objective/train/tokens_used": 1296773600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24329133331775665,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050550937652588,
      "objective/train/weighted_lm_loss": 2.702876329421997,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0168063640594482,
      "theoretical_loss": 3.567291793157486,
      "tokens_seen": 1276313600
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003097492477432297,
      "loss": 2.9249,
      "theoretical_loss": 3.567291793157486,
      "tokens_seen": 1276313600
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030973921765295887,
      "loss": 2.8776,
      "theoretical_loss": 3.567275138990113,
      "tokens_seen": 1276379136
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003097291875626881,
      "loss": 2.8301,
      "theoretical_loss": 3.5672584859172485,
      "tokens_seen": 1276444672
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030971915747241723,
      "loss": 2.8542,
      "theoretical_loss": 3.5672418339387635,
      "tokens_seen": 1276510208
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030970912738214647,
      "loss": 2.6809,
      "theoretical_loss": 3.56722518305453,
      "tokens_seen": 1276575744
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003096990972918756,
      "loss": 2.86,
      "theoretical_loss": 3.5672085332644197,
      "tokens_seen": 1276641280
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030968906720160483,
      "loss": 3.0101,
      "theoretical_loss": 3.567191884568305,
      "tokens_seen": 1276706816
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.000309679037111334,
      "loss": 2.6132,
      "theoretical_loss": 3.567175236966058,
      "tokens_seen": 1276772352
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003096690070210632,
      "loss": 2.8621,
      "theoretical_loss": 3.5671585904575505,
      "tokens_seen": 1276837888
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003096589769307924,
      "loss": 2.8175,
      "theoretical_loss": 3.5671419450426534,
      "tokens_seen": 1276903424
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003096489468405216,
      "loss": 2.7211,
      "theoretical_loss": 3.5671253007212407,
      "tokens_seen": 1276968960
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030963891675025074,
      "loss": 3.2215,
      "theoretical_loss": 3.567108657493183,
      "tokens_seen": 1277034496
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030962888665997997,
      "loss": 2.9728,
      "theoretical_loss": 3.5670920153583534,
      "tokens_seen": 1277100032
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003096188565697091,
      "loss": 2.7885,
      "theoretical_loss": 3.567075374316623,
      "tokens_seen": 1277165568
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030960882647943833,
      "loss": 2.895,
      "theoretical_loss": 3.5670587343678646,
      "tokens_seen": 1277231104
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003095987963891675,
      "loss": 2.8902,
      "theoretical_loss": 3.56704209551195,
      "tokens_seen": 1277296640
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003095887662988967,
      "loss": 2.733,
      "theoretical_loss": 3.567025457748751,
      "tokens_seen": 1277362176
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003095787362086259,
      "loss": 2.9257,
      "theoretical_loss": 3.567008821078141,
      "tokens_seen": 1277427712
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030956870611835506,
      "loss": 2.8224,
      "theoretical_loss": 3.5669921854999913,
      "tokens_seen": 1277493248
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030955867602808424,
      "loss": 2.8904,
      "theoretical_loss": 3.566975551014174,
      "tokens_seen": 1277558784
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003095486459378135,
      "loss": 2.8281,
      "theoretical_loss": 3.5669589176205623,
      "tokens_seen": 1277624320
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003095386158475426,
      "loss": 2.7411,
      "theoretical_loss": 3.566942285319027,
      "tokens_seen": 1277689856
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030952858575727184,
      "loss": 2.7561,
      "theoretical_loss": 3.566925654109441,
      "tokens_seen": 1277755392
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030951855566700096,
      "loss": 2.8423,
      "theoretical_loss": 3.566909023991678,
      "tokens_seen": 1277820928
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003095085255767302,
      "loss": 2.9628,
      "theoretical_loss": 3.566892394965608,
      "tokens_seen": 1277886464
    },
    {
      "epoch": 0.39,
      "objective/train/advantage_avg": 0.49586355686187744,
      "objective/train/docs_used": 725220,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.92828106880188,
      "objective/train/original_loss": 2.9282805919647217,
      "objective/train/theoretical_loss": 3.566875767031105,
      "objective/train/tokens_used": 1298412000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2459433376789093,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0508365631103516,
      "objective/train/weighted_lm_loss": 3.0771162509918213,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 3.566875767031105,
      "tokens_seen": 1277952000
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003094984954864594,
      "loss": 2.822,
      "theoretical_loss": 3.566875767031105,
      "tokens_seen": 1277952000
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030948846539618856,
      "loss": 2.8454,
      "theoretical_loss": 3.5668591401880407,
      "tokens_seen": 1278017536
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030947843530591774,
      "loss": 2.7964,
      "theoretical_loss": 3.5668425144362876,
      "tokens_seen": 1278083072
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.000309468405215647,
      "loss": 2.8356,
      "theoretical_loss": 3.5668258897757186,
      "tokens_seen": 1278148608
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003094583751253761,
      "loss": 2.8254,
      "theoretical_loss": 3.5668092662062048,
      "tokens_seen": 1278214144
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030944834503510534,
      "loss": 2.6375,
      "theoretical_loss": 3.56679264372762,
      "tokens_seen": 1278279680
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030943831494483447,
      "loss": 2.6788,
      "theoretical_loss": 3.5667760223398366,
      "tokens_seen": 1278345216
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003094282848545637,
      "loss": 2.9613,
      "theoretical_loss": 3.5667594020427265,
      "tokens_seen": 1278410752
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003094182547642929,
      "loss": 3.0208,
      "theoretical_loss": 3.5667427828361626,
      "tokens_seen": 1278476288
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030940822467402207,
      "loss": 2.8736,
      "theoretical_loss": 3.5667261647200172,
      "tokens_seen": 1278541824
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030939819458375125,
      "loss": 2.9506,
      "theoretical_loss": 3.5667095476941633,
      "tokens_seen": 1278607360
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030938816449348043,
      "loss": 3.1014,
      "theoretical_loss": 3.566692931758473,
      "tokens_seen": 1278672896
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003093781344032096,
      "loss": 2.8284,
      "theoretical_loss": 3.566676316912819,
      "tokens_seen": 1278738432
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030936810431293884,
      "loss": 2.8745,
      "theoretical_loss": 3.5666597031570735,
      "tokens_seen": 1278803968
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030935807422266797,
      "loss": 2.7374,
      "theoretical_loss": 3.56664309049111,
      "tokens_seen": 1278869504
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003093480441323972,
      "loss": 2.8089,
      "theoretical_loss": 3.5666264789148014,
      "tokens_seen": 1278935040
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030933801404212633,
      "loss": 2.9336,
      "theoretical_loss": 3.5666098684280194,
      "tokens_seen": 1279000576
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030932798395185557,
      "loss": 3.0295,
      "theoretical_loss": 3.566593259030637,
      "tokens_seen": 1279066112
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030931795386158475,
      "loss": 3.2052,
      "theoretical_loss": 3.566576650722528,
      "tokens_seen": 1279131648
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030930792377131393,
      "loss": 2.8276,
      "theoretical_loss": 3.5665600435035634,
      "tokens_seen": 1279197184
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030929789368104317,
      "loss": 2.8438,
      "theoretical_loss": 3.566543437373617,
      "tokens_seen": 1279262720
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030928786359077235,
      "loss": 3.0023,
      "theoretical_loss": 3.5665268323325616,
      "tokens_seen": 1279328256
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030927783350050153,
      "loss": 3.2238,
      "theoretical_loss": 3.5665102283802694,
      "tokens_seen": 1279393792
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003092678034102307,
      "loss": 2.9107,
      "theoretical_loss": 3.5664936255166144,
      "tokens_seen": 1279459328
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003092577733199599,
      "loss": 3.0168,
      "theoretical_loss": 3.5664770237414687,
      "tokens_seen": 1279524864
    },
    {
      "epoch": 0.39,
      "objective/train/advantage_avg": 0.4942437410354614,
      "objective/train/docs_used": 725568,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.655856132507324,
      "objective/train/original_loss": 2.6558563709259033,
      "objective/train/theoretical_loss": 3.5664604230547052,
      "objective/train/tokens_used": 1300050400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24436666071414948,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506664514541626,
      "objective/train/weighted_lm_loss": 2.7903354167938232,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0304546356201172,
      "theoretical_loss": 3.5664604230547052,
      "tokens_seen": 1279590400
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030924774322968907,
      "loss": 2.8974,
      "theoretical_loss": 3.5664604230547052,
      "tokens_seen": 1279590400
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003092377131394183,
      "loss": 2.7783,
      "theoretical_loss": 3.5664438234561966,
      "tokens_seen": 1279655936
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030922768304914743,
      "loss": 2.8159,
      "theoretical_loss": 3.566427224945816,
      "tokens_seen": 1279721472
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030921765295887667,
      "loss": 2.8635,
      "theoretical_loss": 3.566410627523437,
      "tokens_seen": 1279787008
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003092076228686058,
      "loss": 3.2207,
      "theoretical_loss": 3.566394031188932,
      "tokens_seen": 1279852544
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030919759277833503,
      "loss": 2.9546,
      "theoretical_loss": 3.566377435942174,
      "tokens_seen": 1279918080
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003091875626880642,
      "loss": 2.7832,
      "theoretical_loss": 3.5663608417830357,
      "tokens_seen": 1279983616
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003091775325977934,
      "loss": 2.8279,
      "theoretical_loss": 3.566344248711391,
      "tokens_seen": 1280049152
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003091675025075226,
      "loss": 2.7158,
      "theoretical_loss": 3.566327656727112,
      "tokens_seen": 1280114688
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003091574724172518,
      "loss": 3.1215,
      "theoretical_loss": 3.5663110658300727,
      "tokens_seen": 1280180224
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030914744232698094,
      "loss": 2.8272,
      "theoretical_loss": 3.566294476020145,
      "tokens_seen": 1280245760
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030913741223671017,
      "loss": 2.902,
      "theoretical_loss": 3.5662778872972036,
      "tokens_seen": 1280311296
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003091273821464393,
      "loss": 2.927,
      "theoretical_loss": 3.5662612996611207,
      "tokens_seen": 1280376832
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030911735205616853,
      "loss": 2.7007,
      "theoretical_loss": 3.566244713111769,
      "tokens_seen": 1280442368
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003091073219658977,
      "loss": 2.7463,
      "theoretical_loss": 3.5662281276490226,
      "tokens_seen": 1280507904
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003090972918756269,
      "loss": 2.9426,
      "theoretical_loss": 3.5662115432727544,
      "tokens_seen": 1280573440
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003090872617853561,
      "loss": 2.966,
      "theoretical_loss": 3.5661949599828375,
      "tokens_seen": 1280638976
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030907723169508526,
      "loss": 3.0682,
      "theoretical_loss": 3.5661783777791456,
      "tokens_seen": 1280704512
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030906720160481444,
      "loss": 2.9409,
      "theoretical_loss": 3.566161796661551,
      "tokens_seen": 1280770048
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003090571715145437,
      "loss": 2.984,
      "theoretical_loss": 3.566145216629928,
      "tokens_seen": 1280835584
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003090471414242728,
      "loss": 2.9215,
      "theoretical_loss": 3.56612863768415,
      "tokens_seen": 1280901120
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030903711133400204,
      "loss": 2.8708,
      "theoretical_loss": 3.566112059824089,
      "tokens_seen": 1280966656
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030902708124373116,
      "loss": 2.8571,
      "theoretical_loss": 3.5660954830496197,
      "tokens_seen": 1281032192
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003090170511534604,
      "loss": 2.9714,
      "theoretical_loss": 3.5660789073606147,
      "tokens_seen": 1281097728
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003090070210631896,
      "loss": 3.1185,
      "theoretical_loss": 3.566062332756948,
      "tokens_seen": 1281163264
    },
    {
      "epoch": 0.39,
      "objective/train/advantage_avg": 0.49556005001068115,
      "objective/train/docs_used": 726917,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.0980520248413086,
      "objective/train/original_loss": 3.0980513095855713,
      "objective/train/theoretical_loss": 3.5660457592384924,
      "objective/train/tokens_used": 1301688800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24565348029136658,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050804853439331,
      "objective/train/weighted_lm_loss": 3.255378246307373,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0396549701690674,
      "theoretical_loss": 3.5660457592384924,
      "tokens_seen": 1281228800
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030899699097291876,
      "loss": 2.9123,
      "theoretical_loss": 3.5660457592384924,
      "tokens_seen": 1281228800
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030898696088264794,
      "loss": 2.9392,
      "theoretical_loss": 3.566029186805122,
      "tokens_seen": 1281294336
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003089769307923772,
      "loss": 2.9631,
      "theoretical_loss": 3.56601261545671,
      "tokens_seen": 1281359872
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003089669007021063,
      "loss": 3.0205,
      "theoretical_loss": 3.5659960451931294,
      "tokens_seen": 1281425408
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030895687061183554,
      "loss": 3.0626,
      "theoretical_loss": 3.5659794760142542,
      "tokens_seen": 1281490944
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030894684052156467,
      "loss": 2.8121,
      "theoretical_loss": 3.5659629079199586,
      "tokens_seen": 1281556480
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003089368104312939,
      "loss": 2.7783,
      "theoretical_loss": 3.5659463409101146,
      "tokens_seen": 1281622016
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003089267803410231,
      "loss": 2.5159,
      "theoretical_loss": 3.5659297749845966,
      "tokens_seen": 1281687552
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030891675025075227,
      "loss": 2.931,
      "theoretical_loss": 3.5659132101432784,
      "tokens_seen": 1281753088
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030890672016048145,
      "loss": 2.9462,
      "theoretical_loss": 3.5658966463860335,
      "tokens_seen": 1281818624
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030889669007021063,
      "loss": 2.922,
      "theoretical_loss": 3.5658800837127353,
      "tokens_seen": 1281884160
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003088866599799398,
      "loss": 2.8722,
      "theoretical_loss": 3.5658635221232573,
      "tokens_seen": 1281949696
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030887662988966904,
      "loss": 3.0121,
      "theoretical_loss": 3.5658469616174737,
      "tokens_seen": 1282015232
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030886659979939817,
      "loss": 2.6831,
      "theoretical_loss": 3.5658304021952576,
      "tokens_seen": 1282080768
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003088565697091274,
      "loss": 3.0104,
      "theoretical_loss": 3.5658138438564837,
      "tokens_seen": 1282146304
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030884653961885653,
      "loss": 3.0397,
      "theoretical_loss": 3.5657972866010246,
      "tokens_seen": 1282211840
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030883650952858577,
      "loss": 2.8206,
      "theoretical_loss": 3.5657807304287545,
      "tokens_seen": 1282277376
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030882647943831495,
      "loss": 2.8355,
      "theoretical_loss": 3.565764175339547,
      "tokens_seen": 1282342912
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030881644934804413,
      "loss": 3.1077,
      "theoretical_loss": 3.565747621333277,
      "tokens_seen": 1282408448
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003088064192577733,
      "loss": 2.9457,
      "theoretical_loss": 3.565731068409817,
      "tokens_seen": 1282473984
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030879638916750255,
      "loss": 3.1809,
      "theoretical_loss": 3.5657145165690407,
      "tokens_seen": 1282539520
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003087863590772317,
      "loss": 3.0323,
      "theoretical_loss": 3.5656979658108234,
      "tokens_seen": 1282605056
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003087763289869609,
      "loss": 2.9073,
      "theoretical_loss": 3.565681416135038,
      "tokens_seen": 1282670592
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030876629889669004,
      "loss": 3.0153,
      "theoretical_loss": 3.5656648675415585,
      "tokens_seen": 1282736128
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030875626880641927,
      "loss": 2.8781,
      "theoretical_loss": 3.5656483200302587,
      "tokens_seen": 1282801664
    },
    {
      "epoch": 0.39,
      "objective/train/advantage_avg": 0.49519845843315125,
      "objective/train/docs_used": 727506,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.7813215255737305,
      "objective/train/original_loss": 2.7813215255737305,
      "objective/train/theoretical_loss": 3.5656317736010132,
      "objective/train/tokens_used": 1303327200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24531130492687225,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050766944885254,
      "objective/train/weighted_lm_loss": 2.9225118160247803,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0396549701690674,
      "theoretical_loss": 3.5656317736010132,
      "tokens_seen": 1282867200
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030874623871614845,
      "loss": 2.8373,
      "theoretical_loss": 3.5656317736010132,
      "tokens_seen": 1282867200
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030873620862587763,
      "loss": 2.8344,
      "theoretical_loss": 3.5656152282536953,
      "tokens_seen": 1282932736
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003087261785356068,
      "loss": 2.6885,
      "theoretical_loss": 3.5655986839881786,
      "tokens_seen": 1282998272
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.000308716148445336,
      "loss": 2.9701,
      "theoretical_loss": 3.5655821408043384,
      "tokens_seen": 1283063808
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003087061183550652,
      "loss": 2.7423,
      "theoretical_loss": 3.565565598702048,
      "tokens_seen": 1283129344
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003086960882647944,
      "loss": 2.8152,
      "theoretical_loss": 3.5655490576811815,
      "tokens_seen": 1283194880
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030868605817452354,
      "loss": 3.0796,
      "theoretical_loss": 3.5655325177416124,
      "tokens_seen": 1283260416
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003086760280842528,
      "loss": 2.8813,
      "theoretical_loss": 3.5655159788832163,
      "tokens_seen": 1283325952
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003086659979939819,
      "loss": 2.9676,
      "theoretical_loss": 3.565499441105866,
      "tokens_seen": 1283391488
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030865596790371114,
      "loss": 2.8272,
      "theoretical_loss": 3.565482904409436,
      "tokens_seen": 1283457024
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003086459378134403,
      "loss": 2.9001,
      "theoretical_loss": 3.5654663687938006,
      "tokens_seen": 1283522560
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003086359077231695,
      "loss": 2.9048,
      "theoretical_loss": 3.5654498342588337,
      "tokens_seen": 1283588096
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003086258776328987,
      "loss": 2.8152,
      "theoretical_loss": 3.56543330080441,
      "tokens_seen": 1283653632
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003086158475426279,
      "loss": 3.0889,
      "theoretical_loss": 3.5654167684304032,
      "tokens_seen": 1283719168
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030860581745235704,
      "loss": 2.5823,
      "theoretical_loss": 3.5654002371366875,
      "tokens_seen": 1283784704
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003085957873620863,
      "loss": 2.8651,
      "theoretical_loss": 3.5653837069231384,
      "tokens_seen": 1283850240
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003085857572718154,
      "loss": 3.0076,
      "theoretical_loss": 3.5653671777896285,
      "tokens_seen": 1283915776
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030857572718154464,
      "loss": 3.0885,
      "theoretical_loss": 3.5653506497360326,
      "tokens_seen": 1283981312
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003085656970912738,
      "loss": 2.9934,
      "theoretical_loss": 3.565334122762226,
      "tokens_seen": 1284046848
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.000308555667001003,
      "loss": 3.084,
      "theoretical_loss": 3.5653175968680815,
      "tokens_seen": 1284112384
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030854563691073224,
      "loss": 2.7348,
      "theoretical_loss": 3.565301072053475,
      "tokens_seen": 1284177920
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030853560682046137,
      "loss": 2.7953,
      "theoretical_loss": 3.56528454831828,
      "tokens_seen": 1284243456
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003085255767301906,
      "loss": 2.5664,
      "theoretical_loss": 3.565268025662371,
      "tokens_seen": 1284308992
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003085155466399198,
      "loss": 2.9433,
      "theoretical_loss": 3.5652515040856225,
      "tokens_seen": 1284374528
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030850551654964896,
      "loss": 2.7906,
      "theoretical_loss": 3.565234983587909,
      "tokens_seen": 1284440064
    },
    {
      "epoch": 0.39,
      "objective/train/advantage_avg": 0.49206870794296265,
      "objective/train/docs_used": 728764,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.0037403106689453,
      "objective/train/original_loss": 3.0037407875061035,
      "objective/train/theoretical_loss": 3.5652184641691047,
      "objective/train/tokens_used": 1304965600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24225276708602905,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0504382848739624,
      "objective/train/weighted_lm_loss": 3.155045747756958,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0417574644088745,
      "theoretical_loss": 3.5652184641691047,
      "tokens_seen": 1284505600
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030849548645937814,
      "loss": 3.0271,
      "theoretical_loss": 3.5652184641691047,
      "tokens_seen": 1284505600
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003084854563691074,
      "loss": 2.8789,
      "theoretical_loss": 3.5652019458290845,
      "tokens_seen": 1284571136
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003084754262788365,
      "loss": 2.8036,
      "theoretical_loss": 3.565185428567723,
      "tokens_seen": 1284636672
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030846539618856574,
      "loss": 2.9076,
      "theoretical_loss": 3.5651689123848946,
      "tokens_seen": 1284702208
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030845536609829487,
      "loss": 3.0687,
      "theoretical_loss": 3.5651523972804737,
      "tokens_seen": 1284767744
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003084453360080241,
      "loss": 2.6894,
      "theoretical_loss": 3.5651358832543347,
      "tokens_seen": 1284833280
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003084353059177533,
      "loss": 2.9293,
      "theoretical_loss": 3.565119370306353,
      "tokens_seen": 1284898816
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030842527582748247,
      "loss": 2.7479,
      "theoretical_loss": 3.5651028584364024,
      "tokens_seen": 1284964352
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030841524573721165,
      "loss": 2.7788,
      "theoretical_loss": 3.5650863476443577,
      "tokens_seen": 1285029888
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030840521564694083,
      "loss": 3.0044,
      "theoretical_loss": 3.5650698379300936,
      "tokens_seen": 1285095424
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030839518555667,
      "loss": 2.8749,
      "theoretical_loss": 3.565053329293485,
      "tokens_seen": 1285160960
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030838515546639924,
      "loss": 2.9814,
      "theoretical_loss": 3.565036821734407,
      "tokens_seen": 1285226496
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030837512537612837,
      "loss": 2.8876,
      "theoretical_loss": 3.565020315252733,
      "tokens_seen": 1285292032
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003083650952858576,
      "loss": 2.7169,
      "theoretical_loss": 3.565003809848339,
      "tokens_seen": 1285357568
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030835506519558673,
      "loss": 3.005,
      "theoretical_loss": 3.5649873055210994,
      "tokens_seen": 1285423104
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030834503510531597,
      "loss": 2.8037,
      "theoretical_loss": 3.564970802270889,
      "tokens_seen": 1285488640
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030833500501504515,
      "loss": 2.9947,
      "theoretical_loss": 3.5649543000975825,
      "tokens_seen": 1285554176
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030832497492477433,
      "loss": 3.0846,
      "theoretical_loss": 3.5649377990010542,
      "tokens_seen": 1285619712
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003083149448345035,
      "loss": 3.0513,
      "theoretical_loss": 3.5649212989811803,
      "tokens_seen": 1285685248
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030830491474423275,
      "loss": 2.8212,
      "theoretical_loss": 3.5649048000378345,
      "tokens_seen": 1285750784
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003082948846539619,
      "loss": 2.9255,
      "theoretical_loss": 3.5648883021708926,
      "tokens_seen": 1285816320
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003082848545636911,
      "loss": 2.7639,
      "theoretical_loss": 3.5648718053802284,
      "tokens_seen": 1285881856
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030827482447342024,
      "loss": 2.7565,
      "theoretical_loss": 3.564855309665718,
      "tokens_seen": 1285947392
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030826479438314947,
      "loss": 2.7884,
      "theoretical_loss": 3.564838815027236,
      "tokens_seen": 1286012928
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030825476429287865,
      "loss": 3.0488,
      "theoretical_loss": 3.5648223214646566,
      "tokens_seen": 1286078464
    },
    {
      "epoch": 0.39,
      "objective/train/advantage_avg": 0.4935792088508606,
      "objective/train/docs_used": 729512,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.3926758766174316,
      "objective/train/original_loss": 2.3926753997802734,
      "objective/train/theoretical_loss": 3.564805828977856,
      "objective/train/tokens_used": 1306604000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24375119805335999,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505969524383545,
      "objective/train/weighted_lm_loss": 2.513692855834961,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0338950157165527,
      "theoretical_loss": 3.564805828977856,
      "tokens_seen": 1286144000
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030824473420260783,
      "loss": 2.8575,
      "theoretical_loss": 3.564805828977856,
      "tokens_seen": 1286144000
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.000308234704112337,
      "loss": 2.8303,
      "theoretical_loss": 3.5647893375667077,
      "tokens_seen": 1286209536
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003082246740220662,
      "loss": 2.7996,
      "theoretical_loss": 3.5647728472310884,
      "tokens_seen": 1286275072
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003082146439317954,
      "loss": 2.905,
      "theoretical_loss": 3.5647563579708725,
      "tokens_seen": 1286340608
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003082046138415246,
      "loss": 2.8682,
      "theoretical_loss": 3.5647398697859343,
      "tokens_seen": 1286406144
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030819458375125374,
      "loss": 2.8032,
      "theoretical_loss": 3.56472338267615,
      "tokens_seen": 1286471680
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.000308184553660983,
      "loss": 2.8824,
      "theoretical_loss": 3.564706896641395,
      "tokens_seen": 1286537216
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003081745235707121,
      "loss": 2.8238,
      "theoretical_loss": 3.564690411681543,
      "tokens_seen": 1286602752
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030816449348044134,
      "loss": 2.957,
      "theoretical_loss": 3.5646739277964707,
      "tokens_seen": 1286668288
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003081544633901705,
      "loss": 2.6756,
      "theoretical_loss": 3.564657444986052,
      "tokens_seen": 1286733824
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003081444332998997,
      "loss": 2.7302,
      "theoretical_loss": 3.564640963250163,
      "tokens_seen": 1286799360
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003081344032096289,
      "loss": 2.9937,
      "theoretical_loss": 3.564624482588678,
      "tokens_seen": 1286864896
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003081243731193581,
      "loss": 2.8139,
      "theoretical_loss": 3.5646080030014735,
      "tokens_seen": 1286930432
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030811434302908724,
      "loss": 2.9317,
      "theoretical_loss": 3.564591524488424,
      "tokens_seen": 1286995968
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003081043129388165,
      "loss": 2.9211,
      "theoretical_loss": 3.564575047049405,
      "tokens_seen": 1287061504
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003080942828485456,
      "loss": 2.8482,
      "theoretical_loss": 3.564558570684291,
      "tokens_seen": 1287127040
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030808425275827484,
      "loss": 2.8931,
      "theoretical_loss": 3.5645420953929587,
      "tokens_seen": 1287192576
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.000308074222668004,
      "loss": 3.1444,
      "theoretical_loss": 3.564525621175283,
      "tokens_seen": 1287258112
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003080641925777332,
      "loss": 2.7336,
      "theoretical_loss": 3.5645091480311386,
      "tokens_seen": 1287323648
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003080541624874624,
      "loss": 3.0149,
      "theoretical_loss": 3.5644926759604014,
      "tokens_seen": 1287389184
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030804413239719157,
      "loss": 2.9468,
      "theoretical_loss": 3.564476204962947,
      "tokens_seen": 1287454720
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030803410230692075,
      "loss": 2.922,
      "theoretical_loss": 3.5644597350386507,
      "tokens_seen": 1287520256
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030802407221665,
      "loss": 2.9223,
      "theoretical_loss": 3.564443266187388,
      "tokens_seen": 1287585792
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003080140421263791,
      "loss": 2.8589,
      "theoretical_loss": 3.564426798409034,
      "tokens_seen": 1287651328
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030800401203610834,
      "loss": 2.9867,
      "theoretical_loss": 3.564410331703465,
      "tokens_seen": 1287716864
    },
    {
      "epoch": 0.39,
      "objective/train/advantage_avg": 0.49352025985717773,
      "objective/train/docs_used": 730476,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.775486469268799,
      "objective/train/original_loss": 2.775486469268799,
      "objective/train/theoretical_loss": 3.5643938660705556,
      "objective/train/tokens_used": 1308242400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24363847076892853,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505905151367188,
      "objective/train/weighted_lm_loss": 2.915661096572876,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0425469875335693,
      "theoretical_loss": 3.5643938660705556,
      "tokens_seen": 1287782400
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003079939819458375,
      "loss": 3.098,
      "theoretical_loss": 3.5643938660705556,
      "tokens_seen": 1287782400
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003079839518555667,
      "loss": 2.9802,
      "theoretical_loss": 3.564377401510182,
      "tokens_seen": 1287847936
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003079739217652959,
      "loss": 2.9512,
      "theoretical_loss": 3.564360938022219,
      "tokens_seen": 1287913472
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030796389167502507,
      "loss": 2.807,
      "theoretical_loss": 3.564344475606543,
      "tokens_seen": 1287979008
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030795386158475425,
      "loss": 2.8748,
      "theoretical_loss": 3.5643280142630296,
      "tokens_seen": 1288044544
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003079438314944835,
      "loss": 2.9747,
      "theoretical_loss": 3.564311553991554,
      "tokens_seen": 1288110080
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003079338014042126,
      "loss": 2.9299,
      "theoretical_loss": 3.5642950947919916,
      "tokens_seen": 1288175616
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030792377131394185,
      "loss": 2.8388,
      "theoretical_loss": 3.564278636664219,
      "tokens_seen": 1288241152
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.000307913741223671,
      "loss": 2.9618,
      "theoretical_loss": 3.564262179608111,
      "tokens_seen": 1288306688
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003079037111334002,
      "loss": 2.9273,
      "theoretical_loss": 3.564245723623544,
      "tokens_seen": 1288372224
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003078936810431294,
      "loss": 3.0874,
      "theoretical_loss": 3.5642292687103927,
      "tokens_seen": 1288437760
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030788365095285857,
      "loss": 2.9005,
      "theoretical_loss": 3.564212814868534,
      "tokens_seen": 1288503296
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030787362086258775,
      "loss": 3.0743,
      "theoretical_loss": 3.564196362097843,
      "tokens_seen": 1288568832
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030786359077231693,
      "loss": 2.9143,
      "theoretical_loss": 3.564179910398196,
      "tokens_seen": 1288634368
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003078535606820461,
      "loss": 3.0168,
      "theoretical_loss": 3.5641634597694685,
      "tokens_seen": 1288699904
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030784353059177535,
      "loss": 2.9038,
      "theoretical_loss": 3.5641470102115362,
      "tokens_seen": 1288765440
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003078335005015045,
      "loss": 2.9784,
      "theoretical_loss": 3.564130561724275,
      "tokens_seen": 1288830976
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003078234704112337,
      "loss": 2.9279,
      "theoretical_loss": 3.5641141143075608,
      "tokens_seen": 1288896512
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003078134403209629,
      "loss": 3.0273,
      "theoretical_loss": 3.5640976679612697,
      "tokens_seen": 1288962048
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003078034102306921,
      "loss": 3.0047,
      "theoretical_loss": 3.5640812226852776,
      "tokens_seen": 1289027584
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003077933801404213,
      "loss": 2.6727,
      "theoretical_loss": 3.56406477847946,
      "tokens_seen": 1289093120
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030778335005015044,
      "loss": 2.8436,
      "theoretical_loss": 3.5640483353436934,
      "tokens_seen": 1289158656
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030777331995987967,
      "loss": 3.0492,
      "theoretical_loss": 3.5640318932778534,
      "tokens_seen": 1289224192
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030776328986960885,
      "loss": 2.8144,
      "theoretical_loss": 3.5640154522818164,
      "tokens_seen": 1289289728
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030775325977933803,
      "loss": 2.9148,
      "theoretical_loss": 3.563999012355458,
      "tokens_seen": 1289355264
    },
    {
      "epoch": 0.39,
      "objective/train/advantage_avg": 0.49363210797309875,
      "objective/train/docs_used": 731120,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.9378554821014404,
      "objective/train/original_loss": 2.9378552436828613,
      "objective/train/theoretical_loss": 3.563982573498654,
      "objective/train/tokens_used": 1309880800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2437596470117569,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506023168563843,
      "objective/train/weighted_lm_loss": 3.086427688598633,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0425469875335693,
      "theoretical_loss": 3.563982573498654,
      "tokens_seen": 1289420800
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003077432296890672,
      "loss": 2.9982,
      "theoretical_loss": 3.563982573498654,
      "tokens_seen": 1289420800
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003077331995987964,
      "loss": 2.8117,
      "theoretical_loss": 3.5639661357112815,
      "tokens_seen": 1289486336
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003077231695085256,
      "loss": 2.7378,
      "theoretical_loss": 3.563949698993216,
      "tokens_seen": 1289551872
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003077131394182548,
      "loss": 2.7324,
      "theoretical_loss": 3.563933263344333,
      "tokens_seen": 1289617408
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030770310932798394,
      "loss": 2.9701,
      "theoretical_loss": 3.5639168287645093,
      "tokens_seen": 1289682944
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003076930792377132,
      "loss": 2.8273,
      "theoretical_loss": 3.5639003952536212,
      "tokens_seen": 1289748480
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003076830491474423,
      "loss": 2.8574,
      "theoretical_loss": 3.5638839628115444,
      "tokens_seen": 1289814016
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030767301905717154,
      "loss": 2.5725,
      "theoretical_loss": 3.5638675314381554,
      "tokens_seen": 1289879552
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003076629889669007,
      "loss": 2.9199,
      "theoretical_loss": 3.56385110113333,
      "tokens_seen": 1289945088
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003076529588766299,
      "loss": 3.0936,
      "theoretical_loss": 3.563834671896945,
      "tokens_seen": 1290010624
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003076429287863591,
      "loss": 2.9538,
      "theoretical_loss": 3.563818243728876,
      "tokens_seen": 1290076160
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003076328986960883,
      "loss": 2.7066,
      "theoretical_loss": 3.5638018166289998,
      "tokens_seen": 1290141696
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030762286860581744,
      "loss": 2.6404,
      "theoretical_loss": 3.5637853905971926,
      "tokens_seen": 1290207232
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003076128385155467,
      "loss": 2.6721,
      "theoretical_loss": 3.5637689656333302,
      "tokens_seen": 1290272768
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003076028084252758,
      "loss": 2.9595,
      "theoretical_loss": 3.5637525417372897,
      "tokens_seen": 1290338304
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030759277833500504,
      "loss": 2.7717,
      "theoretical_loss": 3.563736118908947,
      "tokens_seen": 1290403840
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003075827482447342,
      "loss": 2.8869,
      "theoretical_loss": 3.563719697148179,
      "tokens_seen": 1290469376
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003075727181544634,
      "loss": 2.9508,
      "theoretical_loss": 3.563703276454861,
      "tokens_seen": 1290534912
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003075626880641926,
      "loss": 2.8392,
      "theoretical_loss": 3.56368685682887,
      "tokens_seen": 1290600448
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030755265797392177,
      "loss": 2.6802,
      "theoretical_loss": 3.563670438270083,
      "tokens_seen": 1290665984
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030754262788365095,
      "loss": 3.099,
      "theoretical_loss": 3.563654020778376,
      "tokens_seen": 1290731520
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003075325977933802,
      "loss": 2.9172,
      "theoretical_loss": 3.563637604353625,
      "tokens_seen": 1290797056
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003075225677031093,
      "loss": 3.093,
      "theoretical_loss": 3.5636211889957066,
      "tokens_seen": 1290862592
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030751253761283854,
      "loss": 2.9599,
      "theoretical_loss": 3.5636047747044985,
      "tokens_seen": 1290928128
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003075025075225677,
      "loss": 2.9574,
      "theoretical_loss": 3.5635883614798756,
      "tokens_seen": 1290993664
    },
    {
      "epoch": 0.39,
      "objective/train/advantage_avg": 0.4943446218967438,
      "objective/train/docs_used": 732356,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.656567335128784,
      "objective/train/original_loss": 2.6565675735473633,
      "objective/train/theoretical_loss": 3.5635719493217155,
      "objective/train/tokens_used": 1311519200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24443094432353973,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050676941871643,
      "objective/train/weighted_lm_loss": 2.7908756732940674,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0457526445388794,
      "theoretical_loss": 3.5635719493217155,
      "tokens_seen": 1291059200
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003074924774322969,
      "loss": 2.9155,
      "theoretical_loss": 3.5635719493217155,
      "tokens_seen": 1291059200
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003074824473420261,
      "loss": 2.9503,
      "theoretical_loss": 3.5635555382298945,
      "tokens_seen": 1291124736
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030747241725175527,
      "loss": 2.8555,
      "theoretical_loss": 3.563539128204289,
      "tokens_seen": 1291190272
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030746238716148445,
      "loss": 2.82,
      "theoretical_loss": 3.563522719244776,
      "tokens_seen": 1291255808
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003074523570712137,
      "loss": 2.8792,
      "theoretical_loss": 3.563506311351232,
      "tokens_seen": 1291321344
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003074423269809428,
      "loss": 2.8807,
      "theoretical_loss": 3.563489904523533,
      "tokens_seen": 1291386880
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030743229689067205,
      "loss": 3.0387,
      "theoretical_loss": 3.563473498761557,
      "tokens_seen": 1291452416
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003074222668004012,
      "loss": 2.647,
      "theoretical_loss": 3.56345709406518,
      "tokens_seen": 1291517952
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003074122367101304,
      "loss": 2.9115,
      "theoretical_loss": 3.5634406904342786,
      "tokens_seen": 1291583488
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003074022066198596,
      "loss": 2.9853,
      "theoretical_loss": 3.5634242878687292,
      "tokens_seen": 1291649024
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030739217652958877,
      "loss": 2.9899,
      "theoretical_loss": 3.5634078863684095,
      "tokens_seen": 1291714560
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030738214643931795,
      "loss": 3.137,
      "theoretical_loss": 3.563391485933195,
      "tokens_seen": 1291780096
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030737211634904713,
      "loss": 2.9495,
      "theoretical_loss": 3.563375086562964,
      "tokens_seen": 1291845632
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003073620862587763,
      "loss": 2.8986,
      "theoretical_loss": 3.5633586882575923,
      "tokens_seen": 1291911168
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030735205616850555,
      "loss": 2.7777,
      "theoretical_loss": 3.563342291016957,
      "tokens_seen": 1291976704
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003073420260782347,
      "loss": 2.894,
      "theoretical_loss": 3.5633258948409354,
      "tokens_seen": 1292042240
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003073319959879639,
      "loss": 3.0013,
      "theoretical_loss": 3.563309499729404,
      "tokens_seen": 1292107776
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003073219658976931,
      "loss": 2.8793,
      "theoretical_loss": 3.5632931056822392,
      "tokens_seen": 1292173312
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003073119358074223,
      "loss": 2.8988,
      "theoretical_loss": 3.563276712699319,
      "tokens_seen": 1292238848
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030730190571715146,
      "loss": 2.982,
      "theoretical_loss": 3.563260320780519,
      "tokens_seen": 1292304384
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030729187562688064,
      "loss": 2.8779,
      "theoretical_loss": 3.5632439299257177,
      "tokens_seen": 1292369920
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003072818455366098,
      "loss": 2.9324,
      "theoretical_loss": 3.5632275401347906,
      "tokens_seen": 1292435456
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030727181544633905,
      "loss": 2.9774,
      "theoretical_loss": 3.563211151407616,
      "tokens_seen": 1292500992
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003072617853560682,
      "loss": 2.8754,
      "theoretical_loss": 3.56319476374407,
      "tokens_seen": 1292566528
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003072517552657974,
      "loss": 2.9391,
      "theoretical_loss": 3.56317837714403,
      "tokens_seen": 1292632064
    },
    {
      "epoch": 0.39,
      "objective/train/advantage_avg": 0.4920934736728668,
      "objective/train/docs_used": 733109,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.886225938796997,
      "objective/train/original_loss": 2.886225938796997,
      "objective/train/theoretical_loss": 3.563161991607373,
      "objective/train/tokens_used": 1313157600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24225148558616638,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0504406690597534,
      "objective/train/weighted_lm_loss": 3.0317935943603516,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0425469875335693,
      "theoretical_loss": 3.563161991607373,
      "tokens_seen": 1292697600
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030724172517552654,
      "loss": 3.0961,
      "theoretical_loss": 3.563161991607373,
      "tokens_seen": 1292697600
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003072316950852558,
      "loss": 2.9932,
      "theoretical_loss": 3.5631456071339764,
      "tokens_seen": 1292763136
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030722166499498496,
      "loss": 2.9081,
      "theoretical_loss": 3.563129223723717,
      "tokens_seen": 1292828672
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030721163490471414,
      "loss": 2.9755,
      "theoretical_loss": 3.563112841376472,
      "tokens_seen": 1292894208
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003072016048144433,
      "loss": 2.8432,
      "theoretical_loss": 3.5630964600921184,
      "tokens_seen": 1292959744
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003071915747241725,
      "loss": 2.659,
      "theoretical_loss": 3.5630800798705335,
      "tokens_seen": 1293025280
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003071815446339017,
      "loss": 2.9428,
      "theoretical_loss": 3.5630637007115946,
      "tokens_seen": 1293090816
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003071715145436309,
      "loss": 2.8738,
      "theoretical_loss": 3.563047322615179,
      "tokens_seen": 1293156352
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030716148445336005,
      "loss": 2.9361,
      "theoretical_loss": 3.5630309455811635,
      "tokens_seen": 1293221888
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003071514543630893,
      "loss": 2.8886,
      "theoretical_loss": 3.5630145696094258,
      "tokens_seen": 1293287424
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030714142427281846,
      "loss": 2.6652,
      "theoretical_loss": 3.5629981946998424,
      "tokens_seen": 1293352960
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030713139418254764,
      "loss": 3.0195,
      "theoretical_loss": 3.5629818208522916,
      "tokens_seen": 1293418496
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003071213640922768,
      "loss": 2.881,
      "theoretical_loss": 3.5629654480666506,
      "tokens_seen": 1293484032
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.000307111334002006,
      "loss": 2.997,
      "theoretical_loss": 3.562949076342796,
      "tokens_seen": 1293549568
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003071013039117352,
      "loss": 2.8748,
      "theoretical_loss": 3.5629327056806055,
      "tokens_seen": 1293615104
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003070912738214644,
      "loss": 2.9137,
      "theoretical_loss": 3.5629163360799563,
      "tokens_seen": 1293680640
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030708124373119355,
      "loss": 3.0492,
      "theoretical_loss": 3.5628999675407265,
      "tokens_seen": 1293746176
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003070712136409228,
      "loss": 2.9539,
      "theoretical_loss": 3.5628836000627926,
      "tokens_seen": 1293811712
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003070611835506519,
      "loss": 2.8564,
      "theoretical_loss": 3.562867233646033,
      "tokens_seen": 1293877248
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030705115346038115,
      "loss": 2.8471,
      "theoretical_loss": 3.562850868290324,
      "tokens_seen": 1293942784
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003070411233701104,
      "loss": 2.9761,
      "theoretical_loss": 3.5628345039955445,
      "tokens_seen": 1294008320
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003070310932798395,
      "loss": 2.996,
      "theoretical_loss": 3.5628181407615704,
      "tokens_seen": 1294073856
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030702106318956874,
      "loss": 3.0215,
      "theoretical_loss": 3.5628017785882804,
      "tokens_seen": 1294139392
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003070110330992979,
      "loss": 2.901,
      "theoretical_loss": 3.5627854174755518,
      "tokens_seen": 1294204928
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003070010030090271,
      "loss": 2.9487,
      "theoretical_loss": 3.562769057423262,
      "tokens_seen": 1294270464
    },
    {
      "epoch": 0.39,
      "objective/train/advantage_avg": 0.49379560351371765,
      "objective/train/docs_used": 734261,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.435786247253418,
      "objective/train/original_loss": 2.435786247253418,
      "objective/train/theoretical_loss": 3.5627526984312885,
      "objective/train/tokens_used": 1314796000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2439122200012207,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506194829940796,
      "objective/train/weighted_lm_loss": 2.5590336322784424,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0417574644088745,
      "theoretical_loss": 3.5627526984312885,
      "tokens_seen": 1294336000
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003069909729187563,
      "loss": 2.667,
      "theoretical_loss": 3.5627526984312885,
      "tokens_seen": 1294336000
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030698094282848547,
      "loss": 2.5764,
      "theoretical_loss": 3.5627363404995087,
      "tokens_seen": 1294401536
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030697091273821465,
      "loss": 2.9496,
      "theoretical_loss": 3.562719983627801,
      "tokens_seen": 1294467072
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003069608826479439,
      "loss": 2.6097,
      "theoretical_loss": 3.5627036278160418,
      "tokens_seen": 1294532608
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.000306950852557673,
      "loss": 2.6928,
      "theoretical_loss": 3.5626872730641104,
      "tokens_seen": 1294598144
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030694082246740225,
      "loss": 2.817,
      "theoretical_loss": 3.562670919371883,
      "tokens_seen": 1294663680
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003069307923771314,
      "loss": 2.8066,
      "theoretical_loss": 3.562654566739238,
      "tokens_seen": 1294729216
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003069207622868606,
      "loss": 2.6992,
      "theoretical_loss": 3.5626382151660527,
      "tokens_seen": 1294794752
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003069107321965898,
      "loss": 2.8515,
      "theoretical_loss": 3.562621864652206,
      "tokens_seen": 1294860288
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030690070210631897,
      "loss": 2.9734,
      "theoretical_loss": 3.562605515197574,
      "tokens_seen": 1294925824
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030689067201604815,
      "loss": 2.9891,
      "theoretical_loss": 3.5625891668020353,
      "tokens_seen": 1294991360
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030688064192577733,
      "loss": 2.8418,
      "theoretical_loss": 3.562572819465468,
      "tokens_seen": 1295056896
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003068706118355065,
      "loss": 2.8774,
      "theoretical_loss": 3.5625564731877493,
      "tokens_seen": 1295122432
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030686058174523575,
      "loss": 2.7452,
      "theoretical_loss": 3.5625401279687576,
      "tokens_seen": 1295187968
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003068505516549649,
      "loss": 3.0876,
      "theoretical_loss": 3.5625237838083708,
      "tokens_seen": 1295253504
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003068405215646941,
      "loss": 2.9117,
      "theoretical_loss": 3.562507440706466,
      "tokens_seen": 1295319040
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003068304914744233,
      "loss": 2.5279,
      "theoretical_loss": 3.562491098662922,
      "tokens_seen": 1295384576
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003068204613841525,
      "loss": 3.0456,
      "theoretical_loss": 3.5624747576776157,
      "tokens_seen": 1295450112
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030681043129388166,
      "loss": 2.7351,
      "theoretical_loss": 3.562458417750426,
      "tokens_seen": 1295515648
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030680040120361084,
      "loss": 2.924,
      "theoretical_loss": 3.5624420788812308,
      "tokens_seen": 1295581184
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030679037111334,
      "loss": 2.7635,
      "theoretical_loss": 3.5624257410699074,
      "tokens_seen": 1295646720
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030678034102306925,
      "loss": 2.7193,
      "theoretical_loss": 3.5624094043163343,
      "tokens_seen": 1295712256
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003067703109327984,
      "loss": 2.9585,
      "theoretical_loss": 3.5623930686203895,
      "tokens_seen": 1295777792
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003067602808425276,
      "loss": 2.9628,
      "theoretical_loss": 3.562376733981951,
      "tokens_seen": 1295843328
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030675025075225674,
      "loss": 3.2008,
      "theoretical_loss": 3.5623604004008964,
      "tokens_seen": 1295908864
    },
    {
      "epoch": 0.39,
      "objective/train/advantage_avg": 0.49555012583732605,
      "objective/train/docs_used": 734748,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.8208982944488525,
      "objective/train/original_loss": 2.8208980560302734,
      "objective/train/theoretical_loss": 3.562344067877105,
      "objective/train/tokens_used": 1316434400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24561043083667755,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0508036613464355,
      "objective/train/weighted_lm_loss": 2.9642088413238525,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.04635488986969,
      "theoretical_loss": 3.562344067877105,
      "tokens_seen": 1295974400
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.000306740220661986,
      "loss": 2.9647,
      "theoretical_loss": 3.562344067877105,
      "tokens_seen": 1295974400
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030673019057171516,
      "loss": 2.7244,
      "theoretical_loss": 3.5623277364104537,
      "tokens_seen": 1296039936
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030672016048144434,
      "loss": 2.9413,
      "theoretical_loss": 3.5623114060008207,
      "tokens_seen": 1296105472
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003067101303911735,
      "loss": 2.8765,
      "theoretical_loss": 3.562295076648085,
      "tokens_seen": 1296171008
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003067101303911735,
      "loss": 3.0967,
      "theoretical_loss": 3.5622787483521243,
      "tokens_seen": 1296236544
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003067001003009027,
      "loss": 2.8251,
      "theoretical_loss": 3.562262421112816,
      "tokens_seen": 1296302080
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003066900702106319,
      "loss": 2.7346,
      "theoretical_loss": 3.56224609493004,
      "tokens_seen": 1296367616
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003066800401203611,
      "loss": 2.8263,
      "theoretical_loss": 3.5622297698036736,
      "tokens_seen": 1296433152
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030667001003009025,
      "loss": 2.7082,
      "theoretical_loss": 3.5622134457335943,
      "tokens_seen": 1296498688
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003066599799398195,
      "loss": 2.8082,
      "theoretical_loss": 3.5621971227196814,
      "tokens_seen": 1296564224
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030664994984954866,
      "loss": 2.9528,
      "theoretical_loss": 3.562180800761813,
      "tokens_seen": 1296629760
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030663991975927784,
      "loss": 2.8698,
      "theoretical_loss": 3.562164479859867,
      "tokens_seen": 1296695296
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.000306629889669007,
      "loss": 2.9841,
      "theoretical_loss": 3.5621481600137224,
      "tokens_seen": 1296760832
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003066198595787362,
      "loss": 2.9234,
      "theoretical_loss": 3.5621318412232568,
      "tokens_seen": 1296826368
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003066098294884654,
      "loss": 2.869,
      "theoretical_loss": 3.562115523488349,
      "tokens_seen": 1296891904
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003065997993981946,
      "loss": 2.7881,
      "theoretical_loss": 3.5620992068088775,
      "tokens_seen": 1296957440
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030658976930792375,
      "loss": 2.7342,
      "theoretical_loss": 3.56208289118472,
      "tokens_seen": 1297022976
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.000306579739217653,
      "loss": 2.8741,
      "theoretical_loss": 3.562066576615756,
      "tokens_seen": 1297088512
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003065697091273821,
      "loss": 3.0412,
      "theoretical_loss": 3.5620502631018636,
      "tokens_seen": 1297154048
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030655967903711135,
      "loss": 2.8647,
      "theoretical_loss": 3.5620339506429204,
      "tokens_seen": 1297219584
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030654964894684053,
      "loss": 2.8991,
      "theoretical_loss": 3.562017639238806,
      "tokens_seen": 1297285120
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003065396188565697,
      "loss": 2.8246,
      "theoretical_loss": 3.5620013288893984,
      "tokens_seen": 1297350656
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003065295887662989,
      "loss": 2.4593,
      "theoretical_loss": 3.5619850195945757,
      "tokens_seen": 1297416192
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003065195586760281,
      "loss": 2.9414,
      "theoretical_loss": 3.5619687113542176,
      "tokens_seen": 1297481728
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030650952858575725,
      "loss": 2.6502,
      "theoretical_loss": 3.561952404168202,
      "tokens_seen": 1297547264
    },
    {
      "epoch": 0.39,
      "objective/train/advantage_avg": 0.49200573563575745,
      "objective/train/docs_used": 735602,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.9423274993896484,
      "objective/train/original_loss": 2.942328453063965,
      "objective/train/theoretical_loss": 3.5619360980364068,
      "objective/train/tokens_used": 1318072800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24216966331005096,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050431489944458,
      "objective/train/weighted_lm_loss": 3.090743064880371,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0408108234405518,
      "theoretical_loss": 3.5619360980364068,
      "tokens_seen": 1297612800
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003064994984954865,
      "loss": 2.8987,
      "theoretical_loss": 3.5619360980364068,
      "tokens_seen": 1297612800
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003064894684052156,
      "loss": 2.777,
      "theoretical_loss": 3.5619197929587116,
      "tokens_seen": 1297678336
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030647943831494485,
      "loss": 3.0158,
      "theoretical_loss": 3.561903488934995,
      "tokens_seen": 1297743872
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030646940822467403,
      "loss": 2.6863,
      "theoretical_loss": 3.5618871859651353,
      "tokens_seen": 1297809408
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003064593781344032,
      "loss": 2.7461,
      "theoretical_loss": 3.561870884049011,
      "tokens_seen": 1297874944
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003064493480441324,
      "loss": 2.7215,
      "theoretical_loss": 3.561854583186501,
      "tokens_seen": 1297940480
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003064393179538616,
      "loss": 2.9588,
      "theoretical_loss": 3.561838283377485,
      "tokens_seen": 1298006016
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030642928786359076,
      "loss": 2.7041,
      "theoretical_loss": 3.5618219846218393,
      "tokens_seen": 1298071552
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030641925777332,
      "loss": 2.8687,
      "theoretical_loss": 3.5618056869194454,
      "tokens_seen": 1298137088
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003064092276830491,
      "loss": 2.6458,
      "theoretical_loss": 3.56178939027018,
      "tokens_seen": 1298202624
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030639919759277835,
      "loss": 2.8662,
      "theoretical_loss": 3.561773094673923,
      "tokens_seen": 1298268160
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003063891675025075,
      "loss": 2.7844,
      "theoretical_loss": 3.561756800130553,
      "tokens_seen": 1298333696
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003063791374122367,
      "loss": 3.1528,
      "theoretical_loss": 3.5617405066399486,
      "tokens_seen": 1298399232
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003063691073219659,
      "loss": 2.7217,
      "theoretical_loss": 3.5617242142019885,
      "tokens_seen": 1298464768
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003063590772316951,
      "loss": 3.1301,
      "theoretical_loss": 3.5617079228165522,
      "tokens_seen": 1298530304
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030634904714142426,
      "loss": 2.7424,
      "theoretical_loss": 3.5616916324835177,
      "tokens_seen": 1298595840
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003063390170511535,
      "loss": 3.0166,
      "theoretical_loss": 3.561675343202765,
      "tokens_seen": 1298661376
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003063289869608826,
      "loss": 3.0777,
      "theoretical_loss": 3.5616590549741725,
      "tokens_seen": 1298726912
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030631895687061186,
      "loss": 3.0141,
      "theoretical_loss": 3.5616427677976192,
      "tokens_seen": 1298792448
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.000306308926780341,
      "loss": 3.0421,
      "theoretical_loss": 3.5616264816729837,
      "tokens_seen": 1298857984
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003062988966900702,
      "loss": 2.973,
      "theoretical_loss": 3.561610196600145,
      "tokens_seen": 1298923520
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030628886659979945,
      "loss": 3.0104,
      "theoretical_loss": 3.561593912578983,
      "tokens_seen": 1298989056
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003062788365095286,
      "loss": 2.9434,
      "theoretical_loss": 3.561577629609376,
      "tokens_seen": 1299054592
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003062688064192578,
      "loss": 3.1599,
      "theoretical_loss": 3.5615613476912027,
      "tokens_seen": 1299120128
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030625877632898694,
      "loss": 2.8786,
      "theoretical_loss": 3.561545066824343,
      "tokens_seen": 1299185664
    },
    {
      "epoch": 0.39,
      "objective/train/advantage_avg": 0.49461960792541504,
      "objective/train/docs_used": 736089,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.637462854385376,
      "objective/train/original_loss": 2.637463092803955,
      "objective/train/theoretical_loss": 3.5615287870086756,
      "objective/train/tokens_used": 1319711200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24472400546073914,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507060289382935,
      "objective/train/weighted_lm_loss": 2.7712199687957764,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0417574644088745,
      "theoretical_loss": 3.5615287870086756,
      "tokens_seen": 1299251200
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003062487462387162,
      "loss": 3.0309,
      "theoretical_loss": 3.5615287870086756,
      "tokens_seen": 1299251200
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030623871614844536,
      "loss": 3.0148,
      "theoretical_loss": 3.5615125082440793,
      "tokens_seen": 1299316736
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030622868605817454,
      "loss": 3.0852,
      "theoretical_loss": 3.561496230530434,
      "tokens_seen": 1299382272
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003062186559679037,
      "loss": 3.0599,
      "theoretical_loss": 3.5614799538676185,
      "tokens_seen": 1299447808
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003062086258776329,
      "loss": 3.004,
      "theoretical_loss": 3.561463678255512,
      "tokens_seen": 1299513344
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003061985957873621,
      "loss": 2.8924,
      "theoretical_loss": 3.561447403693993,
      "tokens_seen": 1299578880
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003061885656970913,
      "loss": 2.8316,
      "theoretical_loss": 3.5614311301829416,
      "tokens_seen": 1299644416
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030617853560682045,
      "loss": 3.0341,
      "theoretical_loss": 3.561414857722237,
      "tokens_seen": 1299709952
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003061685055165497,
      "loss": 2.9638,
      "theoretical_loss": 3.561398586311758,
      "tokens_seen": 1299775488
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030615847542627886,
      "loss": 3.0364,
      "theoretical_loss": 3.5613823159513838,
      "tokens_seen": 1299841024
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030614844533600804,
      "loss": 2.9468,
      "theoretical_loss": 3.5613660466409938,
      "tokens_seen": 1299906560
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003061384152457372,
      "loss": 2.957,
      "theoretical_loss": 3.561349778380468,
      "tokens_seen": 1299972096
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003061283851554664,
      "loss": 2.7402,
      "theoretical_loss": 3.5613335111696847,
      "tokens_seen": 1300037632
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003061183550651956,
      "loss": 2.9469,
      "theoretical_loss": 3.561317245008524,
      "tokens_seen": 1300103168
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003061083249749248,
      "loss": 2.948,
      "theoretical_loss": 3.561300979896865,
      "tokens_seen": 1300168704
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030609829488465395,
      "loss": 2.8336,
      "theoretical_loss": 3.561284715834587,
      "tokens_seen": 1300234240
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003060882647943832,
      "loss": 3.1149,
      "theoretical_loss": 3.5612684528215697,
      "tokens_seen": 1300299776
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003060782347041123,
      "loss": 3.0048,
      "theoretical_loss": 3.561252190857692,
      "tokens_seen": 1300365312
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030606820461384155,
      "loss": 2.9639,
      "theoretical_loss": 3.5612359299428338,
      "tokens_seen": 1300430848
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030605817452357073,
      "loss": 3.2176,
      "theoretical_loss": 3.5612196700768743,
      "tokens_seen": 1300496384
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003060481444332999,
      "loss": 3.1677,
      "theoretical_loss": 3.5612034112596933,
      "tokens_seen": 1300561920
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003060381143430291,
      "loss": 2.8607,
      "theoretical_loss": 3.5611871534911703,
      "tokens_seen": 1300627456
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003060280842527583,
      "loss": 2.9181,
      "theoretical_loss": 3.561170896771184,
      "tokens_seen": 1300692992
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030601805416248745,
      "loss": 3.0842,
      "theoretical_loss": 3.5611546410996153,
      "tokens_seen": 1300758528
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003060080240722167,
      "loss": 2.9626,
      "theoretical_loss": 3.561138386476343,
      "tokens_seen": 1300824064
    },
    {
      "epoch": 0.39,
      "objective/train/advantage_avg": 0.49522367119789124,
      "objective/train/docs_used": 737292,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 1.8815025091171265,
      "objective/train/original_loss": 1.8815025091171265,
      "objective/train/theoretical_loss": 3.5611221329012466,
      "objective/train/tokens_used": 1321349600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24528788030147552,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050769329071045,
      "objective/train/weighted_lm_loss": 1.9770714044570923,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0475232601165771,
      "theoretical_loss": 3.5611221329012466,
      "tokens_seen": 1300889600
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003059979939819458,
      "loss": 2.5198,
      "theoretical_loss": 3.5611221329012466,
      "tokens_seen": 1300889600
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030598796389167505,
      "loss": 3.0399,
      "theoretical_loss": 3.561105880374206,
      "tokens_seen": 1300955136
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030597793380140423,
      "loss": 2.7313,
      "theoretical_loss": 3.5610896288951004,
      "tokens_seen": 1301020672
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003059679037111334,
      "loss": 3.0702,
      "theoretical_loss": 3.5610733784638104,
      "tokens_seen": 1301086208
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003059578736208626,
      "loss": 2.917,
      "theoretical_loss": 3.561057129080215,
      "tokens_seen": 1301151744
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003059478435305918,
      "loss": 3.1443,
      "theoretical_loss": 3.5610408807441933,
      "tokens_seen": 1301217280
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030593781344032096,
      "loss": 2.9103,
      "theoretical_loss": 3.5610246334556255,
      "tokens_seen": 1301282816
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003059277833500502,
      "loss": 3.0004,
      "theoretical_loss": 3.5610083872143923,
      "tokens_seen": 1301348352
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003059177532597793,
      "loss": 2.7617,
      "theoretical_loss": 3.560992142020372,
      "tokens_seen": 1301413888
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030590772316950855,
      "loss": 2.6498,
      "theoretical_loss": 3.560975897873445,
      "tokens_seen": 1301479424
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003058976930792377,
      "loss": 3.1149,
      "theoretical_loss": 3.5609596547734914,
      "tokens_seen": 1301544960
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003058876629889669,
      "loss": 2.9483,
      "theoretical_loss": 3.5609434127203903,
      "tokens_seen": 1301610496
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003058776328986961,
      "loss": 2.7823,
      "theoretical_loss": 3.560927171714022,
      "tokens_seen": 1301676032
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003058676028084253,
      "loss": 2.7404,
      "theoretical_loss": 3.5609109317542664,
      "tokens_seen": 1301741568
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030585757271815446,
      "loss": 3.1074,
      "theoretical_loss": 3.5608946928410035,
      "tokens_seen": 1301807104
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003058475426278837,
      "loss": 2.9482,
      "theoretical_loss": 3.5608784549741124,
      "tokens_seen": 1301872640
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003058375125376128,
      "loss": 3.0424,
      "theoretical_loss": 3.5608622181534733,
      "tokens_seen": 1301938176
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030582748244734206,
      "loss": 2.8569,
      "theoretical_loss": 3.560845982378967,
      "tokens_seen": 1302003712
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003058174523570712,
      "loss": 2.9006,
      "theoretical_loss": 3.560829747650472,
      "tokens_seen": 1302069248
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003058074222668004,
      "loss": 3.1691,
      "theoretical_loss": 3.5608135139678696,
      "tokens_seen": 1302134784
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003057973921765296,
      "loss": 3.118,
      "theoretical_loss": 3.5607972813310393,
      "tokens_seen": 1302200320
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003057873620862588,
      "loss": 2.9756,
      "theoretical_loss": 3.5607810497398606,
      "tokens_seen": 1302265856
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030577733199598796,
      "loss": 2.7986,
      "theoretical_loss": 3.5607648191942145,
      "tokens_seen": 1302331392
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030576730190571714,
      "loss": 2.7466,
      "theoretical_loss": 3.56074858969398,
      "tokens_seen": 1302396928
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003057572718154463,
      "loss": 2.8396,
      "theoretical_loss": 3.560732361239038,
      "tokens_seen": 1302462464
    },
    {
      "epoch": 0.39,
      "objective/train/advantage_avg": 0.4942096471786499,
      "objective/train/docs_used": 737760,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.5456202030181885,
      "objective/train/original_loss": 2.545619487762451,
      "objective/train/theoretical_loss": 3.5607161338292683,
      "objective/train/tokens_used": 1322988000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2442919760942459,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506627559661865,
      "objective/train/weighted_lm_loss": 2.6744868755340576,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0432155132293701,
      "theoretical_loss": 3.5607161338292683,
      "tokens_seen": 1302528000
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030574724172517556,
      "loss": 2.8162,
      "theoretical_loss": 3.5607161338292683,
      "tokens_seen": 1302528000
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003057372116349047,
      "loss": 3.0097,
      "theoretical_loss": 3.5606999074645507,
      "tokens_seen": 1302593536
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003057271815446339,
      "loss": 2.8728,
      "theoretical_loss": 3.5606836821447656,
      "tokens_seen": 1302659072
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030571715145436305,
      "loss": 3.0564,
      "theoretical_loss": 3.560667457869793,
      "tokens_seen": 1302724608
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003057071213640923,
      "loss": 2.9943,
      "theoretical_loss": 3.5606512346395136,
      "tokens_seen": 1302790144
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030569709127382147,
      "loss": 2.8783,
      "theoretical_loss": 3.5606350124538073,
      "tokens_seen": 1302855680
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030568706118355065,
      "loss": 3.0305,
      "theoretical_loss": 3.5606187913125535,
      "tokens_seen": 1302921216
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030567703109327983,
      "loss": 2.7673,
      "theoretical_loss": 3.5606025712156333,
      "tokens_seen": 1302986752
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030566700100300906,
      "loss": 2.9458,
      "theoretical_loss": 3.560586352162927,
      "tokens_seen": 1303052288
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003056569709127382,
      "loss": 2.8028,
      "theoretical_loss": 3.5605701341543146,
      "tokens_seen": 1303117824
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003056469408224674,
      "loss": 2.9419,
      "theoretical_loss": 3.5605539171896767,
      "tokens_seen": 1303183360
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030563691073219655,
      "loss": 3.0565,
      "theoretical_loss": 3.560537701268893,
      "tokens_seen": 1303248896
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.0003056268806419258,
      "loss": 2.9963,
      "theoretical_loss": 3.5605214863918437,
      "tokens_seen": 1303314432
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030561685055165497,
      "loss": 2.8998,
      "theoretical_loss": 3.56050527255841,
      "tokens_seen": 1303379968
    },
    {
      "epoch": 0.39,
      "learning_rate": 0.00030560682046138415,
      "loss": 2.9291,
      "theoretical_loss": 3.5604890597684715,
      "tokens_seen": 1303445504
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030559679037111333,
      "loss": 2.9096,
      "theoretical_loss": 3.5604728480219094,
      "tokens_seen": 1303511040
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003055867602808425,
      "loss": 2.6512,
      "theoretical_loss": 3.5604566373186035,
      "tokens_seen": 1303576576
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003055767301905717,
      "loss": 2.9301,
      "theoretical_loss": 3.560440427658434,
      "tokens_seen": 1303642112
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030556670010030093,
      "loss": 3.0549,
      "theoretical_loss": 3.560424219041282,
      "tokens_seen": 1303707648
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030555667001003006,
      "loss": 3.0315,
      "theoretical_loss": 3.560408011467027,
      "tokens_seen": 1303773184
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003055466399197593,
      "loss": 2.6396,
      "theoretical_loss": 3.560391804935551,
      "tokens_seen": 1303838720
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003055366098294885,
      "loss": 2.8094,
      "theoretical_loss": 3.5603755994467328,
      "tokens_seen": 1303904256
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030552657973921765,
      "loss": 3.1012,
      "theoretical_loss": 3.5603593950004546,
      "tokens_seen": 1303969792
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003055165496489469,
      "loss": 2.9095,
      "theoretical_loss": 3.5603431915965955,
      "tokens_seen": 1304035328
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.000305506519558676,
      "loss": 3.2549,
      "theoretical_loss": 3.5603269892350364,
      "tokens_seen": 1304100864
    },
    {
      "epoch": 0.4,
      "objective/train/advantage_avg": 0.49449700117111206,
      "objective/train/docs_used": 739145,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.7589046955108643,
      "objective/train/original_loss": 2.7589049339294434,
      "objective/train/theoretical_loss": 3.5603107879156584,
      "objective/train/tokens_used": 1324626400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24459508061408997,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506930351257324,
      "objective/train/weighted_lm_loss": 2.89860463142395,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0363596677780151,
      "theoretical_loss": 3.5603107879156584,
      "tokens_seen": 1304166400
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030549648946840525,
      "loss": 2.9824,
      "theoretical_loss": 3.5603107879156584,
      "tokens_seen": 1304166400
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030548645937813443,
      "loss": 2.987,
      "theoretical_loss": 3.5602945876383423,
      "tokens_seen": 1304231936
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003054764292878636,
      "loss": 2.8963,
      "theoretical_loss": 3.5602783884029674,
      "tokens_seen": 1304297472
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003054663991975928,
      "loss": 2.8833,
      "theoretical_loss": 3.5602621902094156,
      "tokens_seen": 1304363008
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.000305456369107322,
      "loss": 2.8516,
      "theoretical_loss": 3.560245993057567,
      "tokens_seen": 1304428544
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030544633901705116,
      "loss": 2.995,
      "theoretical_loss": 3.5602297969473025,
      "tokens_seen": 1304494080
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003054363089267804,
      "loss": 2.9217,
      "theoretical_loss": 3.5602136018785027,
      "tokens_seen": 1304559616
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003054262788365095,
      "loss": 2.883,
      "theoretical_loss": 3.5601974078510477,
      "tokens_seen": 1304625152
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030541624874623875,
      "loss": 2.8482,
      "theoretical_loss": 3.5601812148648193,
      "tokens_seen": 1304690688
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003054062186559679,
      "loss": 2.8311,
      "theoretical_loss": 3.5601650229196977,
      "tokens_seen": 1304756224
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003053961885656971,
      "loss": 3.0566,
      "theoretical_loss": 3.560148832015564,
      "tokens_seen": 1304821760
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003053861584754263,
      "loss": 2.8887,
      "theoretical_loss": 3.5601326421522987,
      "tokens_seen": 1304887296
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003053761283851555,
      "loss": 2.9427,
      "theoretical_loss": 3.5601164533297824,
      "tokens_seen": 1304952832
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030536609829488466,
      "loss": 2.7599,
      "theoretical_loss": 3.560100265547897,
      "tokens_seen": 1305018368
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003053560682046139,
      "loss": 2.6777,
      "theoretical_loss": 3.560084078806522,
      "tokens_seen": 1305083904
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.000305346038114343,
      "loss": 2.7493,
      "theoretical_loss": 3.5600678931055385,
      "tokens_seen": 1305149440
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030533600802407226,
      "loss": 2.9042,
      "theoretical_loss": 3.560051708444828,
      "tokens_seen": 1305214976
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003053259779338014,
      "loss": 2.7217,
      "theoretical_loss": 3.560035524824271,
      "tokens_seen": 1305280512
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003053159478435306,
      "loss": 3.0881,
      "theoretical_loss": 3.5600193422437494,
      "tokens_seen": 1305346048
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003053059177532598,
      "loss": 2.9483,
      "theoretical_loss": 3.5600031607031424,
      "tokens_seen": 1305411584
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.000305295887662989,
      "loss": 2.9021,
      "theoretical_loss": 3.5599869802023325,
      "tokens_seen": 1305477120
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030528585757271816,
      "loss": 2.9463,
      "theoretical_loss": 3.5599708007412,
      "tokens_seen": 1305542656
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030527582748244734,
      "loss": 3.0789,
      "theoretical_loss": 3.5599546223196255,
      "tokens_seen": 1305608192
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003052657973921765,
      "loss": 2.7297,
      "theoretical_loss": 3.559938444937491,
      "tokens_seen": 1305673728
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030525576730190576,
      "loss": 2.9267,
      "theoretical_loss": 3.5599222685946765,
      "tokens_seen": 1305739264
    },
    {
      "epoch": 0.4,
      "objective/train/advantage_avg": 0.4959104061126709,
      "objective/train/docs_used": 739729,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.8422152996063232,
      "objective/train/original_loss": 2.8422152996063232,
      "objective/train/theoretical_loss": 3.5599060932910644,
      "objective/train/tokens_used": 1326264800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24595364928245544,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0508413314819336,
      "objective/train/weighted_lm_loss": 2.9868273735046387,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0417574644088745,
      "theoretical_loss": 3.5599060932910644,
      "tokens_seen": 1305804800
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003052457372116349,
      "loss": 2.8228,
      "theoretical_loss": 3.5599060932910644,
      "tokens_seen": 1305804800
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003052357071213641,
      "loss": 2.9438,
      "theoretical_loss": 3.559889919026535,
      "tokens_seen": 1305870336
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030522567703109325,
      "loss": 2.8269,
      "theoretical_loss": 3.559873745800969,
      "tokens_seen": 1305935872
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003052156469408225,
      "loss": 2.9251,
      "theoretical_loss": 3.559857573614248,
      "tokens_seen": 1306001408
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030520561685055167,
      "loss": 3.0675,
      "theoretical_loss": 3.5598414024662537,
      "tokens_seen": 1306066944
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030519558676028085,
      "loss": 2.8779,
      "theoretical_loss": 3.559825232356866,
      "tokens_seen": 1306132480
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030518555667001003,
      "loss": 2.9658,
      "theoretical_loss": 3.559809063285967,
      "tokens_seen": 1306198016
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030517552657973926,
      "loss": 3.0008,
      "theoretical_loss": 3.559792895253438,
      "tokens_seen": 1306263552
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003051654964894684,
      "loss": 2.7828,
      "theoretical_loss": 3.5597767282591595,
      "tokens_seen": 1306329088
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003051554663991976,
      "loss": 2.8003,
      "theoretical_loss": 3.5597605623030133,
      "tokens_seen": 1306394624
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030514543630892675,
      "loss": 2.8172,
      "theoretical_loss": 3.559744397384881,
      "tokens_seen": 1306460160
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.000305135406218656,
      "loss": 2.8212,
      "theoretical_loss": 3.5597282335046425,
      "tokens_seen": 1306525696
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030512537612838517,
      "loss": 3.2818,
      "theoretical_loss": 3.5597120706621803,
      "tokens_seen": 1306591232
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030511534603811435,
      "loss": 2.8746,
      "theoretical_loss": 3.5596959088573756,
      "tokens_seen": 1306656768
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030510531594784353,
      "loss": 3.1057,
      "theoretical_loss": 3.55967974809011,
      "tokens_seen": 1306722304
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003050952858575727,
      "loss": 2.9186,
      "theoretical_loss": 3.559663588360264,
      "tokens_seen": 1306787840
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003050852557673019,
      "loss": 3.0048,
      "theoretical_loss": 3.559647429667719,
      "tokens_seen": 1306853376
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030507522567703113,
      "loss": 3.0919,
      "theoretical_loss": 3.559631272012357,
      "tokens_seen": 1306918912
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030506519558676026,
      "loss": 2.8717,
      "theoretical_loss": 3.5596151153940596,
      "tokens_seen": 1306984448
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003050551654964895,
      "loss": 3.0109,
      "theoretical_loss": 3.559598959812708,
      "tokens_seen": 1307049984
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003050451354062186,
      "loss": 2.9667,
      "theoretical_loss": 3.5595828052681835,
      "tokens_seen": 1307115520
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030503510531594785,
      "loss": 2.9676,
      "theoretical_loss": 3.5595666517603677,
      "tokens_seen": 1307181056
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030502507522567703,
      "loss": 3.1703,
      "theoretical_loss": 3.5595504992891414,
      "tokens_seen": 1307246592
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003050150451354062,
      "loss": 2.8991,
      "theoretical_loss": 3.5595343478543873,
      "tokens_seen": 1307312128
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003050050150451354,
      "loss": 2.9212,
      "theoretical_loss": 3.559518197455986,
      "tokens_seen": 1307377664
    },
    {
      "epoch": 0.4,
      "objective/train/advantage_avg": 0.49340760707855225,
      "objective/train/docs_used": 740313,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.916780710220337,
      "objective/train/original_loss": 2.916780948638916,
      "objective/train/theoretical_loss": 3.5595020480938198,
      "objective/train/tokens_used": 1327903200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24352940917015076,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505785942077637,
      "objective/train/weighted_lm_loss": 3.0644490718841553,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0396549701690674,
      "theoretical_loss": 3.5595020480938198,
      "tokens_seen": 1307443200
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030499498495486463,
      "loss": 2.9558,
      "theoretical_loss": 3.5595020480938198,
      "tokens_seen": 1307443200
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030498495486459376,
      "loss": 2.9455,
      "theoretical_loss": 3.5594858997677696,
      "tokens_seen": 1307508736
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.000304974924774323,
      "loss": 2.9475,
      "theoretical_loss": 3.5594697524777175,
      "tokens_seen": 1307574272
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003049648946840521,
      "loss": 2.8968,
      "theoretical_loss": 3.5594536062235447,
      "tokens_seen": 1307639808
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030495486459378136,
      "loss": 2.9839,
      "theoretical_loss": 3.5594374610051336,
      "tokens_seen": 1307705344
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030494483450351054,
      "loss": 3.1646,
      "theoretical_loss": 3.5594213168223647,
      "tokens_seen": 1307770880
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003049348044132397,
      "loss": 3.1091,
      "theoretical_loss": 3.5594051736751204,
      "tokens_seen": 1307836416
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003049247743229689,
      "loss": 2.6379,
      "theoretical_loss": 3.5593890315632826,
      "tokens_seen": 1307901952
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003049147442326981,
      "loss": 3.0906,
      "theoretical_loss": 3.5593728904867326,
      "tokens_seen": 1307967488
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030490471414242726,
      "loss": 2.7997,
      "theoretical_loss": 3.5593567504453523,
      "tokens_seen": 1308033024
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003048946840521565,
      "loss": 2.8159,
      "theoretical_loss": 3.559340611439023,
      "tokens_seen": 1308098560
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003048846539618856,
      "loss": 2.668,
      "theoretical_loss": 3.559324473467627,
      "tokens_seen": 1308164096
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030487462387161486,
      "loss": 3.2276,
      "theoretical_loss": 3.5593083365310463,
      "tokens_seen": 1308229632
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.000304864593781344,
      "loss": 2.9471,
      "theoretical_loss": 3.5592922006291623,
      "tokens_seen": 1308295168
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003048545636910732,
      "loss": 2.8407,
      "theoretical_loss": 3.559276065761856,
      "tokens_seen": 1308360704
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003048445336008024,
      "loss": 2.8382,
      "theoretical_loss": 3.5592599319290112,
      "tokens_seen": 1308426240
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003048345035105316,
      "loss": 2.9233,
      "theoretical_loss": 3.5592437991305084,
      "tokens_seen": 1308491776
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030482447342026076,
      "loss": 3.1848,
      "theoretical_loss": 3.5592276673662298,
      "tokens_seen": 1308557312
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030481444332999,
      "loss": 2.6564,
      "theoretical_loss": 3.559211536636057,
      "tokens_seen": 1308622848
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003048044132397192,
      "loss": 2.9318,
      "theoretical_loss": 3.559195406939873,
      "tokens_seen": 1308688384
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030479438314944836,
      "loss": 2.8643,
      "theoretical_loss": 3.5591792782775578,
      "tokens_seen": 1308753920
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030478435305917754,
      "loss": 2.9347,
      "theoretical_loss": 3.5591631506489954,
      "tokens_seen": 1308819456
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003047743229689067,
      "loss": 3.0208,
      "theoretical_loss": 3.5591470240540666,
      "tokens_seen": 1308884992
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030476429287863596,
      "loss": 2.7535,
      "theoretical_loss": 3.5591308984926537,
      "tokens_seen": 1308950528
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003047542627883651,
      "loss": 2.741,
      "theoretical_loss": 3.5591147739646387,
      "tokens_seen": 1309016064
    },
    {
      "epoch": 0.4,
      "objective/train/advantage_avg": 0.4944700598716736,
      "objective/train/docs_used": 741446,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.1900689601898193,
      "objective/train/original_loss": 3.190068483352661,
      "objective/train/theoretical_loss": 3.5590986504699043,
      "objective/train/tokens_used": 1329541600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24455223977565765,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506901741027832,
      "objective/train/weighted_lm_loss": 3.3516974449157715,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0432155132293701,
      "theoretical_loss": 3.5590986504699043,
      "tokens_seen": 1309081600
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003047442326980943,
      "loss": 3.1121,
      "theoretical_loss": 3.5590986504699043,
      "tokens_seen": 1309081600
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030473420260782345,
      "loss": 3.0215,
      "theoretical_loss": 3.5590825280083314,
      "tokens_seen": 1309147136
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003047241725175527,
      "loss": 2.7226,
      "theoretical_loss": 3.5590664065798023,
      "tokens_seen": 1309212672
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030471414242728187,
      "loss": 2.7988,
      "theoretical_loss": 3.5590502861842,
      "tokens_seen": 1309278208
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030470411233701105,
      "loss": 2.6655,
      "theoretical_loss": 3.559034166821406,
      "tokens_seen": 1309343744
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030469408224674023,
      "loss": 3.1259,
      "theoretical_loss": 3.5590180484913025,
      "tokens_seen": 1309409280
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030468405215646946,
      "loss": 2.8663,
      "theoretical_loss": 3.5590019311937713,
      "tokens_seen": 1309474816
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003046740220661986,
      "loss": 3.046,
      "theoretical_loss": 3.5589858149286955,
      "tokens_seen": 1309540352
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003046639919759278,
      "loss": 2.9421,
      "theoretical_loss": 3.5589696996959566,
      "tokens_seen": 1309605888
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030465396188565695,
      "loss": 3.0282,
      "theoretical_loss": 3.5589535854954364,
      "tokens_seen": 1309671424
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003046439317953862,
      "loss": 3.0503,
      "theoretical_loss": 3.558937472327018,
      "tokens_seen": 1309736960
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030463390170511537,
      "loss": 2.8707,
      "theoretical_loss": 3.558921360190584,
      "tokens_seen": 1309802496
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030462387161484455,
      "loss": 3.0646,
      "theoretical_loss": 3.558905249086015,
      "tokens_seen": 1309868032
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030461384152457373,
      "loss": 2.8704,
      "theoretical_loss": 3.5588891390131945,
      "tokens_seen": 1309933568
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003046038114343029,
      "loss": 2.659,
      "theoretical_loss": 3.558873029972005,
      "tokens_seen": 1309999104
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003045937813440321,
      "loss": 2.8338,
      "theoretical_loss": 3.5588569219623283,
      "tokens_seen": 1310064640
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030458375125376133,
      "loss": 3.0443,
      "theoretical_loss": 3.5588408149840474,
      "tokens_seen": 1310130176
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030457372116349046,
      "loss": 2.8807,
      "theoretical_loss": 3.5588247090370437,
      "tokens_seen": 1310195712
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003045636910732197,
      "loss": 3.102,
      "theoretical_loss": 3.5588086041212,
      "tokens_seen": 1310261248
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003045536609829488,
      "loss": 2.7183,
      "theoretical_loss": 3.558792500236399,
      "tokens_seen": 1310326784
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030454363089267805,
      "loss": 3.0526,
      "theoretical_loss": 3.5587763973825224,
      "tokens_seen": 1310392320
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030453360080240723,
      "loss": 3.0582,
      "theoretical_loss": 3.5587602955594537,
      "tokens_seen": 1310457856
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003045235707121364,
      "loss": 2.6334,
      "theoretical_loss": 3.558744194767075,
      "tokens_seen": 1310523392
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003045135406218656,
      "loss": 2.8252,
      "theoretical_loss": 3.558728095005268,
      "tokens_seen": 1310588928
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030450351053159483,
      "loss": 2.8271,
      "theoretical_loss": 3.5587119962739164,
      "tokens_seen": 1310654464
    },
    {
      "debugging/Self-BLEU-5": 0.3406839562282708,
      "debugging/distinct-1-grams": 0.809715869747467,
      "debugging/distinct-2-grams": 0.978648923303283,
      "debugging/entropy-1-grams": 5.376185509300293,
      "debugging/entropy-2-grams": 5.999122918373571,
      "debugging/length": 409.2857142857143,
      "debugging/num_segments": 7,
      "debugging/raw_token_scores_avg": 0.00914314016699791,
      "debugging/raw_token_scores_std": 0.015953432768583298,
      "debugging/score": 0.008424260124251317,
      "debugging/score_std": 0.009798326703924899,
      "epoch": 0.4,
      "objective/train/advantage_avg": 0.4908560812473297,
      "objective/train/docs_used": 742066,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.0459959506988525,
      "objective/train/original_loss": 3.0459959506988525,
      "objective/train/theoretical_loss": 3.5586958985729016,
      "objective/train/tokens_used": 1331180000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24119439721107483,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.05031156539917,
      "objective/train/weighted_lm_loss": 3.1993701457977295,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 3.5586958985729016,
      "tokens_seen": 1310720000
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030449348044132396,
      "loss": 2.9979,
      "theoretical_loss": 3.5586958985729016,
      "tokens_seen": 1310720000
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003044834503510532,
      "loss": 2.7406,
      "theoretical_loss": 3.558679801902107,
      "tokens_seen": 1310785536
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003044734202607823,
      "loss": 2.8924,
      "theoretical_loss": 3.558663706261415,
      "tokens_seen": 1310851072
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030446339017051156,
      "loss": 3.1604,
      "theoretical_loss": 3.558647611650708,
      "tokens_seen": 1310916608
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030445336008024074,
      "loss": 2.8368,
      "theoretical_loss": 3.558631518069869,
      "tokens_seen": 1310982144
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003044433299899699,
      "loss": 3.0039,
      "theoretical_loss": 3.5586154255187794,
      "tokens_seen": 1311047680
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003044332998996991,
      "loss": 2.9641,
      "theoretical_loss": 3.558599333997323,
      "tokens_seen": 1311113216
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003044232698094283,
      "loss": 3.0507,
      "theoretical_loss": 3.5585832435053826,
      "tokens_seen": 1311178752
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030441323971915746,
      "loss": 2.9595,
      "theoretical_loss": 3.55856715404284,
      "tokens_seen": 1311244288
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003044032096288867,
      "loss": 2.732,
      "theoretical_loss": 3.5585510656095787,
      "tokens_seen": 1311309824
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003043931795386158,
      "loss": 3.0259,
      "theoretical_loss": 3.558534978205481,
      "tokens_seen": 1311375360
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030438314944834506,
      "loss": 2.9058,
      "theoretical_loss": 3.5585188918304294,
      "tokens_seen": 1311440896
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003043731193580742,
      "loss": 2.7384,
      "theoretical_loss": 3.558502806484307,
      "tokens_seen": 1311506432
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003043630892678034,
      "loss": 2.9939,
      "theoretical_loss": 3.5584867221669967,
      "tokens_seen": 1311571968
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003043530591775326,
      "loss": 2.9001,
      "theoretical_loss": 3.558470638878381,
      "tokens_seen": 1311637504
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003043430290872618,
      "loss": 2.7002,
      "theoretical_loss": 3.5584545566183436,
      "tokens_seen": 1311703040
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030433299899699097,
      "loss": 2.8745,
      "theoretical_loss": 3.558438475386766,
      "tokens_seen": 1311768576
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003043229689067202,
      "loss": 2.8749,
      "theoretical_loss": 3.5584223951835314,
      "tokens_seen": 1311834112
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030431293881644933,
      "loss": 2.8828,
      "theoretical_loss": 3.558406316008523,
      "tokens_seen": 1311899648
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030430290872617856,
      "loss": 3.0321,
      "theoretical_loss": 3.558390237861624,
      "tokens_seen": 1311965184
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003042928786359077,
      "loss": 2.8026,
      "theoretical_loss": 3.5583741607427166,
      "tokens_seen": 1312030720
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003042828485456369,
      "loss": 2.7595,
      "theoretical_loss": 3.558358084651684,
      "tokens_seen": 1312096256
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003042728184553661,
      "loss": 3.007,
      "theoretical_loss": 3.5583420095884097,
      "tokens_seen": 1312161792
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003042627883650953,
      "loss": 3.279,
      "theoretical_loss": 3.5583259355527757,
      "tokens_seen": 1312227328
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030425275827482447,
      "loss": 3.0328,
      "theoretical_loss": 3.5583098625446654,
      "tokens_seen": 1312292864
    },
    {
      "epoch": 0.4,
      "objective/train/advantage_avg": 0.49451467394828796,
      "objective/train/docs_used": 743478,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.6003925800323486,
      "objective/train/original_loss": 2.6003923416137695,
      "objective/train/theoretical_loss": 3.558293790563962,
      "objective/train/tokens_used": 1332818400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24460865557193756,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506948232650757,
      "objective/train/weighted_lm_loss": 2.732067346572876,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0417574644088745,
      "theoretical_loss": 3.558293790563962,
      "tokens_seen": 1312358400
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030424272818455365,
      "loss": 2.7155,
      "theoretical_loss": 3.558293790563962,
      "tokens_seen": 1312358400
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030423269809428283,
      "loss": 2.7988,
      "theoretical_loss": 3.5582777196105484,
      "tokens_seen": 1312423936
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030422266800401207,
      "loss": 2.9254,
      "theoretical_loss": 3.5582616496843076,
      "tokens_seen": 1312489472
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003042126379137412,
      "loss": 3.0304,
      "theoretical_loss": 3.5582455807851225,
      "tokens_seen": 1312555008
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030420260782347043,
      "loss": 2.736,
      "theoretical_loss": 3.5582295129128765,
      "tokens_seen": 1312620544
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003041925777331996,
      "loss": 2.5803,
      "theoretical_loss": 3.5582134460674526,
      "tokens_seen": 1312686080
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003041825476429288,
      "loss": 2.91,
      "theoretical_loss": 3.558197380248734,
      "tokens_seen": 1312751616
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030417251755265797,
      "loss": 2.8692,
      "theoretical_loss": 3.5581813154566038,
      "tokens_seen": 1312817152
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030416248746238715,
      "loss": 2.8057,
      "theoretical_loss": 3.5581652516909448,
      "tokens_seen": 1312882688
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030415245737211633,
      "loss": 3.0923,
      "theoretical_loss": 3.558149188951641,
      "tokens_seen": 1312948224
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030414242728184557,
      "loss": 2.7201,
      "theoretical_loss": 3.5581331272385746,
      "tokens_seen": 1313013760
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003041323971915747,
      "loss": 2.6182,
      "theoretical_loss": 3.5581170665516293,
      "tokens_seen": 1313079296
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030412236710130393,
      "loss": 2.9143,
      "theoretical_loss": 3.558101006890688,
      "tokens_seen": 1313144832
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030411233701103306,
      "loss": 2.907,
      "theoretical_loss": 3.558084948255635,
      "tokens_seen": 1313210368
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003041023069207623,
      "loss": 2.8613,
      "theoretical_loss": 3.5580688906463527,
      "tokens_seen": 1313275904
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003040922768304915,
      "loss": 2.8822,
      "theoretical_loss": 3.558052834062724,
      "tokens_seen": 1313341440
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030408224674022066,
      "loss": 2.9315,
      "theoretical_loss": 3.558036778504633,
      "tokens_seen": 1313406976
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030407221664994984,
      "loss": 2.8476,
      "theoretical_loss": 3.5580207239719632,
      "tokens_seen": 1313472512
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.000304062186559679,
      "loss": 3.0926,
      "theoretical_loss": 3.558004670464597,
      "tokens_seen": 1313538048
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030405215646940825,
      "loss": 2.8458,
      "theoretical_loss": 3.557988617982419,
      "tokens_seen": 1313603584
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030404212637913743,
      "loss": 2.9322,
      "theoretical_loss": 3.5579725665253115,
      "tokens_seen": 1313669120
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003040320962888666,
      "loss": 2.85,
      "theoretical_loss": 3.5579565160931583,
      "tokens_seen": 1313734656
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003040220661985958,
      "loss": 2.8107,
      "theoretical_loss": 3.5579404666858427,
      "tokens_seen": 1313800192
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030401203610832503,
      "loss": 2.8914,
      "theoretical_loss": 3.5579244183032483,
      "tokens_seen": 1313865728
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030400200601805416,
      "loss": 2.8473,
      "theoretical_loss": 3.557908370945259,
      "tokens_seen": 1313931264
    },
    {
      "epoch": 0.4,
      "objective/train/advantage_avg": 0.4943666160106659,
      "objective/train/docs_used": 744202,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.019697666168213,
      "objective/train/original_loss": 3.019697666168213,
      "objective/train/theoretical_loss": 3.5578923246117578,
      "objective/train/tokens_used": 1334456800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24445119500160217,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506792068481445,
      "objective/train/weighted_lm_loss": 3.172801971435547,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0408108234405518,
      "theoretical_loss": 3.5578923246117578,
      "tokens_seen": 1313996800
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003039919759277834,
      "loss": 3.0132,
      "theoretical_loss": 3.5578923246117578,
      "tokens_seen": 1313996800
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003039819458375125,
      "loss": 2.9366,
      "theoretical_loss": 3.5578762793026275,
      "tokens_seen": 1314062336
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030397191574724176,
      "loss": 2.8951,
      "theoretical_loss": 3.5578602350177526,
      "tokens_seen": 1314127872
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030396188565697094,
      "loss": 2.692,
      "theoretical_loss": 3.5578441917570167,
      "tokens_seen": 1314193408
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003039518555667001,
      "loss": 2.724,
      "theoretical_loss": 3.557828149520303,
      "tokens_seen": 1314258944
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003039418254764293,
      "loss": 3.1574,
      "theoretical_loss": 3.557812108307495,
      "tokens_seen": 1314324480
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003039317953861585,
      "loss": 2.8789,
      "theoretical_loss": 3.557796068118477,
      "tokens_seen": 1314390016
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030392176529588766,
      "loss": 3.0779,
      "theoretical_loss": 3.557780028953131,
      "tokens_seen": 1314455552
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003039117352056169,
      "loss": 2.6358,
      "theoretical_loss": 3.5577639908113428,
      "tokens_seen": 1314521088
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.000303901705115346,
      "loss": 2.9693,
      "theoretical_loss": 3.557747953692994,
      "tokens_seen": 1314586624
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030389167502507526,
      "loss": 3.0925,
      "theoretical_loss": 3.5577319175979696,
      "tokens_seen": 1314652160
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003038816449348044,
      "loss": 2.9862,
      "theoretical_loss": 3.5577158825261535,
      "tokens_seen": 1314717696
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003038716148445336,
      "loss": 2.912,
      "theoretical_loss": 3.5576998484774283,
      "tokens_seen": 1314783232
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003038615847542628,
      "loss": 3.0236,
      "theoretical_loss": 3.5576838154516777,
      "tokens_seen": 1314848768
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.000303851554663992,
      "loss": 2.8653,
      "theoretical_loss": 3.557667783448787,
      "tokens_seen": 1314914304
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030384152457372117,
      "loss": 2.908,
      "theoretical_loss": 3.5576517524686384,
      "tokens_seen": 1314979840
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003038314944834504,
      "loss": 2.9662,
      "theoretical_loss": 3.557635722511116,
      "tokens_seen": 1315045376
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030382146439317953,
      "loss": 2.9251,
      "theoretical_loss": 3.557619693576105,
      "tokens_seen": 1315110912
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030381143430290876,
      "loss": 2.9235,
      "theoretical_loss": 3.5576036656634873,
      "tokens_seen": 1315176448
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003038014042126379,
      "loss": 2.7088,
      "theoretical_loss": 3.557587638773147,
      "tokens_seen": 1315241984
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003037913741223671,
      "loss": 2.7774,
      "theoretical_loss": 3.5575716129049697,
      "tokens_seen": 1315307520
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003037813440320963,
      "loss": 2.9451,
      "theoretical_loss": 3.5575555880588374,
      "tokens_seen": 1315373056
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003037713139418255,
      "loss": 2.8058,
      "theoretical_loss": 3.557539564234635,
      "tokens_seen": 1315438592
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030376128385155467,
      "loss": 2.9258,
      "theoretical_loss": 3.557523541432246,
      "tokens_seen": 1315504128
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030375125376128385,
      "loss": 3.105,
      "theoretical_loss": 3.5575075196515544,
      "tokens_seen": 1315569664
    },
    {
      "epoch": 0.4,
      "objective/train/advantage_avg": 0.4953295588493347,
      "objective/train/docs_used": 745555,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.491119146347046,
      "objective/train/original_loss": 2.491119146347046,
      "objective/train/theoretical_loss": 3.5574914988924444,
      "objective/train/tokens_used": 1336095200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24539652466773987,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507804155349731,
      "objective/train/weighted_lm_loss": 2.617626428604126,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0442860126495361,
      "theoretical_loss": 3.5574914988924444,
      "tokens_seen": 1315635200
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030374122367101303,
      "loss": 2.7252,
      "theoretical_loss": 3.5574914988924444,
      "tokens_seen": 1315635200
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030373119358074227,
      "loss": 2.8309,
      "theoretical_loss": 3.5574754791547996,
      "tokens_seen": 1315700736
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003037211634904714,
      "loss": 2.8263,
      "theoretical_loss": 3.557459460438504,
      "tokens_seen": 1315766272
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030371113340020063,
      "loss": 2.8766,
      "theoretical_loss": 3.557443442743442,
      "tokens_seen": 1315831808
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003037011033099298,
      "loss": 2.7736,
      "theoretical_loss": 3.5574274260694976,
      "tokens_seen": 1315897344
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.000303691073219659,
      "loss": 2.9458,
      "theoretical_loss": 3.5574114104165546,
      "tokens_seen": 1315962880
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030368104312938817,
      "loss": 2.5352,
      "theoretical_loss": 3.5573953957844973,
      "tokens_seen": 1316028416
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030367101303911735,
      "loss": 2.9175,
      "theoretical_loss": 3.5573793821732096,
      "tokens_seen": 1316093952
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030366098294884653,
      "loss": 2.6652,
      "theoretical_loss": 3.5573633695825757,
      "tokens_seen": 1316159488
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030365095285857577,
      "loss": 3.0214,
      "theoretical_loss": 3.5573473580124797,
      "tokens_seen": 1316225024
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003036409227683049,
      "loss": 2.8698,
      "theoretical_loss": 3.5573313474628057,
      "tokens_seen": 1316290560
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030363089267803413,
      "loss": 2.8679,
      "theoretical_loss": 3.5573153379334377,
      "tokens_seen": 1316356096
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030362086258776326,
      "loss": 2.774,
      "theoretical_loss": 3.5572993294242607,
      "tokens_seen": 1316421632
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003036108324974925,
      "loss": 2.9652,
      "theoretical_loss": 3.557283321935158,
      "tokens_seen": 1316487168
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003036008024072217,
      "loss": 2.7378,
      "theoretical_loss": 3.557267315466014,
      "tokens_seen": 1316552704
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030359077231695086,
      "loss": 2.9294,
      "theoretical_loss": 3.5572513100167136,
      "tokens_seen": 1316618240
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030358074222668004,
      "loss": 2.7818,
      "theoretical_loss": 3.55723530558714,
      "tokens_seen": 1316683776
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003035707121364092,
      "loss": 2.7375,
      "theoretical_loss": 3.5572193021771783,
      "tokens_seen": 1316749312
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003035606820461384,
      "loss": 2.9986,
      "theoretical_loss": 3.5572032997867122,
      "tokens_seen": 1316814848
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030355065195586763,
      "loss": 2.9496,
      "theoretical_loss": 3.5571872984156268,
      "tokens_seen": 1316880384
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030354062186559676,
      "loss": 2.9725,
      "theoretical_loss": 3.557171298063806,
      "tokens_seen": 1316945920
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.000303530591775326,
      "loss": 2.9432,
      "theoretical_loss": 3.557155298731134,
      "tokens_seen": 1317011456
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003035205616850552,
      "loss": 3.0359,
      "theoretical_loss": 3.5571393004174947,
      "tokens_seen": 1317076992
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030351053159478436,
      "loss": 3.2399,
      "theoretical_loss": 3.557123303122774,
      "tokens_seen": 1317142528
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030350050150451354,
      "loss": 3.2213,
      "theoretical_loss": 3.557107306846855,
      "tokens_seen": 1317208064
    },
    {
      "epoch": 0.4,
      "objective/train/advantage_avg": 0.4951379597187042,
      "objective/train/docs_used": 746236,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.062016010284424,
      "objective/train/original_loss": 3.0620157718658447,
      "objective/train/theoretical_loss": 3.5570913115896228,
      "objective/train/tokens_used": 1337733600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24521109461784363,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050760269165039,
      "objective/train/weighted_lm_loss": 3.217428684234619,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0432155132293701,
      "theoretical_loss": 3.5570913115896228,
      "tokens_seen": 1317273600
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003034904714142427,
      "loss": 3.0154,
      "theoretical_loss": 3.5570913115896228,
      "tokens_seen": 1317273600
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003034804413239719,
      "loss": 2.9306,
      "theoretical_loss": 3.5570753173509617,
      "tokens_seen": 1317339136
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030347041123370114,
      "loss": 2.7394,
      "theoretical_loss": 3.5570593241307558,
      "tokens_seen": 1317404672
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030346038114343026,
      "loss": 2.649,
      "theoretical_loss": 3.55704333192889,
      "tokens_seen": 1317470208
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003034503510531595,
      "loss": 2.794,
      "theoretical_loss": 3.5570273407452486,
      "tokens_seen": 1317535744
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003034403209628886,
      "loss": 2.9163,
      "theoretical_loss": 3.557011350579716,
      "tokens_seen": 1317601280
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030343029087261786,
      "loss": 2.7982,
      "theoretical_loss": 3.556995361432177,
      "tokens_seen": 1317666816
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030342026078234704,
      "loss": 2.8739,
      "theoretical_loss": 3.5569793733025166,
      "tokens_seen": 1317732352
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003034102306920762,
      "loss": 2.7086,
      "theoretical_loss": 3.5569633861906187,
      "tokens_seen": 1317797888
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003034002006018054,
      "loss": 2.9174,
      "theoretical_loss": 3.5569474000963677,
      "tokens_seen": 1317863424
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003033901705115346,
      "loss": 3.0333,
      "theoretical_loss": 3.556931415019649,
      "tokens_seen": 1317928960
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030338014042126377,
      "loss": 2.7609,
      "theoretical_loss": 3.5569154309603466,
      "tokens_seen": 1317994496
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.000303370110330993,
      "loss": 2.9811,
      "theoretical_loss": 3.5568994479183456,
      "tokens_seen": 1318060032
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030336008024072213,
      "loss": 3.0442,
      "theoretical_loss": 3.556883465893531,
      "tokens_seen": 1318125568
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030335005015045137,
      "loss": 2.8679,
      "theoretical_loss": 3.556867484885786,
      "tokens_seen": 1318191104
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030334002006018055,
      "loss": 2.9547,
      "theoretical_loss": 3.556851504894997,
      "tokens_seen": 1318256640
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030332998996990973,
      "loss": 2.7334,
      "theoretical_loss": 3.556835525921048,
      "tokens_seen": 1318322176
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003033199598796389,
      "loss": 2.8497,
      "theoretical_loss": 3.5568195479638236,
      "tokens_seen": 1318387712
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003033099297893681,
      "loss": 2.7482,
      "theoretical_loss": 3.5568035710232087,
      "tokens_seen": 1318453248
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003032998996990973,
      "loss": 2.9528,
      "theoretical_loss": 3.556787595099088,
      "tokens_seen": 1318518784
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003032898696088265,
      "loss": 2.7134,
      "theoretical_loss": 3.556771620191347,
      "tokens_seen": 1318584320
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003032798395185557,
      "loss": 2.8358,
      "theoretical_loss": 3.55675564629987,
      "tokens_seen": 1318649856
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030326980942828487,
      "loss": 2.9264,
      "theoretical_loss": 3.5567396734245413,
      "tokens_seen": 1318715392
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030325977933801405,
      "loss": 2.7948,
      "theoretical_loss": 3.5567237015652466,
      "tokens_seen": 1318780928
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030324974924774323,
      "loss": 2.5629,
      "theoretical_loss": 3.556707730721871,
      "tokens_seen": 1318846464
    },
    {
      "epoch": 0.4,
      "objective/train/advantage_avg": 0.49318379163742065,
      "objective/train/docs_used": 747487,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.653420925140381,
      "objective/train/original_loss": 2.653420925140381,
      "objective/train/theoretical_loss": 3.5566917608942976,
      "objective/train/tokens_used": 1339372000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24333502352237701,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505553483963013,
      "objective/train/weighted_lm_loss": 2.787405014038086,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 3.5566917608942976,
      "tokens_seen": 1318912000
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030323971915747247,
      "loss": 2.6388,
      "theoretical_loss": 3.5566917608942976,
      "tokens_seen": 1318912000
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003032296890672016,
      "loss": 2.7751,
      "theoretical_loss": 3.5566757920824132,
      "tokens_seen": 1318977536
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030321965897693083,
      "loss": 2.7466,
      "theoretical_loss": 3.5566598242861027,
      "tokens_seen": 1319043072
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030320962888666,
      "loss": 2.6024,
      "theoretical_loss": 3.55664385750525,
      "tokens_seen": 1319108608
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003031995987963892,
      "loss": 2.7504,
      "theoretical_loss": 3.5566278917397405,
      "tokens_seen": 1319174144
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030318956870611837,
      "loss": 2.8788,
      "theoretical_loss": 3.5566119269894596,
      "tokens_seen": 1319239680
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030317953861584755,
      "loss": 2.9419,
      "theoretical_loss": 3.556595963254292,
      "tokens_seen": 1319305216
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030316950852557673,
      "loss": 2.8988,
      "theoretical_loss": 3.5565800005341224,
      "tokens_seen": 1319370752
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030315947843530597,
      "loss": 2.6996,
      "theoretical_loss": 3.556564038828837,
      "tokens_seen": 1319436288
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003031494483450351,
      "loss": 2.8217,
      "theoretical_loss": 3.556548078138319,
      "tokens_seen": 1319501824
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030313941825476433,
      "loss": 2.8246,
      "theoretical_loss": 3.5565321184624548,
      "tokens_seen": 1319567360
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030312938816449346,
      "loss": 3.0003,
      "theoretical_loss": 3.55651615980113,
      "tokens_seen": 1319632896
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003031193580742227,
      "loss": 2.7507,
      "theoretical_loss": 3.5565002021542282,
      "tokens_seen": 1319698432
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003031093279839519,
      "loss": 3.0523,
      "theoretical_loss": 3.5564842455216352,
      "tokens_seen": 1319763968
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030309929789368106,
      "loss": 3.0275,
      "theoretical_loss": 3.556468289903237,
      "tokens_seen": 1319829504
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030308926780341024,
      "loss": 2.9339,
      "theoretical_loss": 3.5564523352989177,
      "tokens_seen": 1319895040
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003030792377131394,
      "loss": 2.97,
      "theoretical_loss": 3.556436381708563,
      "tokens_seen": 1319960576
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003030692076228686,
      "loss": 3.0415,
      "theoretical_loss": 3.556420429132058,
      "tokens_seen": 1320026112
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030305917753259783,
      "loss": 2.7642,
      "theoretical_loss": 3.556404477569288,
      "tokens_seen": 1320091648
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030304914744232696,
      "loss": 2.7147,
      "theoretical_loss": 3.556388527020138,
      "tokens_seen": 1320157184
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003030391173520562,
      "loss": 2.8898,
      "theoretical_loss": 3.556372577484493,
      "tokens_seen": 1320222720
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003030290872617854,
      "loss": 2.771,
      "theoretical_loss": 3.5563566289622397,
      "tokens_seen": 1320288256
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030301905717151456,
      "loss": 2.6932,
      "theoretical_loss": 3.556340681453262,
      "tokens_seen": 1320353792
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030300902708124374,
      "loss": 2.6104,
      "theoretical_loss": 3.5563247349574456,
      "tokens_seen": 1320419328
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003029989969909729,
      "loss": 2.8443,
      "theoretical_loss": 3.5563087894746763,
      "tokens_seen": 1320484864
    },
    {
      "epoch": 0.4,
      "objective/train/advantage_avg": 0.4911225140094757,
      "objective/train/docs_used": 748180,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.773440361022949,
      "objective/train/original_loss": 2.77344012260437,
      "objective/train/theoretical_loss": 3.5562928450048386,
      "objective/train/tokens_used": 1341010400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.241318479180336,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0503387451171875,
      "objective/train/weighted_lm_loss": 2.9126834869384766,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0408108234405518,
      "theoretical_loss": 3.5562928450048386,
      "tokens_seen": 1320550400
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003029889669007021,
      "loss": 2.7279,
      "theoretical_loss": 3.5562928450048386,
      "tokens_seen": 1320550400
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030297893681043134,
      "loss": 2.7141,
      "theoretical_loss": 3.556276901547819,
      "tokens_seen": 1320615936
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030296890672016046,
      "loss": 2.7588,
      "theoretical_loss": 3.5562609591035024,
      "tokens_seen": 1320681472
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003029588766298897,
      "loss": 2.9805,
      "theoretical_loss": 3.556245017671774,
      "tokens_seen": 1320747008
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003029488465396188,
      "loss": 2.9411,
      "theoretical_loss": 3.556229077252519,
      "tokens_seen": 1320812544
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030293881644934806,
      "loss": 2.7302,
      "theoretical_loss": 3.556213137845624,
      "tokens_seen": 1320878080
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030292878635907724,
      "loss": 2.9116,
      "theoretical_loss": 3.556197199450973,
      "tokens_seen": 1320943616
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003029187562688064,
      "loss": 2.9181,
      "theoretical_loss": 3.5561812620684528,
      "tokens_seen": 1321009152
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003029087261785356,
      "loss": 2.9902,
      "theoretical_loss": 3.5561653256979477,
      "tokens_seen": 1321074688
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003028986960882648,
      "loss": 2.8371,
      "theoretical_loss": 3.556149390339345,
      "tokens_seen": 1321140224
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030288866599799397,
      "loss": 3.2292,
      "theoretical_loss": 3.556133455992528,
      "tokens_seen": 1321205760
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003028786359077232,
      "loss": 2.9402,
      "theoretical_loss": 3.5561175226573845,
      "tokens_seen": 1321271296
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030286860581745233,
      "loss": 2.8438,
      "theoretical_loss": 3.5561015903337987,
      "tokens_seen": 1321336832
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030285857572718157,
      "loss": 2.9187,
      "theoretical_loss": 3.5560856590216563,
      "tokens_seen": 1321402368
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030284854563691075,
      "loss": 2.6576,
      "theoretical_loss": 3.5560697287208427,
      "tokens_seen": 1321467904
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030283851554663993,
      "loss": 2.786,
      "theoretical_loss": 3.556053799431245,
      "tokens_seen": 1321533440
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003028284854563691,
      "loss": 2.9495,
      "theoretical_loss": 3.556037871152747,
      "tokens_seen": 1321598976
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003028184553660983,
      "loss": 3.0332,
      "theoretical_loss": 3.556021943885236,
      "tokens_seen": 1321664512
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030280842527582747,
      "loss": 3.0271,
      "theoretical_loss": 3.556006017628597,
      "tokens_seen": 1321730048
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003027983951855567,
      "loss": 3.0598,
      "theoretical_loss": 3.5559900923827152,
      "tokens_seen": 1321795584
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030278836509528583,
      "loss": 2.9292,
      "theoretical_loss": 3.5559741681474772,
      "tokens_seen": 1321861120
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030277833500501507,
      "loss": 2.7156,
      "theoretical_loss": 3.5559582449227687,
      "tokens_seen": 1321926656
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003027683049147442,
      "loss": 2.9441,
      "theoretical_loss": 3.555942322708475,
      "tokens_seen": 1321992192
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030275827482447343,
      "loss": 2.6426,
      "theoretical_loss": 3.5559264015044816,
      "tokens_seen": 1322057728
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003027482447342026,
      "loss": 2.9632,
      "theoretical_loss": 3.5559104813106757,
      "tokens_seen": 1322123264
    },
    {
      "epoch": 0.4,
      "objective/train/advantage_avg": 0.4953859746456146,
      "objective/train/docs_used": 749233,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.74743390083313,
      "objective/train/original_loss": 2.747433662414551,
      "objective/train/theoretical_loss": 3.5558945621269413,
      "objective/train/tokens_used": 1342648800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24544572830200195,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507862567901611,
      "objective/train/weighted_lm_loss": 2.886876344680786,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0437887907028198,
      "theoretical_loss": 3.5558945621269413,
      "tokens_seen": 1322188800
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003027382146439318,
      "loss": 2.9411,
      "theoretical_loss": 3.5558945621269413,
      "tokens_seen": 1322188800
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.000302728184553661,
      "loss": 2.8972,
      "theoretical_loss": 3.5558786439531653,
      "tokens_seen": 1322254336
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003027181544633902,
      "loss": 2.8125,
      "theoretical_loss": 3.5558627267892344,
      "tokens_seen": 1322319872
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030270812437311934,
      "loss": 2.8009,
      "theoretical_loss": 3.5558468106350327,
      "tokens_seen": 1322385408
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030269809428284857,
      "loss": 2.9212,
      "theoretical_loss": 3.5558308954904474,
      "tokens_seen": 1322450944
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003026880641925777,
      "loss": 2.7085,
      "theoretical_loss": 3.555814981355364,
      "tokens_seen": 1322516480
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030267803410230693,
      "loss": 2.7367,
      "theoretical_loss": 3.555799068229668,
      "tokens_seen": 1322582016
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003026680040120361,
      "loss": 2.869,
      "theoretical_loss": 3.5557831561132462,
      "tokens_seen": 1322647552
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003026579739217653,
      "loss": 2.9436,
      "theoretical_loss": 3.5557672450059843,
      "tokens_seen": 1322713088
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003026479438314945,
      "loss": 2.9746,
      "theoretical_loss": 3.555751334907768,
      "tokens_seen": 1322778624
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030263791374122366,
      "loss": 2.8538,
      "theoretical_loss": 3.555735425818484,
      "tokens_seen": 1322844160
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030262788365095284,
      "loss": 2.7003,
      "theoretical_loss": 3.555719517738017,
      "tokens_seen": 1322909696
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003026178535606821,
      "loss": 2.981,
      "theoretical_loss": 3.5557036106662547,
      "tokens_seen": 1322975232
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003026078234704112,
      "loss": 2.738,
      "theoretical_loss": 3.555687704603082,
      "tokens_seen": 1323040768
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030259779338014044,
      "loss": 2.9315,
      "theoretical_loss": 3.5556717995483855,
      "tokens_seen": 1323106304
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030258776328986956,
      "loss": 3.1324,
      "theoretical_loss": 3.555655895502051,
      "tokens_seen": 1323171840
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003025777331995988,
      "loss": 2.8663,
      "theoretical_loss": 3.555639992463965,
      "tokens_seen": 1323237376
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.000302567703109328,
      "loss": 2.5923,
      "theoretical_loss": 3.555624090434014,
      "tokens_seen": 1323302912
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030255767301905716,
      "loss": 2.6454,
      "theoretical_loss": 3.555608189412083,
      "tokens_seen": 1323368448
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003025476429287864,
      "loss": 3.0122,
      "theoretical_loss": 3.555592289398059,
      "tokens_seen": 1323433984
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003025376128385156,
      "loss": 3.1495,
      "theoretical_loss": 3.555576390391828,
      "tokens_seen": 1323499520
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030252758274824476,
      "loss": 2.9967,
      "theoretical_loss": 3.555560492393276,
      "tokens_seen": 1323565056
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030251755265797394,
      "loss": 2.7763,
      "theoretical_loss": 3.5555445954022904,
      "tokens_seen": 1323630592
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003025075225677031,
      "loss": 2.9306,
      "theoretical_loss": 3.555528699418756,
      "tokens_seen": 1323696128
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003024974924774323,
      "loss": 3.0874,
      "theoretical_loss": 3.55551280444256,
      "tokens_seen": 1323761664
    },
    {
      "epoch": 0.4,
      "objective/train/advantage_avg": 0.49440979957580566,
      "objective/train/docs_used": 749806,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.68034029006958,
      "objective/train/original_loss": 2.68034029006958,
      "objective/train/theoretical_loss": 3.555496910473588,
      "objective/train/tokens_used": 1344287200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24452422559261322,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050683856010437,
      "objective/train/weighted_lm_loss": 2.8160133361816406,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0304546356201172,
      "theoretical_loss": 3.555496910473588,
      "tokens_seen": 1323827200
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030248746238716154,
      "loss": 2.8187,
      "theoretical_loss": 3.555496910473588,
      "tokens_seen": 1323827200
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030247743229689066,
      "loss": 2.909,
      "theoretical_loss": 3.5554810175117266,
      "tokens_seen": 1323892736
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003024674022066199,
      "loss": 2.7892,
      "theoretical_loss": 3.5554651255568626,
      "tokens_seen": 1323958272
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030245737211634903,
      "loss": 2.6844,
      "theoretical_loss": 3.555449234608882,
      "tokens_seen": 1324023808
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030244734202607826,
      "loss": 2.8732,
      "theoretical_loss": 3.5554333446676707,
      "tokens_seen": 1324089344
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030243731193580744,
      "loss": 3.0431,
      "theoretical_loss": 3.5554174557331155,
      "tokens_seen": 1324154880
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003024272818455366,
      "loss": 3.0356,
      "theoretical_loss": 3.555401567805103,
      "tokens_seen": 1324220416
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003024172517552658,
      "loss": 2.9014,
      "theoretical_loss": 3.5553856808835196,
      "tokens_seen": 1324285952
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.000302407221664995,
      "loss": 2.4818,
      "theoretical_loss": 3.555369794968252,
      "tokens_seen": 1324351488
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030239719157472417,
      "loss": 2.896,
      "theoretical_loss": 3.5553539100591856,
      "tokens_seen": 1324417024
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003023871614844534,
      "loss": 2.9241,
      "theoretical_loss": 3.5553380261562078,
      "tokens_seen": 1324482560
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030237713139418253,
      "loss": 2.7935,
      "theoretical_loss": 3.555322143259205,
      "tokens_seen": 1324548096
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030236710130391177,
      "loss": 3.053,
      "theoretical_loss": 3.5553062613680635,
      "tokens_seen": 1324613632
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030235707121364095,
      "loss": 2.5179,
      "theoretical_loss": 3.5552903804826697,
      "tokens_seen": 1324679168
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030234704112337013,
      "loss": 2.9773,
      "theoretical_loss": 3.555274500602911,
      "tokens_seen": 1324744704
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003023370110330993,
      "loss": 2.7759,
      "theoretical_loss": 3.5552586217286724,
      "tokens_seen": 1324810240
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003023269809428285,
      "loss": 2.5937,
      "theoretical_loss": 3.555242743859842,
      "tokens_seen": 1324875776
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030231695085255767,
      "loss": 2.9323,
      "theoretical_loss": 3.555226866996306,
      "tokens_seen": 1324941312
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003023069207622869,
      "loss": 2.7781,
      "theoretical_loss": 3.555210991137951,
      "tokens_seen": 1325006848
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030229689067201603,
      "loss": 3.0223,
      "theoretical_loss": 3.555195116284663,
      "tokens_seen": 1325072384
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030228686058174527,
      "loss": 2.6596,
      "theoretical_loss": 3.5551792424363295,
      "tokens_seen": 1325137920
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003022768304914744,
      "loss": 2.941,
      "theoretical_loss": 3.5551633695928366,
      "tokens_seen": 1325203456
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030226680040120363,
      "loss": 2.7344,
      "theoretical_loss": 3.5551474977540716,
      "tokens_seen": 1325268992
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003022567703109328,
      "loss": 2.8106,
      "theoretical_loss": 3.555131626919921,
      "tokens_seen": 1325334528
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.000302246740220662,
      "loss": 2.8836,
      "theoretical_loss": 3.555115757090271,
      "tokens_seen": 1325400064
    },
    {
      "epoch": 0.4,
      "objective/train/advantage_avg": 0.4937264323234558,
      "objective/train/docs_used": 750658,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.4140923023223877,
      "objective/train/original_loss": 2.414092540740967,
      "objective/train/theoretical_loss": 3.5550998882650093,
      "objective/train/tokens_used": 1345925600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24384434521198273,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506123304367065,
      "objective/train/weighted_lm_loss": 2.5359904766082764,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0417574644088745,
      "theoretical_loss": 3.5550998882650093,
      "tokens_seen": 1325465600
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003022367101303912,
      "loss": 2.5518,
      "theoretical_loss": 3.5550998882650093,
      "tokens_seen": 1325465600
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003022266800401204,
      "loss": 3.2376,
      "theoretical_loss": 3.5550840204440224,
      "tokens_seen": 1325531136
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030221664994984954,
      "loss": 2.6893,
      "theoretical_loss": 3.555068153627196,
      "tokens_seen": 1325596672
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030220661985957877,
      "loss": 2.7307,
      "theoretical_loss": 3.5550522878144184,
      "tokens_seen": 1325662208
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003021965897693079,
      "loss": 2.9818,
      "theoretical_loss": 3.5550364230055758,
      "tokens_seen": 1325727744
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030218655967903713,
      "loss": 2.8704,
      "theoretical_loss": 3.555020559200555,
      "tokens_seen": 1325793280
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003021765295887663,
      "loss": 2.8494,
      "theoretical_loss": 3.555004696399243,
      "tokens_seen": 1325858816
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003021664994984955,
      "loss": 2.9207,
      "theoretical_loss": 3.5549888346015264,
      "tokens_seen": 1325924352
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003021564694082247,
      "loss": 2.8431,
      "theoretical_loss": 3.554972973807293,
      "tokens_seen": 1325989888
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030214643931795386,
      "loss": 2.9152,
      "theoretical_loss": 3.554957114016428,
      "tokens_seen": 1326055424
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030213640922768304,
      "loss": 2.7617,
      "theoretical_loss": 3.5549412552288207,
      "tokens_seen": 1326120960
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003021263791374123,
      "loss": 2.775,
      "theoretical_loss": 3.5549253974443555,
      "tokens_seen": 1326186496
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003021163490471414,
      "loss": 2.9124,
      "theoretical_loss": 3.554909540662922,
      "tokens_seen": 1326252032
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030210631895687064,
      "loss": 2.9446,
      "theoretical_loss": 3.554893684884405,
      "tokens_seen": 1326317568
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030209628886659976,
      "loss": 3.0104,
      "theoretical_loss": 3.5548778301086923,
      "tokens_seen": 1326383104
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.000302086258776329,
      "loss": 2.9342,
      "theoretical_loss": 3.554861976335671,
      "tokens_seen": 1326448640
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003020762286860582,
      "loss": 2.8486,
      "theoretical_loss": 3.5548461235652287,
      "tokens_seen": 1326514176
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030206619859578736,
      "loss": 2.9823,
      "theoretical_loss": 3.5548302717972513,
      "tokens_seen": 1326579712
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030205616850551654,
      "loss": 2.9811,
      "theoretical_loss": 3.554814421031627,
      "tokens_seen": 1326645248
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003020461384152458,
      "loss": 2.9924,
      "theoretical_loss": 3.554798571268242,
      "tokens_seen": 1326710784
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003020361083249749,
      "loss": 2.4893,
      "theoretical_loss": 3.554782722506984,
      "tokens_seen": 1326776320
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030202607823470414,
      "loss": 2.5855,
      "theoretical_loss": 3.55476687474774,
      "tokens_seen": 1326841856
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030201604814443327,
      "loss": 2.8996,
      "theoretical_loss": 3.554751027990397,
      "tokens_seen": 1326907392
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003020060180541625,
      "loss": 2.8505,
      "theoretical_loss": 3.5547351822348423,
      "tokens_seen": 1326972928
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003019959879638917,
      "loss": 3.0077,
      "theoretical_loss": 3.5547193374809636,
      "tokens_seen": 1327038464
    },
    {
      "epoch": 0.4,
      "objective/train/advantage_avg": 0.49426817893981934,
      "objective/train/docs_used": 751457,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.896923780441284,
      "objective/train/original_loss": 2.896923780441284,
      "objective/train/theoretical_loss": 3.5547034937286472,
      "objective/train/tokens_used": 1347564000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24436092376708984,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506690740585327,
      "objective/train/weighted_lm_loss": 3.0436878204345703,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0428944826126099,
      "theoretical_loss": 3.5547034937286472,
      "tokens_seen": 1327104000
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030198595787362087,
      "loss": 2.8154,
      "theoretical_loss": 3.5547034937286472,
      "tokens_seen": 1327104000
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030197592778335005,
      "loss": 2.9268,
      "theoretical_loss": 3.554687650977781,
      "tokens_seen": 1327169536
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030196589769307923,
      "loss": 2.6363,
      "theoretical_loss": 3.5546718092282514,
      "tokens_seen": 1327235072
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003019558676028084,
      "loss": 2.7523,
      "theoretical_loss": 3.5546559684799472,
      "tokens_seen": 1327300608
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030194583751253764,
      "loss": 3.0527,
      "theoretical_loss": 3.5546401287327543,
      "tokens_seen": 1327366144
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030193580742226677,
      "loss": 2.9638,
      "theoretical_loss": 3.5546242899865605,
      "tokens_seen": 1327431680
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.000301925777331996,
      "loss": 2.9171,
      "theoretical_loss": 3.5546084522412533,
      "tokens_seen": 1327497216
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030191574724172513,
      "loss": 2.8834,
      "theoretical_loss": 3.55459261549672,
      "tokens_seen": 1327562752
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030190571715145437,
      "loss": 2.468,
      "theoretical_loss": 3.554576779752848,
      "tokens_seen": 1327628288
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030189568706118355,
      "loss": 2.7384,
      "theoretical_loss": 3.554560945009524,
      "tokens_seen": 1327693824
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030188565697091273,
      "loss": 3.0641,
      "theoretical_loss": 3.554545111266636,
      "tokens_seen": 1327759360
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003018756268806419,
      "loss": 2.7802,
      "theoretical_loss": 3.5545292785240714,
      "tokens_seen": 1327824896
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030186559679037115,
      "loss": 2.972,
      "theoretical_loss": 3.5545134467817174,
      "tokens_seen": 1327890432
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003018555667001003,
      "loss": 2.9482,
      "theoretical_loss": 3.554497616039462,
      "tokens_seen": 1327955968
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003018455366098295,
      "loss": 2.8153,
      "theoretical_loss": 3.554481786297192,
      "tokens_seen": 1328021504
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030183550651955864,
      "loss": 2.9736,
      "theoretical_loss": 3.5544659575547954,
      "tokens_seen": 1328087040
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030182547642928787,
      "loss": 2.9896,
      "theoretical_loss": 3.5544501298121594,
      "tokens_seen": 1328152576
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030181544633901705,
      "loss": 2.6453,
      "theoretical_loss": 3.5544343030691716,
      "tokens_seen": 1328218112
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030180541624874623,
      "loss": 2.9779,
      "theoretical_loss": 3.5544184773257195,
      "tokens_seen": 1328283648
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030179538615847547,
      "loss": 3.0892,
      "theoretical_loss": 3.5544026525816905,
      "tokens_seen": 1328349184
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003017853560682046,
      "loss": 2.525,
      "theoretical_loss": 3.5543868288369733,
      "tokens_seen": 1328414720
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030177532597793383,
      "loss": 2.9019,
      "theoretical_loss": 3.5543710060914537,
      "tokens_seen": 1328480256
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.000301765295887663,
      "loss": 2.8842,
      "theoretical_loss": 3.5543551843450203,
      "tokens_seen": 1328545792
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003017552657973922,
      "loss": 3.0107,
      "theoretical_loss": 3.5543393635975615,
      "tokens_seen": 1328611328
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003017452357071214,
      "loss": 2.82,
      "theoretical_loss": 3.5543235438489633,
      "tokens_seen": 1328676864
    },
    {
      "epoch": 0.4,
      "objective/train/advantage_avg": 0.4955544173717499,
      "objective/train/docs_used": 752138,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.7155673503875732,
      "objective/train/original_loss": 2.7155673503875732,
      "objective/train/theoretical_loss": 3.5543077250991146,
      "objective/train/tokens_used": 1349202400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24562297761440277,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0508040189743042,
      "objective/train/weighted_lm_loss": 2.8533716201782227,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0396549701690674,
      "theoretical_loss": 3.5543077250991146,
      "tokens_seen": 1328742400
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003017352056168506,
      "loss": 2.8639,
      "theoretical_loss": 3.5543077250991146,
      "tokens_seen": 1328742400
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030172517552657974,
      "loss": 2.6614,
      "theoretical_loss": 3.5542919073479022,
      "tokens_seen": 1328807936
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030171514543630897,
      "loss": 3.0466,
      "theoretical_loss": 3.5542760905952147,
      "tokens_seen": 1328873472
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003017051153460381,
      "loss": 3.0317,
      "theoretical_loss": 3.5542602748409395,
      "tokens_seen": 1328939008
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030169508525576733,
      "loss": 3.0212,
      "theoretical_loss": 3.5542444600849636,
      "tokens_seen": 1329004544
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003016850551654965,
      "loss": 2.7777,
      "theoretical_loss": 3.554228646327176,
      "tokens_seen": 1329070080
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003016750250752257,
      "loss": 2.8606,
      "theoretical_loss": 3.5542128335674636,
      "tokens_seen": 1329135616
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003016649949849549,
      "loss": 2.8532,
      "theoretical_loss": 3.554197021805715,
      "tokens_seen": 1329201152
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030165496489468406,
      "loss": 2.9963,
      "theoretical_loss": 3.5541812110418167,
      "tokens_seen": 1329266688
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030164493480441324,
      "loss": 2.6525,
      "theoretical_loss": 3.5541654012756583,
      "tokens_seen": 1329332224
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003016349047141425,
      "loss": 2.8583,
      "theoretical_loss": 3.554149592507126,
      "tokens_seen": 1329397760
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003016248746238716,
      "loss": 2.6197,
      "theoretical_loss": 3.5541337847361087,
      "tokens_seen": 1329463296
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030161484453360084,
      "loss": 2.7856,
      "theoretical_loss": 3.554117977962494,
      "tokens_seen": 1329528832
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030160481444332996,
      "loss": 2.8393,
      "theoretical_loss": 3.5541021721861696,
      "tokens_seen": 1329594368
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003015947843530592,
      "loss": 2.9963,
      "theoretical_loss": 3.5540863674070238,
      "tokens_seen": 1329659904
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003015847542627884,
      "loss": 2.8342,
      "theoretical_loss": 3.5540705636249443,
      "tokens_seen": 1329725440
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030157472417251756,
      "loss": 2.8927,
      "theoretical_loss": 3.554054760839819,
      "tokens_seen": 1329790976
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030156469408224674,
      "loss": 3.1101,
      "theoretical_loss": 3.554038959051536,
      "tokens_seen": 1329856512
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.000301554663991976,
      "loss": 2.5986,
      "theoretical_loss": 3.5540231582599833,
      "tokens_seen": 1329922048
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003015446339017051,
      "loss": 2.993,
      "theoretical_loss": 3.554007358465049,
      "tokens_seen": 1329987584
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030153460381143434,
      "loss": 2.8792,
      "theoretical_loss": 3.553991559666621,
      "tokens_seen": 1330053120
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030152457372116347,
      "loss": 2.835,
      "theoretical_loss": 3.5539757618645873,
      "tokens_seen": 1330118656
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003015145436308927,
      "loss": 2.8979,
      "theoretical_loss": 3.5539599650588363,
      "tokens_seen": 1330184192
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003015045135406219,
      "loss": 3.0106,
      "theoretical_loss": 3.553944169249256,
      "tokens_seen": 1330249728
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030149448345035107,
      "loss": 2.7598,
      "theoretical_loss": 3.553928374435734,
      "tokens_seen": 1330315264
    },
    {
      "epoch": 0.4,
      "objective/train/advantage_avg": 0.4953942894935608,
      "objective/train/docs_used": 753163,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.194594621658325,
      "objective/train/original_loss": 3.194594621658325,
      "objective/train/theoretical_loss": 3.5539125806181584,
      "objective/train/tokens_used": 1350840800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24552972614765167,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507875680923462,
      "objective/train/weighted_lm_loss": 3.3569045066833496,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0253151655197144,
      "theoretical_loss": 3.5539125806181584,
      "tokens_seen": 1330380800
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030148445336008025,
      "loss": 2.8602,
      "theoretical_loss": 3.5539125806181584,
      "tokens_seen": 1330380800
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030147442326980943,
      "loss": 3.0129,
      "theoretical_loss": 3.5538967877964183,
      "tokens_seen": 1330446336
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003014643931795386,
      "loss": 3.0533,
      "theoretical_loss": 3.553880995970401,
      "tokens_seen": 1330511872
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030145436308926784,
      "loss": 2.6389,
      "theoretical_loss": 3.5538652051399957,
      "tokens_seen": 1330577408
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030144433299899697,
      "loss": 2.8258,
      "theoretical_loss": 3.5538494153050895,
      "tokens_seen": 1330642944
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003014343029087262,
      "loss": 2.7927,
      "theoretical_loss": 3.5538336264655705,
      "tokens_seen": 1330708480
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030142427281845533,
      "loss": 2.8202,
      "theoretical_loss": 3.5538178386213275,
      "tokens_seen": 1330774016
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030141424272818457,
      "loss": 2.905,
      "theoretical_loss": 3.5538020517722493,
      "tokens_seen": 1330839552
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030140421263791375,
      "loss": 2.9649,
      "theoretical_loss": 3.553786265918223,
      "tokens_seen": 1330905088
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030139418254764293,
      "loss": 2.6982,
      "theoretical_loss": 3.5537704810591375,
      "tokens_seen": 1330970624
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003013841524573721,
      "loss": 3.0263,
      "theoretical_loss": 3.5537546971948815,
      "tokens_seen": 1331036160
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030137412236710135,
      "loss": 2.9706,
      "theoretical_loss": 3.5537389143253426,
      "tokens_seen": 1331101696
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003013640922768305,
      "loss": 3.1117,
      "theoretical_loss": 3.5537231324504095,
      "tokens_seen": 1331167232
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003013540621865597,
      "loss": 2.7145,
      "theoretical_loss": 3.5537073515699698,
      "tokens_seen": 1331232768
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030134403209628884,
      "loss": 2.681,
      "theoretical_loss": 3.5536915716839133,
      "tokens_seen": 1331298304
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030133400200601807,
      "loss": 2.8114,
      "theoretical_loss": 3.553675792792127,
      "tokens_seen": 1331363840
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030132397191574725,
      "loss": 2.9936,
      "theoretical_loss": 3.5536600148945006,
      "tokens_seen": 1331429376
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030131394182547643,
      "loss": 2.7925,
      "theoretical_loss": 3.5536442379909214,
      "tokens_seen": 1331494912
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003013039117352056,
      "loss": 2.7832,
      "theoretical_loss": 3.553628462081279,
      "tokens_seen": 1331560448
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003012938816449348,
      "loss": 2.8645,
      "theoretical_loss": 3.5536126871654607,
      "tokens_seen": 1331625984
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.000301283851554664,
      "loss": 2.9089,
      "theoretical_loss": 3.5535969132433554,
      "tokens_seen": 1331691520
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003012738214643932,
      "loss": 2.8147,
      "theoretical_loss": 3.5535811403148516,
      "tokens_seen": 1331757056
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030126379137412234,
      "loss": 2.9929,
      "theoretical_loss": 3.5535653683798385,
      "tokens_seen": 1331822592
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003012537612838516,
      "loss": 2.8867,
      "theoretical_loss": 3.5535495974382036,
      "tokens_seen": 1331888128
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003012437311935807,
      "loss": 2.6582,
      "theoretical_loss": 3.553533827489836,
      "tokens_seen": 1331953664
    },
    {
      "epoch": 0.4,
      "objective/train/advantage_avg": 0.4953325092792511,
      "objective/train/docs_used": 753870,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.723559856414795,
      "objective/train/original_loss": 2.723559617996216,
      "objective/train/theoretical_loss": 3.553518058534624,
      "objective/train/tokens_used": 1352479200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24540911614894867,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507807731628418,
      "objective/train/weighted_lm_loss": 2.861851930618286,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.038975715637207,
      "theoretical_loss": 3.553518058534624,
      "tokens_seen": 1332019200
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030123370110330994,
      "loss": 2.9985,
      "theoretical_loss": 3.553518058534624,
      "tokens_seen": 1332019200
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003012236710130391,
      "loss": 2.9034,
      "theoretical_loss": 3.5535022905724567,
      "tokens_seen": 1332084736
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003012136409227683,
      "loss": 2.8317,
      "theoretical_loss": 3.5534865236032225,
      "tokens_seen": 1332150272
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003012036108324975,
      "loss": 3.2594,
      "theoretical_loss": 3.5534707576268096,
      "tokens_seen": 1332215808
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003011935807422267,
      "loss": 2.7773,
      "theoretical_loss": 3.553454992643107,
      "tokens_seen": 1332281344
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030118355065195584,
      "loss": 2.98,
      "theoretical_loss": 3.5534392286520036,
      "tokens_seen": 1332346880
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003011735205616851,
      "loss": 2.8828,
      "theoretical_loss": 3.5534234656533874,
      "tokens_seen": 1332412416
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003011634904714142,
      "loss": 2.889,
      "theoretical_loss": 3.553407703647148,
      "tokens_seen": 1332477952
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030115346038114344,
      "loss": 2.8798,
      "theoretical_loss": 3.5533919426331737,
      "tokens_seen": 1332543488
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003011434302908726,
      "loss": 2.8996,
      "theoretical_loss": 3.553376182611353,
      "tokens_seen": 1332609024
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003011334002006018,
      "loss": 2.5455,
      "theoretical_loss": 3.553360423581575,
      "tokens_seen": 1332674560
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.000301123370110331,
      "loss": 3.1227,
      "theoretical_loss": 3.5533446655437277,
      "tokens_seen": 1332740096
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030111334002006016,
      "loss": 2.8868,
      "theoretical_loss": 3.5533289084977007,
      "tokens_seen": 1332805632
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030110330992978935,
      "loss": 2.8771,
      "theoretical_loss": 3.553313152443383,
      "tokens_seen": 1332871168
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003010932798395186,
      "loss": 2.9884,
      "theoretical_loss": 3.5532973973806627,
      "tokens_seen": 1332936704
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003010832497492477,
      "loss": 2.9111,
      "theoretical_loss": 3.553281643309429,
      "tokens_seen": 1333002240
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030107321965897694,
      "loss": 3.0882,
      "theoretical_loss": 3.553265890229571,
      "tokens_seen": 1333067776
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030106318956870607,
      "loss": 2.762,
      "theoretical_loss": 3.553250138140977,
      "tokens_seen": 1333133312
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003010531594784353,
      "loss": 2.6814,
      "theoretical_loss": 3.5532343870435366,
      "tokens_seen": 1333198848
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030104312938816454,
      "loss": 2.838,
      "theoretical_loss": 3.5532186369371384,
      "tokens_seen": 1333264384
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030103309929789367,
      "loss": 3.1008,
      "theoretical_loss": 3.553202887821671,
      "tokens_seen": 1333329920
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003010230692076229,
      "loss": 2.9039,
      "theoretical_loss": 3.553187139697024,
      "tokens_seen": 1333395456
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003010130391173521,
      "loss": 2.8976,
      "theoretical_loss": 3.5531713925630855,
      "tokens_seen": 1333460992
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030100300902708127,
      "loss": 2.8339,
      "theoretical_loss": 3.5531556464197456,
      "tokens_seen": 1333526528
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030099297893681045,
      "loss": 2.7447,
      "theoretical_loss": 3.553139901266892,
      "tokens_seen": 1333592064
    },
    {
      "epoch": 0.4,
      "objective/train/advantage_avg": 0.4881908893585205,
      "objective/train/docs_used": 755212,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.722550868988037,
      "objective/train/original_loss": 2.722550868988037,
      "objective/train/theoretical_loss": 3.5531241571044148,
      "objective/train/tokens_used": 1354117600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.23885828256607056,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0500330924987793,
      "objective/train/weighted_lm_loss": 2.8600103855133057,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.041465401649475,
      "theoretical_loss": 3.5531241571044148,
      "tokens_seen": 1333657600
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030098294884653963,
      "loss": 3.0424,
      "theoretical_loss": 3.5531241571044148,
      "tokens_seen": 1333657600
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003009729187562688,
      "loss": 2.8577,
      "theoretical_loss": 3.5531084139322027,
      "tokens_seen": 1333723136
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030096288866599804,
      "loss": 2.8246,
      "theoretical_loss": 3.5530926717501448,
      "tokens_seen": 1333788672
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030095285857572717,
      "loss": 2.9305,
      "theoretical_loss": 3.55307693055813,
      "tokens_seen": 1333854208
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003009428284854564,
      "loss": 2.9409,
      "theoretical_loss": 3.5530611903560474,
      "tokens_seen": 1333919744
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030093279839518553,
      "loss": 2.7161,
      "theoretical_loss": 3.5530454511437863,
      "tokens_seen": 1333985280
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030092276830491477,
      "loss": 2.9114,
      "theoretical_loss": 3.5530297129212354,
      "tokens_seen": 1334050816
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030091273821464395,
      "loss": 2.7608,
      "theoretical_loss": 3.553013975688285,
      "tokens_seen": 1334116352
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030090270812437313,
      "loss": 2.7329,
      "theoretical_loss": 3.5529982394448227,
      "tokens_seen": 1334181888
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003008926780341023,
      "loss": 2.819,
      "theoretical_loss": 3.552982504190739,
      "tokens_seen": 1334247424
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030088264794383155,
      "loss": 2.6038,
      "theoretical_loss": 3.5529667699259226,
      "tokens_seen": 1334312960
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003008726178535607,
      "loss": 2.8323,
      "theoretical_loss": 3.552951036650262,
      "tokens_seen": 1334378496
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003008625877632899,
      "loss": 2.9869,
      "theoretical_loss": 3.552935304363648,
      "tokens_seen": 1334444032
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030085255767301904,
      "loss": 2.8766,
      "theoretical_loss": 3.552919573065968,
      "tokens_seen": 1334509568
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030084252758274827,
      "loss": 2.8286,
      "theoretical_loss": 3.5529038427571127,
      "tokens_seen": 1334575104
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030083249749247745,
      "loss": 2.8724,
      "theoretical_loss": 3.552888113436971,
      "tokens_seen": 1334640640
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030082246740220663,
      "loss": 2.8761,
      "theoretical_loss": 3.5528723851054327,
      "tokens_seen": 1334706176
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003008124373119358,
      "loss": 2.9596,
      "theoretical_loss": 3.552856657762386,
      "tokens_seen": 1334771712
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.000300802407221665,
      "loss": 2.6596,
      "theoretical_loss": 3.5528409314077205,
      "tokens_seen": 1334837248
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003007923771313942,
      "loss": 2.9156,
      "theoretical_loss": 3.5528252060413266,
      "tokens_seen": 1334902784
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003007823470411234,
      "loss": 2.6166,
      "theoretical_loss": 3.5528094816630924,
      "tokens_seen": 1334968320
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030077231695085254,
      "loss": 2.5898,
      "theoretical_loss": 3.5527937582729083,
      "tokens_seen": 1335033856
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003007622868605818,
      "loss": 2.6842,
      "theoretical_loss": 3.552778035870663,
      "tokens_seen": 1335099392
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003007522567703109,
      "loss": 2.7666,
      "theoretical_loss": 3.5527623144562464,
      "tokens_seen": 1335164928
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030074222668004014,
      "loss": 2.7794,
      "theoretical_loss": 3.5527465940295473,
      "tokens_seen": 1335230464
    },
    {
      "epoch": 0.4,
      "objective/train/advantage_avg": 0.4953099489212036,
      "objective/train/docs_used": 755960,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.5784409046173096,
      "objective/train/original_loss": 2.5784409046173096,
      "objective/train/theoretical_loss": 3.5527308745904564,
      "objective/train/tokens_used": 1355756000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2453741878271103,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507783889770508,
      "objective/train/weighted_lm_loss": 2.709312677383423,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0449823141098022,
      "theoretical_loss": 3.5527308745904564,
      "tokens_seen": 1335296000
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003007321965897693,
      "loss": 2.8612,
      "theoretical_loss": 3.5527308745904564,
      "tokens_seen": 1335296000
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003007221664994985,
      "loss": 3.0564,
      "theoretical_loss": 3.552715156138862,
      "tokens_seen": 1335361536
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003007121364092277,
      "loss": 2.8051,
      "theoretical_loss": 3.5526994386746544,
      "tokens_seen": 1335427072
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003007021063189569,
      "loss": 3.132,
      "theoretical_loss": 3.552683722197722,
      "tokens_seen": 1335492608
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030069207622868604,
      "loss": 2.9723,
      "theoretical_loss": 3.5526680067079557,
      "tokens_seen": 1335558144
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003006820461384153,
      "loss": 2.9133,
      "theoretical_loss": 3.552652292205244,
      "tokens_seen": 1335623680
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003006720160481444,
      "loss": 2.8827,
      "theoretical_loss": 3.552636578689478,
      "tokens_seen": 1335689216
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030066198595787364,
      "loss": 2.79,
      "theoretical_loss": 3.552620866160545,
      "tokens_seen": 1335754752
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003006519558676028,
      "loss": 2.84,
      "theoretical_loss": 3.5526051546183366,
      "tokens_seen": 1335820288
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.000300641925777332,
      "loss": 2.6883,
      "theoretical_loss": 3.5525894440627415,
      "tokens_seen": 1335885824
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003006318956870612,
      "loss": 2.7569,
      "theoretical_loss": 3.5525737344936497,
      "tokens_seen": 1335951360
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030062186559679036,
      "loss": 2.6814,
      "theoretical_loss": 3.55255802591095,
      "tokens_seen": 1336016896
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030061183550651955,
      "loss": 2.7061,
      "theoretical_loss": 3.5525423183145337,
      "tokens_seen": 1336082432
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003006018054162488,
      "loss": 2.5423,
      "theoretical_loss": 3.5525266117042893,
      "tokens_seen": 1336147968
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003005917753259779,
      "loss": 2.6255,
      "theoretical_loss": 3.5525109060801063,
      "tokens_seen": 1336213504
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030058174523570714,
      "loss": 3.2222,
      "theoretical_loss": 3.552495201441875,
      "tokens_seen": 1336279040
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.00030057171514543627,
      "loss": 2.8907,
      "theoretical_loss": 3.552479497789486,
      "tokens_seen": 1336344576
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003005616850551655,
      "loss": 2.9438,
      "theoretical_loss": 3.5524637951228275,
      "tokens_seen": 1336410112
    },
    {
      "epoch": 0.4,
      "learning_rate": 0.0003005516549648947,
      "loss": 2.5715,
      "theoretical_loss": 3.55244809344179,
      "tokens_seen": 1336475648
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00030054162487462387,
      "loss": 2.7793,
      "theoretical_loss": 3.5524323927462635,
      "tokens_seen": 1336541184
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00030053159478435305,
      "loss": 2.756,
      "theoretical_loss": 3.552416693036137,
      "tokens_seen": 1336606720
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0003005215646940823,
      "loss": 2.8188,
      "theoretical_loss": 3.5524009943113013,
      "tokens_seen": 1336672256
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0003005115346038114,
      "loss": 2.7056,
      "theoretical_loss": 3.5523852965716456,
      "tokens_seen": 1336737792
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00030050150451354065,
      "loss": 2.9825,
      "theoretical_loss": 3.5523695998170606,
      "tokens_seen": 1336803328
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0003004914744232698,
      "loss": 2.7813,
      "theoretical_loss": 3.5523539040474352,
      "tokens_seen": 1336868864
    },
    {
      "epoch": 0.41,
      "objective/train/advantage_avg": 0.49310922622680664,
      "objective/train/docs_used": 757325,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.7950615882873535,
      "objective/train/original_loss": 2.7950618267059326,
      "objective/train/theoretical_loss": 3.5523382092626603,
      "objective/train/tokens_used": 1357394400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24323943257331848,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505473613739014,
      "objective/train/weighted_lm_loss": 2.9362199306488037,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0253151655197144,
      "theoretical_loss": 3.5523382092626603,
      "tokens_seen": 1336934400
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.000300481444332999,
      "loss": 2.7847,
      "theoretical_loss": 3.5523382092626603,
      "tokens_seen": 1336934400
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0003004714142427282,
      "loss": 2.8298,
      "theoretical_loss": 3.552322515462625,
      "tokens_seen": 1336999936
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00030046138415245737,
      "loss": 2.7164,
      "theoretical_loss": 3.5523068226472194,
      "tokens_seen": 1337065472
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00030045135406218655,
      "loss": 2.7335,
      "theoretical_loss": 3.552291130816334,
      "tokens_seen": 1337131008
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00030044132397191573,
      "loss": 3.1408,
      "theoretical_loss": 3.5522754399698586,
      "tokens_seen": 1337196544
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0003004312938816449,
      "loss": 2.9076,
      "theoretical_loss": 3.552259750107682,
      "tokens_seen": 1337262080
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00030042126379137415,
      "loss": 2.9708,
      "theoretical_loss": 3.5522440612296964,
      "tokens_seen": 1337327616
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0003004112337011033,
      "loss": 2.9132,
      "theoretical_loss": 3.5522283733357902,
      "tokens_seen": 1337393152
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0003004012036108325,
      "loss": 2.8242,
      "theoretical_loss": 3.552212686425854,
      "tokens_seen": 1337458688
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0003003911735205617,
      "loss": 2.5365,
      "theoretical_loss": 3.5521970004997776,
      "tokens_seen": 1337524224
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0003003811434302909,
      "loss": 2.9666,
      "theoretical_loss": 3.5521813155574513,
      "tokens_seen": 1337589760
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00030037111334002006,
      "loss": 2.6702,
      "theoretical_loss": 3.5521656315987657,
      "tokens_seen": 1337655296
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00030036108324974924,
      "loss": 2.8619,
      "theoretical_loss": 3.55214994862361,
      "tokens_seen": 1337720832
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0003003510531594784,
      "loss": 2.6485,
      "theoretical_loss": 3.5521342666318745,
      "tokens_seen": 1337786368
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00030034102306920765,
      "loss": 2.958,
      "theoretical_loss": 3.5521185856234503,
      "tokens_seen": 1337851904
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0003003309929789368,
      "loss": 2.9773,
      "theoretical_loss": 3.5521029055982263,
      "tokens_seen": 1337917440
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.000300320962888666,
      "loss": 2.8713,
      "theoretical_loss": 3.552087226556094,
      "tokens_seen": 1337982976
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00030031093279839514,
      "loss": 3.0001,
      "theoretical_loss": 3.5520715484969427,
      "tokens_seen": 1338048512
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0003003009027081244,
      "loss": 3.0317,
      "theoretical_loss": 3.552055871420662,
      "tokens_seen": 1338114048
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0003002908726178536,
      "loss": 3.048,
      "theoretical_loss": 3.5520401953271445,
      "tokens_seen": 1338179584
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00030028084252758274,
      "loss": 2.7465,
      "theoretical_loss": 3.552024520216278,
      "tokens_seen": 1338245120
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.000300270812437312,
      "loss": 2.6631,
      "theoretical_loss": 3.5520088460879538,
      "tokens_seen": 1338310656
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0003002607823470411,
      "loss": 2.8521,
      "theoretical_loss": 3.5519931729420624,
      "tokens_seen": 1338376192
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00030025075225677034,
      "loss": 2.7839,
      "theoretical_loss": 3.551977500778494,
      "tokens_seen": 1338441728
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0003002407221664995,
      "loss": 2.7662,
      "theoretical_loss": 3.551961829597139,
      "tokens_seen": 1338507264
    },
    {
      "epoch": 0.41,
      "objective/train/advantage_avg": 0.4945479929447174,
      "objective/train/docs_used": 758060,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.5743420124053955,
      "objective/train/original_loss": 2.5743417739868164,
      "objective/train/theoretical_loss": 3.551946159397887,
      "objective/train/tokens_used": 1359032800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24462446570396423,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506983995437622,
      "objective/train/weighted_lm_loss": 2.7049171924591064,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0442860126495361,
      "theoretical_loss": 3.551946159397887,
      "tokens_seen": 1338572800
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0003002306920762287,
      "loss": 2.833,
      "theoretical_loss": 3.551946159397887,
      "tokens_seen": 1338572800
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0003002206619859579,
      "loss": 2.7197,
      "theoretical_loss": 3.551930490180629,
      "tokens_seen": 1338638336
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0003002106318956871,
      "loss": 2.9241,
      "theoretical_loss": 3.551914821945256,
      "tokens_seen": 1338703872
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00030020060180541624,
      "loss": 2.8071,
      "theoretical_loss": 3.551899154691657,
      "tokens_seen": 1338769408
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0003001905717151455,
      "loss": 2.8225,
      "theoretical_loss": 3.5518834884197235,
      "tokens_seen": 1338834944
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0003001805416248746,
      "loss": 2.7112,
      "theoretical_loss": 3.5518678231293457,
      "tokens_seen": 1338900480
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00030017051153460384,
      "loss": 2.9986,
      "theoretical_loss": 3.551852158820414,
      "tokens_seen": 1338966016
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.000300160481444333,
      "loss": 2.8904,
      "theoretical_loss": 3.5518364954928185,
      "tokens_seen": 1339031552
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0003001504513540622,
      "loss": 2.6732,
      "theoretical_loss": 3.5518208331464507,
      "tokens_seen": 1339097088
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0003001404212637914,
      "loss": 2.9215,
      "theoretical_loss": 3.5518051717812,
      "tokens_seen": 1339162624
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00030013039117352056,
      "loss": 2.6018,
      "theoretical_loss": 3.5517895113969575,
      "tokens_seen": 1339228160
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00030012036108324975,
      "loss": 2.8143,
      "theoretical_loss": 3.5517738519936133,
      "tokens_seen": 1339293696
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.000300110330992979,
      "loss": 2.7564,
      "theoretical_loss": 3.551758193571059,
      "tokens_seen": 1339359232
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0003001003009027081,
      "loss": 3.1684,
      "theoretical_loss": 3.551742536129184,
      "tokens_seen": 1339424768
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00030009027081243734,
      "loss": 3.0871,
      "theoretical_loss": 3.55172687966788,
      "tokens_seen": 1339490304
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00030008024072216647,
      "loss": 2.8282,
      "theoretical_loss": 3.5517112241870366,
      "tokens_seen": 1339555840
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0003000702106318957,
      "loss": 2.6431,
      "theoretical_loss": 3.5516955696865447,
      "tokens_seen": 1339621376
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0003000601805416249,
      "loss": 2.7553,
      "theoretical_loss": 3.5516799161662957,
      "tokens_seen": 1339686912
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00030005015045135407,
      "loss": 2.5601,
      "theoretical_loss": 3.5516642636261793,
      "tokens_seen": 1339752448
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00030004012036108325,
      "loss": 2.981,
      "theoretical_loss": 3.551648612066087,
      "tokens_seen": 1339817984
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0003000300902708125,
      "loss": 2.7049,
      "theoretical_loss": 3.5516329614859083,
      "tokens_seen": 1339883520
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0003000200601805416,
      "loss": 2.7903,
      "theoretical_loss": 3.551617311885535,
      "tokens_seen": 1339949056
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00030001003009027085,
      "loss": 3.0298,
      "theoretical_loss": 3.551601663264858,
      "tokens_seen": 1340014592
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0003,
      "loss": 2.8917,
      "theoretical_loss": 3.551586015623767,
      "tokens_seen": 1340080128
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002999899699097292,
      "loss": 2.7711,
      "theoretical_loss": 3.5515703689621536,
      "tokens_seen": 1340145664
    },
    {
      "epoch": 0.41,
      "objective/train/advantage_avg": 0.49450618028640747,
      "objective/train/docs_used": 759300,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.625169277191162,
      "objective/train/original_loss": 2.625169277191162,
      "objective/train/theoretical_loss": 3.5515547232799087,
      "objective/train/tokens_used": 1360671200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.244602769613266,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506939888000488,
      "objective/train/weighted_lm_loss": 2.758131742477417,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0408108234405518,
      "theoretical_loss": 3.5515547232799087,
      "tokens_seen": 1340211200
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002999799398194584,
      "loss": 2.7422,
      "theoretical_loss": 3.5515547232799087,
      "tokens_seen": 1340211200
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029996990972918757,
      "loss": 2.8988,
      "theoretical_loss": 3.5515390785769227,
      "tokens_seen": 1340276736
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029995987963891675,
      "loss": 2.6772,
      "theoretical_loss": 3.5515234348530864,
      "tokens_seen": 1340342272
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029994984954864593,
      "loss": 2.9557,
      "theoretical_loss": 3.5515077921082905,
      "tokens_seen": 1340407808
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002999398194583751,
      "loss": 2.716,
      "theoretical_loss": 3.5514921503424266,
      "tokens_seen": 1340473344
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029992978936810435,
      "loss": 2.7549,
      "theoretical_loss": 3.5514765095553846,
      "tokens_seen": 1340538880
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002999197592778335,
      "loss": 3.0217,
      "theoretical_loss": 3.551460869747056,
      "tokens_seen": 1340604416
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002999097291875627,
      "loss": 2.5925,
      "theoretical_loss": 3.5514452309173326,
      "tokens_seen": 1340669952
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002998996990972919,
      "loss": 2.6371,
      "theoretical_loss": 3.5514295930661035,
      "tokens_seen": 1340735488
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002998896690070211,
      "loss": 2.9088,
      "theoretical_loss": 3.5514139561932607,
      "tokens_seen": 1340801024
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029987963891675026,
      "loss": 2.666,
      "theoretical_loss": 3.551398320298695,
      "tokens_seen": 1340866560
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029986960882647944,
      "loss": 2.8632,
      "theoretical_loss": 3.551382685382298,
      "tokens_seen": 1340932096
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002998595787362086,
      "loss": 2.8265,
      "theoretical_loss": 3.5513670514439593,
      "tokens_seen": 1340997632
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029984954864593785,
      "loss": 2.6203,
      "theoretical_loss": 3.551351418483571,
      "tokens_seen": 1341063168
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.000299839518555667,
      "loss": 2.9688,
      "theoretical_loss": 3.5513357865010233,
      "tokens_seen": 1341128704
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002998294884653962,
      "loss": 2.9404,
      "theoretical_loss": 3.5513201554962084,
      "tokens_seen": 1341194240
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029981945837512534,
      "loss": 2.9954,
      "theoretical_loss": 3.5513045254690168,
      "tokens_seen": 1341259776
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002998094282848546,
      "loss": 2.7084,
      "theoretical_loss": 3.551288896419339,
      "tokens_seen": 1341325312
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029979939819458376,
      "loss": 2.7093,
      "theoretical_loss": 3.5512732683470674,
      "tokens_seen": 1341390848
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029978936810431294,
      "loss": 2.7612,
      "theoretical_loss": 3.551257641252092,
      "tokens_seen": 1341456384
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002997793380140421,
      "loss": 2.8637,
      "theoretical_loss": 3.5512420151343047,
      "tokens_seen": 1341521920
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002997693079237713,
      "loss": 2.8464,
      "theoretical_loss": 3.5512263899935963,
      "tokens_seen": 1341587456
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002997592778335005,
      "loss": 2.7996,
      "theoretical_loss": 3.5512107658298575,
      "tokens_seen": 1341652992
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002997492477432297,
      "loss": 2.8916,
      "theoretical_loss": 3.55119514264298,
      "tokens_seen": 1341718528
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029973921765295885,
      "loss": 3.0443,
      "theoretical_loss": 3.5511795204328553,
      "tokens_seen": 1341784064
    },
    {
      "epoch": 0.41,
      "objective/train/advantage_avg": 0.4945753812789917,
      "objective/train/docs_used": 759950,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.769815683364868,
      "objective/train/original_loss": 2.769815444946289,
      "objective/train/theoretical_loss": 3.5511638991993744,
      "objective/train/tokens_used": 1362309600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24466320872306824,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507012605667114,
      "objective/train/weighted_lm_loss": 2.910219669342041,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0253151655197144,
      "theoretical_loss": 3.5511638991993744,
      "tokens_seen": 1341849600
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002997291875626881,
      "loss": 2.9443,
      "theoretical_loss": 3.5511638991993744,
      "tokens_seen": 1341849600
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029971915747241726,
      "loss": 2.8056,
      "theoretical_loss": 3.5511482789424287,
      "tokens_seen": 1341915136
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029970912738214644,
      "loss": 2.8376,
      "theoretical_loss": 3.551132659661909,
      "tokens_seen": 1341980672
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002996990972918756,
      "loss": 2.7754,
      "theoretical_loss": 3.551117041357707,
      "tokens_seen": 1342046208
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002996890672016048,
      "loss": 3.1127,
      "theoretical_loss": 3.5511014240297136,
      "tokens_seen": 1342111744
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.000299679037111334,
      "loss": 2.8003,
      "theoretical_loss": 3.5510858076778202,
      "tokens_seen": 1342177280
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002996690070210632,
      "loss": 2.7068,
      "theoretical_loss": 3.5510701923019186,
      "tokens_seen": 1342242816
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029965897693079235,
      "loss": 2.8129,
      "theoretical_loss": 3.5510545779019,
      "tokens_seen": 1342308352
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002996489468405216,
      "loss": 2.9585,
      "theoretical_loss": 3.551038964477655,
      "tokens_seen": 1342373888
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002996389167502507,
      "loss": 2.7484,
      "theoretical_loss": 3.551023352029076,
      "tokens_seen": 1342439424
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029962888665997995,
      "loss": 2.6009,
      "theoretical_loss": 3.551007740556054,
      "tokens_seen": 1342504960
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029961885656970913,
      "loss": 2.8759,
      "theoretical_loss": 3.5509921300584804,
      "tokens_seen": 1342570496
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002996088264794383,
      "loss": 2.8533,
      "theoretical_loss": 3.5509765205362465,
      "tokens_seen": 1342636032
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002995987963891675,
      "loss": 2.8308,
      "theoretical_loss": 3.5509609119892445,
      "tokens_seen": 1342701568
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029958876629889667,
      "loss": 3.0144,
      "theoretical_loss": 3.5509453044173647,
      "tokens_seen": 1342767104
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029957873620862585,
      "loss": 3.0158,
      "theoretical_loss": 3.5509296978204996,
      "tokens_seen": 1342832640
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002995687061183551,
      "loss": 2.7416,
      "theoretical_loss": 3.55091409219854,
      "tokens_seen": 1342898176
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002995586760280842,
      "loss": 2.8814,
      "theoretical_loss": 3.550898487551378,
      "tokens_seen": 1342963712
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029954864593781345,
      "loss": 2.6444,
      "theoretical_loss": 3.550882883878905,
      "tokens_seen": 1343029248
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002995386158475427,
      "loss": 2.733,
      "theoretical_loss": 3.550867281181012,
      "tokens_seen": 1343094784
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002995285857572718,
      "loss": 2.7876,
      "theoretical_loss": 3.5508516794575913,
      "tokens_seen": 1343160320
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029951855566700105,
      "loss": 2.9119,
      "theoretical_loss": 3.5508360787085342,
      "tokens_seen": 1343225856
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002995085255767302,
      "loss": 3.0147,
      "theoretical_loss": 3.5508204789337325,
      "tokens_seen": 1343291392
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002994984954864594,
      "loss": 2.8989,
      "theoretical_loss": 3.550804880133077,
      "tokens_seen": 1343356928
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002994884653961886,
      "loss": 2.9295,
      "theoretical_loss": 3.550789282306461,
      "tokens_seen": 1343422464
    },
    {
      "epoch": 0.41,
      "objective/train/advantage_avg": 0.4956955015659332,
      "objective/train/docs_used": 761102,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.3120293617248535,
      "objective/train/original_loss": 2.3120293617248535,
      "objective/train/theoretical_loss": 3.550773685453774,
      "objective/train/tokens_used": 1363948000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24577094614505768,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050818920135498,
      "objective/train/weighted_lm_loss": 2.4293549060821533,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0253151655197144,
      "theoretical_loss": 3.550773685453774,
      "tokens_seen": 1343488000
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029947843530591777,
      "loss": 2.7885,
      "theoretical_loss": 3.550773685453774,
      "tokens_seen": 1343488000
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029946840521564695,
      "loss": 2.8473,
      "theoretical_loss": 3.55075808957491,
      "tokens_seen": 1343553536
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029945837512537613,
      "loss": 2.8789,
      "theoretical_loss": 3.550742494669759,
      "tokens_seen": 1343619072
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002994483450351053,
      "loss": 2.8034,
      "theoretical_loss": 3.550726900738213,
      "tokens_seen": 1343684608
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029943831494483455,
      "loss": 2.8434,
      "theoretical_loss": 3.5507113077801646,
      "tokens_seen": 1343750144
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002994282848545637,
      "loss": 2.9742,
      "theoretical_loss": 3.550695715795505,
      "tokens_seen": 1343815680
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002994182547642929,
      "loss": 2.6561,
      "theoretical_loss": 3.5506801247841255,
      "tokens_seen": 1343881216
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002994082246740221,
      "loss": 2.7946,
      "theoretical_loss": 3.5506645347459185,
      "tokens_seen": 1343946752
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002993981945837513,
      "loss": 2.593,
      "theoretical_loss": 3.5506489456807757,
      "tokens_seen": 1344012288
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029938816449348046,
      "loss": 2.8207,
      "theoretical_loss": 3.550633357588589,
      "tokens_seen": 1344077824
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029937813440320964,
      "loss": 2.7097,
      "theoretical_loss": 3.55061777046925,
      "tokens_seen": 1344143360
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002993681043129388,
      "loss": 2.729,
      "theoretical_loss": 3.55060218432265,
      "tokens_seen": 1344208896
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029935807422266805,
      "loss": 2.5919,
      "theoretical_loss": 3.5505865991486827,
      "tokens_seen": 1344274432
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002993480441323972,
      "loss": 2.9217,
      "theoretical_loss": 3.550571014947238,
      "tokens_seen": 1344339968
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002993380140421264,
      "loss": 2.6328,
      "theoretical_loss": 3.550555431718209,
      "tokens_seen": 1344405504
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029932798395185554,
      "loss": 2.728,
      "theoretical_loss": 3.5505398494614866,
      "tokens_seen": 1344471040
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002993179538615848,
      "loss": 2.7998,
      "theoretical_loss": 3.550524268176964,
      "tokens_seen": 1344536576
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029930792377131396,
      "loss": 2.6953,
      "theoretical_loss": 3.550508687864532,
      "tokens_seen": 1344602112
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029929789368104314,
      "loss": 3.0946,
      "theoretical_loss": 3.550493108524084,
      "tokens_seen": 1344667648
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002992878635907723,
      "loss": 2.6819,
      "theoretical_loss": 3.5504775301555105,
      "tokens_seen": 1344733184
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002992778335005015,
      "loss": 2.8569,
      "theoretical_loss": 3.5504619527587042,
      "tokens_seen": 1344798720
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002992678034102307,
      "loss": 2.8851,
      "theoretical_loss": 3.550446376333557,
      "tokens_seen": 1344864256
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002992577733199599,
      "loss": 2.8967,
      "theoretical_loss": 3.5504308008799614,
      "tokens_seen": 1344929792
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029924774322968905,
      "loss": 2.7404,
      "theoretical_loss": 3.550415226397809,
      "tokens_seen": 1344995328
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002992377131394183,
      "loss": 2.9583,
      "theoretical_loss": 3.5503996528869917,
      "tokens_seen": 1345060864
    },
    {
      "epoch": 0.41,
      "objective/train/advantage_avg": 0.49437108635902405,
      "objective/train/docs_used": 761644,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.3978404998779297,
      "objective/train/original_loss": 2.3978404998779297,
      "objective/train/theoretical_loss": 3.550384080347402,
      "objective/train/tokens_used": 1365586400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2444564402103424,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506798028945923,
      "objective/train/weighted_lm_loss": 2.5193002223968506,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0413066148757935,
      "theoretical_loss": 3.550384080347402,
      "tokens_seen": 1345126400
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029922768304914746,
      "loss": 2.7263,
      "theoretical_loss": 3.550384080347402,
      "tokens_seen": 1345126400
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029921765295887664,
      "loss": 2.7357,
      "theoretical_loss": 3.5503685087789316,
      "tokens_seen": 1345191936
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002992076228686058,
      "loss": 2.8477,
      "theoretical_loss": 3.550352938181473,
      "tokens_seen": 1345257472
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.000299197592778335,
      "loss": 2.8026,
      "theoretical_loss": 3.5503373685549184,
      "tokens_seen": 1345323008
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002991875626880642,
      "loss": 2.782,
      "theoretical_loss": 3.5503217998991596,
      "tokens_seen": 1345388544
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002991775325977934,
      "loss": 2.77,
      "theoretical_loss": 3.5503062322140897,
      "tokens_seen": 1345454080
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029916750250752255,
      "loss": 2.5807,
      "theoretical_loss": 3.5502906654996,
      "tokens_seen": 1345519616
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002991574724172518,
      "loss": 3.039,
      "theoretical_loss": 3.550275099755582,
      "tokens_seen": 1345585152
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002991474423269809,
      "loss": 2.9959,
      "theoretical_loss": 3.5502595349819295,
      "tokens_seen": 1345650688
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029913741223671015,
      "loss": 2.7929,
      "theoretical_loss": 3.550243971178534,
      "tokens_seen": 1345716224
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029912738214643933,
      "loss": 2.9559,
      "theoretical_loss": 3.550228408345288,
      "tokens_seen": 1345781760
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002991173520561685,
      "loss": 2.9204,
      "theoretical_loss": 3.5502128464820837,
      "tokens_seen": 1345847296
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002991073219658977,
      "loss": 2.7692,
      "theoretical_loss": 3.5501972855888138,
      "tokens_seen": 1345912832
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029909729187562687,
      "loss": 2.8734,
      "theoretical_loss": 3.5501817256653694,
      "tokens_seen": 1345978368
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029908726178535605,
      "loss": 2.9508,
      "theoretical_loss": 3.550166166711644,
      "tokens_seen": 1346043904
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002990772316950853,
      "loss": 2.7344,
      "theoretical_loss": 3.550150608727529,
      "tokens_seen": 1346109440
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002990672016048144,
      "loss": 2.8945,
      "theoretical_loss": 3.5501350517129184,
      "tokens_seen": 1346174976
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029905717151454365,
      "loss": 2.8241,
      "theoretical_loss": 3.550119495667703,
      "tokens_seen": 1346240512
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029904714142427283,
      "loss": 2.7539,
      "theoretical_loss": 3.550103940591776,
      "tokens_seen": 1346306048
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.000299037111334002,
      "loss": 2.7483,
      "theoretical_loss": 3.5500883864850294,
      "tokens_seen": 1346371584
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002990270812437312,
      "loss": 2.8412,
      "theoretical_loss": 3.5500728333473557,
      "tokens_seen": 1346437120
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002990170511534604,
      "loss": 3.0685,
      "theoretical_loss": 3.550057281178648,
      "tokens_seen": 1346502656
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029900702106318956,
      "loss": 3.0988,
      "theoretical_loss": 3.5500417299787976,
      "tokens_seen": 1346568192
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002989969909729188,
      "loss": 2.9756,
      "theoretical_loss": 3.5500261797476984,
      "tokens_seen": 1346633728
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002989869608826479,
      "loss": 2.6754,
      "theoretical_loss": 3.5500106304852412,
      "tokens_seen": 1346699264
    },
    {
      "epoch": 0.41,
      "objective/train/advantage_avg": 0.4952184557914734,
      "objective/train/docs_used": 762571,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.6544687747955322,
      "objective/train/original_loss": 2.6544687747955322,
      "objective/train/theoretical_loss": 3.5499950821913204,
      "objective/train/tokens_used": 1367224800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24529409408569336,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507688522338867,
      "objective/train/weighted_lm_loss": 2.789095878601074,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0462769269943237,
      "theoretical_loss": 3.5499950821913204,
      "tokens_seen": 1346764800
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029897693079237715,
      "loss": 2.9611,
      "theoretical_loss": 3.5499950821913204,
      "tokens_seen": 1346764800
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002989669007021063,
      "loss": 2.7417,
      "theoretical_loss": 3.5499795348658276,
      "tokens_seen": 1346830336
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002989568706118355,
      "loss": 2.8253,
      "theoretical_loss": 3.5499639885086554,
      "tokens_seen": 1346895872
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002989468405215647,
      "loss": 2.8248,
      "theoretical_loss": 3.549948443119696,
      "tokens_seen": 1346961408
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002989368104312939,
      "loss": 2.9215,
      "theoretical_loss": 3.5499328986988425,
      "tokens_seen": 1347026944
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029892678034102306,
      "loss": 3.1506,
      "theoretical_loss": 3.549917355245987,
      "tokens_seen": 1347092480
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002989167502507523,
      "loss": 2.7203,
      "theoretical_loss": 3.549901812761023,
      "tokens_seen": 1347158016
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002989067201604814,
      "loss": 2.8399,
      "theoretical_loss": 3.5498862712438433,
      "tokens_seen": 1347223552
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029889669007021066,
      "loss": 2.5367,
      "theoretical_loss": 3.549870730694339,
      "tokens_seen": 1347289088
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002988866599799398,
      "loss": 2.8242,
      "theoretical_loss": 3.549855191112404,
      "tokens_seen": 1347354624
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.000298876629889669,
      "loss": 2.8799,
      "theoretical_loss": 3.5498396524979308,
      "tokens_seen": 1347420160
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002988665997993982,
      "loss": 2.9845,
      "theoretical_loss": 3.549824114850812,
      "tokens_seen": 1347485696
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002988565697091274,
      "loss": 2.8194,
      "theoretical_loss": 3.5498085781709405,
      "tokens_seen": 1347551232
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029884653961885656,
      "loss": 2.704,
      "theoretical_loss": 3.5497930424582087,
      "tokens_seen": 1347616768
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029883650952858574,
      "loss": 3.2279,
      "theoretical_loss": 3.54977750771251,
      "tokens_seen": 1347682304
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002988264794383149,
      "loss": 2.8784,
      "theoretical_loss": 3.5497619739337365,
      "tokens_seen": 1347747840
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029881644934804416,
      "loss": 2.8875,
      "theoretical_loss": 3.549746441121781,
      "tokens_seen": 1347813376
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002988064192577733,
      "loss": 2.8455,
      "theoretical_loss": 3.549730909276537,
      "tokens_seen": 1347878912
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002987963891675025,
      "loss": 2.6462,
      "theoretical_loss": 3.549715378397897,
      "tokens_seen": 1347944448
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002987863590772317,
      "loss": 2.8535,
      "theoretical_loss": 3.549699848485753,
      "tokens_seen": 1348009984
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002987763289869609,
      "loss": 3.0165,
      "theoretical_loss": 3.54968431954,
      "tokens_seen": 1348075520
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002987662988966901,
      "loss": 2.9527,
      "theoretical_loss": 3.5496687915605287,
      "tokens_seen": 1348141056
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029875626880641925,
      "loss": 2.7955,
      "theoretical_loss": 3.549653264547233,
      "tokens_seen": 1348206592
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002987462387161485,
      "loss": 2.7863,
      "theoretical_loss": 3.5496377385000057,
      "tokens_seen": 1348272128
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029873620862587766,
      "loss": 2.747,
      "theoretical_loss": 3.54962221341874,
      "tokens_seen": 1348337664
    },
    {
      "epoch": 0.41,
      "objective/train/advantage_avg": 0.49311578273773193,
      "objective/train/docs_used": 763291,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.767502784729004,
      "objective/train/original_loss": 2.767502546310425,
      "objective/train/theoretical_loss": 3.549606689303328,
      "objective/train/tokens_used": 1368863200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2432556450366974,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505480766296387,
      "objective/train/weighted_lm_loss": 2.907355546951294,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0338950157165527,
      "theoretical_loss": 3.549606689303328,
      "tokens_seen": 1348403200
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029872617853560684,
      "loss": 2.7961,
      "theoretical_loss": 3.549606689303328,
      "tokens_seen": 1348403200
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.000298716148445336,
      "loss": 2.8471,
      "theoretical_loss": 3.5495911661536637,
      "tokens_seen": 1348468736
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002987061183550652,
      "loss": 2.8012,
      "theoretical_loss": 3.5495756439696393,
      "tokens_seen": 1348534272
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002986960882647944,
      "loss": 2.8791,
      "theoretical_loss": 3.5495601227511484,
      "tokens_seen": 1348599808
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002986860581745236,
      "loss": 3.1263,
      "theoretical_loss": 3.5495446024980835,
      "tokens_seen": 1348665344
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029867602808425275,
      "loss": 2.969,
      "theoretical_loss": 3.5495290832103383,
      "tokens_seen": 1348730880
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.000298665997993982,
      "loss": 2.897,
      "theoretical_loss": 3.549513564887805,
      "tokens_seen": 1348796416
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002986559679037111,
      "loss": 2.7329,
      "theoretical_loss": 3.5494980475303777,
      "tokens_seen": 1348861952
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029864593781344035,
      "loss": 2.8366,
      "theoretical_loss": 3.549482531137949,
      "tokens_seen": 1348927488
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029863590772316953,
      "loss": 2.538,
      "theoretical_loss": 3.5494670157104116,
      "tokens_seen": 1348993024
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002986258776328987,
      "loss": 2.877,
      "theoretical_loss": 3.549451501247659,
      "tokens_seen": 1349058560
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002986158475426279,
      "loss": 3.0545,
      "theoretical_loss": 3.5494359877495842,
      "tokens_seen": 1349124096
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029860581745235707,
      "loss": 2.6233,
      "theoretical_loss": 3.5494204752160803,
      "tokens_seen": 1349189632
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029859578736208625,
      "loss": 2.661,
      "theoretical_loss": 3.5494049636470413,
      "tokens_seen": 1349255168
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002985857572718155,
      "loss": 2.6979,
      "theoretical_loss": 3.5493894530423593,
      "tokens_seen": 1349320704
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002985757271815446,
      "loss": 2.7765,
      "theoretical_loss": 3.5493739434019282,
      "tokens_seen": 1349386240
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029856569709127385,
      "loss": 3.1333,
      "theoretical_loss": 3.5493584347256406,
      "tokens_seen": 1349451776
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029855566700100303,
      "loss": 2.8845,
      "theoretical_loss": 3.5493429270133907,
      "tokens_seen": 1349517312
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002985456369107322,
      "loss": 2.8721,
      "theoretical_loss": 3.5493274202650706,
      "tokens_seen": 1349582848
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002985356068204614,
      "loss": 2.8778,
      "theoretical_loss": 3.5493119144805743,
      "tokens_seen": 1349648384
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002985255767301906,
      "loss": 2.9509,
      "theoretical_loss": 3.5492964096597954,
      "tokens_seen": 1349713920
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029851554663991976,
      "loss": 2.6988,
      "theoretical_loss": 3.5492809058026262,
      "tokens_seen": 1349779456
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.000298505516549649,
      "loss": 3.0484,
      "theoretical_loss": 3.5492654029089614,
      "tokens_seen": 1349844992
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002984954864593781,
      "loss": 2.9153,
      "theoretical_loss": 3.549249900978693,
      "tokens_seen": 1349910528
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029848545636910735,
      "loss": 2.7767,
      "theoretical_loss": 3.5492344000117146,
      "tokens_seen": 1349976064
    },
    {
      "epoch": 0.41,
      "objective/train/advantage_avg": 0.4929984211921692,
      "objective/train/docs_used": 764652,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.6372923851013184,
      "objective/train/original_loss": 2.6372926235198975,
      "objective/train/theoretical_loss": 3.549218900007921,
      "objective/train/tokens_used": 1370501600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24317175149917603,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505359172821045,
      "objective/train/weighted_lm_loss": 2.7705559730529785,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0304546356201172,
      "theoretical_loss": 3.549218900007921,
      "tokens_seen": 1350041600
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002984754262788365,
      "loss": 2.7091,
      "theoretical_loss": 3.549218900007921,
      "tokens_seen": 1350041600
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002984653961885657,
      "loss": 2.8643,
      "theoretical_loss": 3.5492034009672038,
      "tokens_seen": 1350107136
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002984553660982949,
      "loss": 2.8829,
      "theoretical_loss": 3.5491879028894573,
      "tokens_seen": 1350172672
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002984453360080241,
      "loss": 2.9126,
      "theoretical_loss": 3.5491724057745744,
      "tokens_seen": 1350238208
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029843530591775326,
      "loss": 2.5909,
      "theoretical_loss": 3.5491569096224493,
      "tokens_seen": 1350303744
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002984252758274825,
      "loss": 2.6335,
      "theoretical_loss": 3.5491414144329747,
      "tokens_seen": 1350369280
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002984152457372116,
      "loss": 2.6424,
      "theoretical_loss": 3.549125920206045,
      "tokens_seen": 1350434816
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029840521564694086,
      "loss": 2.8329,
      "theoretical_loss": 3.5491104269415525,
      "tokens_seen": 1350500352
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029839518555667,
      "loss": 2.727,
      "theoretical_loss": 3.549094934639392,
      "tokens_seen": 1350565888
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002983851554663992,
      "loss": 2.9285,
      "theoretical_loss": 3.549079443299456,
      "tokens_seen": 1350631424
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002983751253761284,
      "loss": 2.8137,
      "theoretical_loss": 3.549063952921639,
      "tokens_seen": 1350696960
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002983650952858576,
      "loss": 2.8374,
      "theoretical_loss": 3.5490484635058337,
      "tokens_seen": 1350762496
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029835506519558676,
      "loss": 2.8905,
      "theoretical_loss": 3.549032975051934,
      "tokens_seen": 1350828032
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029834503510531594,
      "loss": 2.9202,
      "theoretical_loss": 3.549017487559833,
      "tokens_seen": 1350893568
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002983350050150451,
      "loss": 2.7875,
      "theoretical_loss": 3.549002001029425,
      "tokens_seen": 1350959104
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029832497492477436,
      "loss": 2.8422,
      "theoretical_loss": 3.548986515460604,
      "tokens_seen": 1351024640
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002983149448345035,
      "loss": 3.0333,
      "theoretical_loss": 3.548971030853263,
      "tokens_seen": 1351090176
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002983049147442327,
      "loss": 2.7616,
      "theoretical_loss": 3.548955547207296,
      "tokens_seen": 1351155712
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029829488465396185,
      "loss": 3.0085,
      "theoretical_loss": 3.548940064522596,
      "tokens_seen": 1351221248
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002982848545636911,
      "loss": 2.9112,
      "theoretical_loss": 3.5489245827990574,
      "tokens_seen": 1351286784
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029827482447342026,
      "loss": 2.7475,
      "theoretical_loss": 3.5489091020365735,
      "tokens_seen": 1351352320
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029826479438314945,
      "loss": 3.023,
      "theoretical_loss": 3.5488936222350382,
      "tokens_seen": 1351417856
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002982547642928786,
      "loss": 2.8226,
      "theoretical_loss": 3.5488781433943455,
      "tokens_seen": 1351483392
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029824473420260786,
      "loss": 2.6705,
      "theoretical_loss": 3.548862665514389,
      "tokens_seen": 1351548928
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.000298234704112337,
      "loss": 3.0667,
      "theoretical_loss": 3.5488471885950625,
      "tokens_seen": 1351614464
    },
    {
      "epoch": 0.41,
      "objective/train/advantage_avg": 0.49586939811706543,
      "objective/train/docs_used": 765372,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.2697947025299072,
      "objective/train/original_loss": 2.269794464111328,
      "objective/train/theoretical_loss": 3.5488317126362596,
      "objective/train/tokens_used": 1372140000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24592556059360504,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0508370399475098,
      "objective/train/weighted_lm_loss": 2.385178565979004,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0417574644088745,
      "theoretical_loss": 3.5488317126362596,
      "tokens_seen": 1351680000
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002982246740220662,
      "loss": 2.498,
      "theoretical_loss": 3.5488317126362596,
      "tokens_seen": 1351680000
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029821464393179535,
      "loss": 2.9262,
      "theoretical_loss": 3.5488162376378742,
      "tokens_seen": 1351745536
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002982046138415246,
      "loss": 2.6358,
      "theoretical_loss": 3.5488007635998007,
      "tokens_seen": 1351811072
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029819458375125377,
      "loss": 2.9878,
      "theoretical_loss": 3.5487852905219324,
      "tokens_seen": 1351876608
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029818455366098295,
      "loss": 2.8229,
      "theoretical_loss": 3.5487698184041627,
      "tokens_seen": 1351942144
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029817452357071213,
      "loss": 2.826,
      "theoretical_loss": 3.548754347246386,
      "tokens_seen": 1352007680
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002981644934804413,
      "loss": 2.7683,
      "theoretical_loss": 3.548738877048497,
      "tokens_seen": 1352073216
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002981544633901705,
      "loss": 2.9007,
      "theoretical_loss": 3.5487234078103884,
      "tokens_seen": 1352138752
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029814443329989973,
      "loss": 2.8235,
      "theoretical_loss": 3.5487079395319547,
      "tokens_seen": 1352204288
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029813440320962885,
      "loss": 3.1799,
      "theoretical_loss": 3.54869247221309,
      "tokens_seen": 1352269824
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002981243731193581,
      "loss": 3.0134,
      "theoretical_loss": 3.5486770058536874,
      "tokens_seen": 1352335360
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002981143430290872,
      "loss": 2.7518,
      "theoretical_loss": 3.548661540453642,
      "tokens_seen": 1352400896
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029810431293881645,
      "loss": 2.8556,
      "theoretical_loss": 3.5486460760128473,
      "tokens_seen": 1352466432
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029809428284854563,
      "loss": 2.5115,
      "theoretical_loss": 3.548630612531197,
      "tokens_seen": 1352531968
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002980842527582748,
      "loss": 2.8313,
      "theoretical_loss": 3.548615150008586,
      "tokens_seen": 1352597504
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.000298074222668004,
      "loss": 2.768,
      "theoretical_loss": 3.5485996884449076,
      "tokens_seen": 1352663040
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029806419257773323,
      "loss": 2.7021,
      "theoretical_loss": 3.548584227840056,
      "tokens_seen": 1352728576
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029805416248746236,
      "loss": 2.9332,
      "theoretical_loss": 3.5485687681939257,
      "tokens_seen": 1352794112
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002980441323971916,
      "loss": 2.8812,
      "theoretical_loss": 3.5485533095064103,
      "tokens_seen": 1352859648
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002980341023069208,
      "loss": 2.8268,
      "theoretical_loss": 3.548537851777404,
      "tokens_seen": 1352925184
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029802407221664996,
      "loss": 2.7848,
      "theoretical_loss": 3.5485223950068017,
      "tokens_seen": 1352990720
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002980140421263792,
      "loss": 2.8893,
      "theoretical_loss": 3.5485069391944966,
      "tokens_seen": 1353056256
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002980040120361083,
      "loss": 3.0181,
      "theoretical_loss": 3.548491484340383,
      "tokens_seen": 1353121792
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029799398194583755,
      "loss": 2.9265,
      "theoretical_loss": 3.5484760304443554,
      "tokens_seen": 1353187328
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002979839518555667,
      "loss": 2.8853,
      "theoretical_loss": 3.5484605775063085,
      "tokens_seen": 1353252864
    },
    {
      "epoch": 0.41,
      "objective/train/advantage_avg": 0.4941771626472473,
      "objective/train/docs_used": 766365,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.7179901599884033,
      "objective/train/original_loss": 2.717989683151245,
      "objective/train/theoretical_loss": 3.5484451255261353,
      "objective/train/tokens_used": 1373778400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24428781867027283,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506595373153687,
      "objective/train/weighted_lm_loss": 2.8556110858917236,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0363596677780151,
      "theoretical_loss": 3.5484451255261353,
      "tokens_seen": 1353318400
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002979739217652959,
      "loss": 2.6898,
      "theoretical_loss": 3.5484451255261353,
      "tokens_seen": 1353318400
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002979638916750251,
      "loss": 2.8517,
      "theoretical_loss": 3.548429674503731,
      "tokens_seen": 1353383936
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002979538615847543,
      "loss": 2.783,
      "theoretical_loss": 3.54841422443899,
      "tokens_seen": 1353449472
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029794383149448346,
      "loss": 3.1019,
      "theoretical_loss": 3.5483987753318056,
      "tokens_seen": 1353515008
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002979338014042127,
      "loss": 2.7833,
      "theoretical_loss": 3.548383327182073,
      "tokens_seen": 1353580544
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002979237713139418,
      "loss": 2.7645,
      "theoretical_loss": 3.5483678799896854,
      "tokens_seen": 1353646080
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029791374122367106,
      "loss": 2.7803,
      "theoretical_loss": 3.5483524337545385,
      "tokens_seen": 1353711616
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002979037111334002,
      "loss": 2.7777,
      "theoretical_loss": 3.5483369884765255,
      "tokens_seen": 1353777152
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002978936810431294,
      "loss": 3.0309,
      "theoretical_loss": 3.5483215441555416,
      "tokens_seen": 1353842688
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002978836509528586,
      "loss": 2.7923,
      "theoretical_loss": 3.548306100791481,
      "tokens_seen": 1353908224
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002978736208625878,
      "loss": 2.7011,
      "theoretical_loss": 3.548290658384238,
      "tokens_seen": 1353973760
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029786359077231696,
      "loss": 2.6775,
      "theoretical_loss": 3.548275216933707,
      "tokens_seen": 1354039296
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029785356068204614,
      "loss": 2.9073,
      "theoretical_loss": 3.5482597764397816,
      "tokens_seen": 1354104832
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002978435305917753,
      "loss": 2.9996,
      "theoretical_loss": 3.548244336902358,
      "tokens_seen": 1354170368
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029783350050150456,
      "loss": 2.5053,
      "theoretical_loss": 3.548228898321329,
      "tokens_seen": 1354235904
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002978234704112337,
      "loss": 2.8622,
      "theoretical_loss": 3.54821346069659,
      "tokens_seen": 1354301440
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002978134403209629,
      "loss": 2.701,
      "theoretical_loss": 3.548198024028035,
      "tokens_seen": 1354366976
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029780341023069205,
      "loss": 2.8305,
      "theoretical_loss": 3.5481825883155595,
      "tokens_seen": 1354432512
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002977933801404213,
      "loss": 2.8654,
      "theoretical_loss": 3.5481671535590564,
      "tokens_seen": 1354498048
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029778335005015046,
      "loss": 3.0525,
      "theoretical_loss": 3.5481517197584216,
      "tokens_seen": 1354563584
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029777331995987965,
      "loss": 2.5219,
      "theoretical_loss": 3.5481362869135493,
      "tokens_seen": 1354629120
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029776328986960883,
      "loss": 2.7986,
      "theoretical_loss": 3.548120855024334,
      "tokens_seen": 1354694656
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029775325977933806,
      "loss": 2.7738,
      "theoretical_loss": 3.5481054240906698,
      "tokens_seen": 1354760192
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002977432296890672,
      "loss": 2.6052,
      "theoretical_loss": 3.5480899941124524,
      "tokens_seen": 1354825728
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002977331995987964,
      "loss": 2.6652,
      "theoretical_loss": 3.5480745650895757,
      "tokens_seen": 1354891264
    },
    {
      "epoch": 0.41,
      "objective/train/advantage_avg": 0.4946807324886322,
      "objective/train/docs_used": 767078,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.775590181350708,
      "objective/train/original_loss": 2.77558970451355,
      "objective/train/theoretical_loss": 3.5480591370219345,
      "objective/train/tokens_used": 1375416800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24477756023406982,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.05071222782135,
      "objective/train/weighted_lm_loss": 2.9162235260009766,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 3.5480591370219345,
      "tokens_seen": 1354956800
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029772316950852555,
      "loss": 3.0188,
      "theoretical_loss": 3.5480591370219345,
      "tokens_seen": 1354956800
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002977131394182548,
      "loss": 2.8177,
      "theoretical_loss": 3.5480437099094226,
      "tokens_seen": 1355022336
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029770310932798397,
      "loss": 2.5643,
      "theoretical_loss": 3.5480282837519366,
      "tokens_seen": 1355087872
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029769307923771315,
      "loss": 2.6658,
      "theoretical_loss": 3.5480128585493693,
      "tokens_seen": 1355153408
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029768304914744233,
      "loss": 2.5719,
      "theoretical_loss": 3.5479974343016165,
      "tokens_seen": 1355218944
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002976730190571715,
      "loss": 2.6539,
      "theoretical_loss": 3.547982011008573,
      "tokens_seen": 1355284480
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002976629889669007,
      "loss": 2.867,
      "theoretical_loss": 3.547966588670133,
      "tokens_seen": 1355350016
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029765295887662993,
      "loss": 2.9948,
      "theoretical_loss": 3.5479511672861914,
      "tokens_seen": 1355415552
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029764292878635905,
      "loss": 2.6165,
      "theoretical_loss": 3.5479357468566426,
      "tokens_seen": 1355481088
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002976328986960883,
      "loss": 2.4948,
      "theoretical_loss": 3.547920327381382,
      "tokens_seen": 1355546624
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002976228686058174,
      "loss": 2.8653,
      "theoretical_loss": 3.5479049088603043,
      "tokens_seen": 1355612160
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029761283851554665,
      "loss": 2.8817,
      "theoretical_loss": 3.5478894912933043,
      "tokens_seen": 1355677696
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029760280842527583,
      "loss": 2.8375,
      "theoretical_loss": 3.5478740746802773,
      "tokens_seen": 1355743232
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.000297592778335005,
      "loss": 2.6454,
      "theoretical_loss": 3.547858659021117,
      "tokens_seen": 1355808768
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002975827482447342,
      "loss": 2.8918,
      "theoretical_loss": 3.547843244315719,
      "tokens_seen": 1355874304
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029757271815446343,
      "loss": 2.7589,
      "theoretical_loss": 3.5478278305639783,
      "tokens_seen": 1355939840
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029756268806419256,
      "loss": 2.8302,
      "theoretical_loss": 3.54781241776579,
      "tokens_seen": 1356005376
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002975526579739218,
      "loss": 2.8077,
      "theoretical_loss": 3.547797005921048,
      "tokens_seen": 1356070912
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002975426278836509,
      "loss": 2.8245,
      "theoretical_loss": 3.5477815950296483,
      "tokens_seen": 1356136448
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029753259779338016,
      "loss": 2.7177,
      "theoretical_loss": 3.5477661850914854,
      "tokens_seen": 1356201984
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029752256770310934,
      "loss": 2.7968,
      "theoretical_loss": 3.5477507761064544,
      "tokens_seen": 1356267520
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002975125376128385,
      "loss": 2.6711,
      "theoretical_loss": 3.5477353680744503,
      "tokens_seen": 1356333056
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002975025075225677,
      "loss": 2.7575,
      "theoretical_loss": 3.5477199609953685,
      "tokens_seen": 1356398592
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002974924774322969,
      "loss": 2.8308,
      "theoretical_loss": 3.547704554869103,
      "tokens_seen": 1356464128
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029748244734202606,
      "loss": 2.7855,
      "theoretical_loss": 3.5476891496955494,
      "tokens_seen": 1356529664
    },
    {
      "epoch": 0.41,
      "objective/train/advantage_avg": 0.4960591197013855,
      "objective/train/docs_used": 768221,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.428018569946289,
      "objective/train/original_loss": 2.42801833152771,
      "objective/train/theoretical_loss": 3.5476737454746035,
      "objective/train/tokens_used": 1377055200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24610623717308044,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0508570671081543,
      "objective/train/weighted_lm_loss": 2.5515620708465576,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0442860126495361,
      "theoretical_loss": 3.5476737454746035,
      "tokens_seen": 1356595200
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002974724172517553,
      "loss": 2.7956,
      "theoretical_loss": 3.5476737454746035,
      "tokens_seen": 1356595200
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002974623871614844,
      "loss": 2.8183,
      "theoretical_loss": 3.5476583422061587,
      "tokens_seen": 1356660736
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029745235707121366,
      "loss": 2.8813,
      "theoretical_loss": 3.5476429398901117,
      "tokens_seen": 1356726272
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002974423269809428,
      "loss": 3.0281,
      "theoretical_loss": 3.547627538526357,
      "tokens_seen": 1356791808
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.000297432296890672,
      "loss": 2.9919,
      "theoretical_loss": 3.5476121381147894,
      "tokens_seen": 1356857344
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002974222668004012,
      "loss": 3.137,
      "theoretical_loss": 3.5475967386553044,
      "tokens_seen": 1356922880
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002974122367101304,
      "loss": 2.8138,
      "theoretical_loss": 3.5475813401477976,
      "tokens_seen": 1356988416
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029740220661985956,
      "loss": 2.7799,
      "theoretical_loss": 3.547565942592163,
      "tokens_seen": 1357053952
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002973921765295888,
      "loss": 2.7979,
      "theoretical_loss": 3.5475505459882974,
      "tokens_seen": 1357119488
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002973821464393179,
      "loss": 2.6886,
      "theoretical_loss": 3.5475351503360946,
      "tokens_seen": 1357185024
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029737211634904716,
      "loss": 2.928,
      "theoretical_loss": 3.5475197556354505,
      "tokens_seen": 1357250560
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002973620862587763,
      "loss": 2.7989,
      "theoretical_loss": 3.5475043618862596,
      "tokens_seen": 1357316096
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002973520561685055,
      "loss": 2.8331,
      "theoretical_loss": 3.5474889690884184,
      "tokens_seen": 1357381632
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002973420260782347,
      "loss": 2.754,
      "theoretical_loss": 3.5474735772418216,
      "tokens_seen": 1357447168
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002973319959879639,
      "loss": 2.8634,
      "theoretical_loss": 3.547458186346364,
      "tokens_seen": 1357512704
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029732196589769307,
      "loss": 2.8146,
      "theoretical_loss": 3.547442796401941,
      "tokens_seen": 1357578240
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029731193580742225,
      "loss": 3.0634,
      "theoretical_loss": 3.547427407408449,
      "tokens_seen": 1357643776
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029730190571715143,
      "loss": 2.6295,
      "theoretical_loss": 3.5474120193657823,
      "tokens_seen": 1357709312
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029729187562688067,
      "loss": 2.7617,
      "theoretical_loss": 3.547396632273837,
      "tokens_seen": 1357774848
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029728184553660985,
      "loss": 2.7644,
      "theoretical_loss": 3.547381246132507,
      "tokens_seen": 1357840384
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029727181544633903,
      "loss": 2.8013,
      "theoretical_loss": 3.5473658609416896,
      "tokens_seen": 1357905920
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029726178535606826,
      "loss": 2.9226,
      "theoretical_loss": 3.547350476701279,
      "tokens_seen": 1357971456
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002972517552657974,
      "loss": 2.9778,
      "theoretical_loss": 3.547335093411171,
      "tokens_seen": 1358036992
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002972417251755266,
      "loss": 2.7786,
      "theoretical_loss": 3.5473197110712613,
      "tokens_seen": 1358102528
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029723169508525575,
      "loss": 2.6847,
      "theoretical_loss": 3.547304329681445,
      "tokens_seen": 1358168064
    },
    {
      "epoch": 0.41,
      "objective/train/advantage_avg": 0.4936213493347168,
      "objective/train/docs_used": 768734,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.7533702850341797,
      "objective/train/original_loss": 2.7533702850341797,
      "objective/train/theoretical_loss": 3.5472889492416173,
      "objective/train/tokens_used": 1378693600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2437574416399002,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506012439727783,
      "objective/train/weighted_lm_loss": 2.8927090167999268,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0363596677780151,
      "theoretical_loss": 3.5472889492416173,
      "tokens_seen": 1358233600
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.000297221664994985,
      "loss": 3.0185,
      "theoretical_loss": 3.5472889492416173,
      "tokens_seen": 1358233600
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029721163490471417,
      "loss": 2.6629,
      "theoretical_loss": 3.547273569751674,
      "tokens_seen": 1358299136
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029720160481444335,
      "loss": 2.9413,
      "theoretical_loss": 3.547258191211511,
      "tokens_seen": 1358364672
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029719157472417253,
      "loss": 2.8461,
      "theoretical_loss": 3.5472428136210237,
      "tokens_seen": 1358430208
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002971815446339017,
      "loss": 2.7914,
      "theoretical_loss": 3.5472274369801067,
      "tokens_seen": 1358495744
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002971715145436309,
      "loss": 2.9517,
      "theoretical_loss": 3.547212061288657,
      "tokens_seen": 1358561280
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029716148445336013,
      "loss": 2.6327,
      "theoretical_loss": 3.547196686546569,
      "tokens_seen": 1358626816
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029715145436308926,
      "loss": 2.7183,
      "theoretical_loss": 3.5471813127537386,
      "tokens_seen": 1358692352
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002971414242728185,
      "loss": 2.7467,
      "theoretical_loss": 3.547165939910062,
      "tokens_seen": 1358757888
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002971313941825476,
      "loss": 2.7175,
      "theoretical_loss": 3.547150568015434,
      "tokens_seen": 1358823424
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029712136409227685,
      "loss": 2.7959,
      "theoretical_loss": 3.547135197069751,
      "tokens_seen": 1358888960
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029711133400200603,
      "loss": 2.7381,
      "theoretical_loss": 3.5471198270729083,
      "tokens_seen": 1358954496
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002971013039117352,
      "loss": 2.9399,
      "theoretical_loss": 3.5471044580248012,
      "tokens_seen": 1359020032
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002970912738214644,
      "loss": 2.8411,
      "theoretical_loss": 3.5470890899253256,
      "tokens_seen": 1359085568
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029708124373119363,
      "loss": 2.9227,
      "theoretical_loss": 3.547073722774378,
      "tokens_seen": 1359151104
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029707121364092276,
      "loss": 2.6273,
      "theoretical_loss": 3.547058356571853,
      "tokens_seen": 1359216640
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.000297061183550652,
      "loss": 3.1104,
      "theoretical_loss": 3.547042991317647,
      "tokens_seen": 1359282176
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002970511534603811,
      "loss": 2.8809,
      "theoretical_loss": 3.547027627011656,
      "tokens_seen": 1359347712
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029704112337011036,
      "loss": 2.4527,
      "theoretical_loss": 3.5470122636537744,
      "tokens_seen": 1359413248
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029703109327983954,
      "loss": 2.8326,
      "theoretical_loss": 3.5469969012438995,
      "tokens_seen": 1359478784
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002970210631895687,
      "loss": 2.7152,
      "theoretical_loss": 3.5469815397819264,
      "tokens_seen": 1359544320
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002970110330992979,
      "loss": 2.8501,
      "theoretical_loss": 3.546966179267751,
      "tokens_seen": 1359609856
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002970010030090271,
      "loss": 3.0159,
      "theoretical_loss": 3.5469508197012694,
      "tokens_seen": 1359675392
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029699097291875626,
      "loss": 2.8186,
      "theoretical_loss": 3.5469354610823776,
      "tokens_seen": 1359740928
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002969809428284855,
      "loss": 2.9901,
      "theoretical_loss": 3.5469201034109705,
      "tokens_seen": 1359806464
    },
    {
      "epoch": 0.41,
      "objective/train/advantage_avg": 0.49728667736053467,
      "objective/train/docs_used": 769315,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.2374396324157715,
      "objective/train/original_loss": 2.2374393939971924,
      "objective/train/theoretical_loss": 3.5469047466869448,
      "objective/train/tokens_used": 1380332000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24731874465942383,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0509860515594482,
      "objective/train/weighted_lm_loss": 2.351376533508301,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0432155132293701,
      "theoretical_loss": 3.5469047466869448,
      "tokens_seen": 1359872000
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002969709127382146,
      "loss": 2.7593,
      "theoretical_loss": 3.5469047466869448,
      "tokens_seen": 1359872000
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029696088264794386,
      "loss": 2.8198,
      "theoretical_loss": 3.546889390910197,
      "tokens_seen": 1359937536
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.000296950852557673,
      "loss": 2.6849,
      "theoretical_loss": 3.546874036080621,
      "tokens_seen": 1360003072
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002969408224674022,
      "loss": 2.7879,
      "theoretical_loss": 3.5468586821981147,
      "tokens_seen": 1360068608
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002969307923771314,
      "loss": 2.9148,
      "theoretical_loss": 3.546843329262573,
      "tokens_seen": 1360134144
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002969207622868606,
      "loss": 2.8317,
      "theoretical_loss": 3.5468279772738924,
      "tokens_seen": 1360199680
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029691073219658976,
      "loss": 2.7458,
      "theoretical_loss": 3.5468126262319686,
      "tokens_seen": 1360265216
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.000296900702106319,
      "loss": 2.7654,
      "theoretical_loss": 3.546797276136698,
      "tokens_seen": 1360330752
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002968906720160481,
      "loss": 2.7484,
      "theoretical_loss": 3.5467819269879763,
      "tokens_seen": 1360396288
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029688064192577736,
      "loss": 2.9133,
      "theoretical_loss": 3.5467665787856997,
      "tokens_seen": 1360461824
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002968706118355065,
      "loss": 3.0732,
      "theoretical_loss": 3.546751231529764,
      "tokens_seen": 1360527360
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002968605817452357,
      "loss": 2.581,
      "theoretical_loss": 3.546735885220065,
      "tokens_seen": 1360592896
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002968505516549649,
      "loss": 2.6893,
      "theoretical_loss": 3.5467205398564996,
      "tokens_seen": 1360658432
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002968405215646941,
      "loss": 2.891,
      "theoretical_loss": 3.546705195438963,
      "tokens_seen": 1360723968
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029683049147442327,
      "loss": 2.6003,
      "theoretical_loss": 3.5466898519673524,
      "tokens_seen": 1360789504
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029682046138415245,
      "loss": 2.8451,
      "theoretical_loss": 3.546674509441563,
      "tokens_seen": 1360855040
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029681043129388163,
      "loss": 2.8204,
      "theoretical_loss": 3.546659167861491,
      "tokens_seen": 1360920576
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029680040120361087,
      "loss": 2.8155,
      "theoretical_loss": 3.5466438272270335,
      "tokens_seen": 1360986112
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029679037111334,
      "loss": 2.7479,
      "theoretical_loss": 3.5466284875380856,
      "tokens_seen": 1361051648
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029678034102306923,
      "loss": 3.0437,
      "theoretical_loss": 3.546613148794544,
      "tokens_seen": 1361117184
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002967703109327984,
      "loss": 2.6676,
      "theoretical_loss": 3.546597810996305,
      "tokens_seen": 1361182720
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002967602808425276,
      "loss": 3.0727,
      "theoretical_loss": 3.546582474143265,
      "tokens_seen": 1361248256
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029675025075225677,
      "loss": 2.6815,
      "theoretical_loss": 3.546567138235319,
      "tokens_seen": 1361313792
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029674022066198595,
      "loss": 2.9356,
      "theoretical_loss": 3.546551803272365,
      "tokens_seen": 1361379328
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029673019057171513,
      "loss": 2.8007,
      "theoretical_loss": 3.546536469254298,
      "tokens_seen": 1361444864
    },
    {
      "epoch": 0.41,
      "objective/train/advantage_avg": 0.49294084310531616,
      "objective/train/docs_used": 770761,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.609768867492676,
      "objective/train/original_loss": 2.609769105911255,
      "objective/train/theoretical_loss": 3.546521136181015,
      "objective/train/tokens_used": 1381970400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24306774139404297,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505295991897583,
      "objective/train/weighted_lm_loss": 2.7417497634887695,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0304546356201172,
      "theoretical_loss": 3.546521136181015,
      "tokens_seen": 1361510400
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029672016048144437,
      "loss": 2.8204,
      "theoretical_loss": 3.546521136181015,
      "tokens_seen": 1361510400
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002967101303911735,
      "loss": 2.8733,
      "theoretical_loss": 3.546505804052413,
      "tokens_seen": 1361575936
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029670010030090273,
      "loss": 2.7904,
      "theoretical_loss": 3.546490472868386,
      "tokens_seen": 1361641472
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029669007021063186,
      "loss": 3.1376,
      "theoretical_loss": 3.5464751426288323,
      "tokens_seen": 1361707008
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002966800401203611,
      "loss": 2.7212,
      "theoretical_loss": 3.546459813333648,
      "tokens_seen": 1361772544
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002966700100300903,
      "loss": 2.8174,
      "theoretical_loss": 3.5464444849827292,
      "tokens_seen": 1361838080
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029665997993981946,
      "loss": 2.8202,
      "theoretical_loss": 3.5464291575759725,
      "tokens_seen": 1361903616
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029664994984954864,
      "loss": 2.8533,
      "theoretical_loss": 3.5464138311132736,
      "tokens_seen": 1361969152
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002966399197592778,
      "loss": 3.0297,
      "theoretical_loss": 3.5463985055945297,
      "tokens_seen": 1362034688
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.000296629889669007,
      "loss": 2.7542,
      "theoretical_loss": 3.546383181019637,
      "tokens_seen": 1362100224
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029661985957873623,
      "loss": 2.6506,
      "theoretical_loss": 3.546367857388492,
      "tokens_seen": 1362165760
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029660982948846536,
      "loss": 2.7992,
      "theoretical_loss": 3.546352534700991,
      "tokens_seen": 1362231296
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002965997993981946,
      "loss": 2.8296,
      "theoretical_loss": 3.546337212957031,
      "tokens_seen": 1362296832
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002965897693079238,
      "loss": 2.7349,
      "theoretical_loss": 3.546321892156508,
      "tokens_seen": 1362362368
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029657973921765296,
      "loss": 2.7542,
      "theoretical_loss": 3.5463065722993186,
      "tokens_seen": 1362427904
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029656970912738214,
      "loss": 2.9248,
      "theoretical_loss": 3.54629125338536,
      "tokens_seen": 1362493440
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002965596790371113,
      "loss": 2.7786,
      "theoretical_loss": 3.5462759354145277,
      "tokens_seen": 1362558976
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002965496489468405,
      "loss": 2.9504,
      "theoretical_loss": 3.546260618386719,
      "tokens_seen": 1362624512
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029653961885656974,
      "loss": 2.6091,
      "theoretical_loss": 3.5462453023018297,
      "tokens_seen": 1362690048
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002965295887662989,
      "loss": 2.6784,
      "theoretical_loss": 3.546229987159758,
      "tokens_seen": 1362755584
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002965195586760281,
      "loss": 2.9067,
      "theoretical_loss": 3.5462146729603985,
      "tokens_seen": 1362821120
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002965095285857573,
      "loss": 2.889,
      "theoretical_loss": 3.546199359703649,
      "tokens_seen": 1362886656
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029649949849548646,
      "loss": 2.9291,
      "theoretical_loss": 3.5461840473894064,
      "tokens_seen": 1362952192
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002964894684052157,
      "loss": 2.6727,
      "theoretical_loss": 3.546168736017567,
      "tokens_seen": 1363017728
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002964794383149448,
      "loss": 2.6895,
      "theoretical_loss": 3.5461534255880274,
      "tokens_seen": 1363083264
    },
    {
      "epoch": 0.41,
      "objective/train/advantage_avg": 0.4941280484199524,
      "objective/train/docs_used": 771430,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.7495861053466797,
      "objective/train/original_loss": 2.749586343765259,
      "objective/train/theoretical_loss": 3.5461381161006846,
      "objective/train/tokens_used": 1383608800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2442336231470108,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506542921066284,
      "objective/train/weighted_lm_loss": 2.8889079093933105,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0338950157165527,
      "theoretical_loss": 3.5461381161006846,
      "tokens_seen": 1363148800
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029646940822467406,
      "loss": 2.755,
      "theoretical_loss": 3.5461381161006846,
      "tokens_seen": 1363148800
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002964593781344032,
      "loss": 3.0268,
      "theoretical_loss": 3.5461228075554345,
      "tokens_seen": 1363214336
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002964493480441324,
      "loss": 2.5676,
      "theoretical_loss": 3.546107499952175,
      "tokens_seen": 1363279872
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002964393179538616,
      "loss": 2.9006,
      "theoretical_loss": 3.546092193290802,
      "tokens_seen": 1363345408
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002964292878635908,
      "loss": 2.7242,
      "theoretical_loss": 3.5460768875712128,
      "tokens_seen": 1363410944
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029641925777331996,
      "loss": 2.8799,
      "theoretical_loss": 3.546061582793304,
      "tokens_seen": 1363476480
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002964092276830492,
      "loss": 2.838,
      "theoretical_loss": 3.5460462789569718,
      "tokens_seen": 1363542016
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002963991975927783,
      "loss": 2.9212,
      "theoretical_loss": 3.5460309760621143,
      "tokens_seen": 1363607552
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029638916750250756,
      "loss": 2.5593,
      "theoretical_loss": 3.5460156741086273,
      "tokens_seen": 1363673088
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002963791374122367,
      "loss": 2.5919,
      "theoretical_loss": 3.5460003730964083,
      "tokens_seen": 1363738624
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002963691073219659,
      "loss": 2.7148,
      "theoretical_loss": 3.5459850730253537,
      "tokens_seen": 1363804160
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002963590772316951,
      "loss": 2.6742,
      "theoretical_loss": 3.54596977389536,
      "tokens_seen": 1363869696
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002963490471414243,
      "loss": 2.7467,
      "theoretical_loss": 3.545954475706325,
      "tokens_seen": 1363935232
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029633901705115347,
      "loss": 2.8211,
      "theoretical_loss": 3.5459391784581458,
      "tokens_seen": 1364000768
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029632898696088265,
      "loss": 2.9238,
      "theoretical_loss": 3.545923882150718,
      "tokens_seen": 1364066304
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029631895687061183,
      "loss": 2.7219,
      "theoretical_loss": 3.54590858678394,
      "tokens_seen": 1364131840
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029630892678034107,
      "loss": 2.5754,
      "theoretical_loss": 3.5458932923577082,
      "tokens_seen": 1364197376
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002962988966900702,
      "loss": 2.7122,
      "theoretical_loss": 3.545877998871919,
      "tokens_seen": 1364262912
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029628886659979943,
      "loss": 3.1641,
      "theoretical_loss": 3.54586270632647,
      "tokens_seen": 1364328448
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002962788365095286,
      "loss": 3.0673,
      "theoretical_loss": 3.5458474147212584,
      "tokens_seen": 1364393984
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002962688064192578,
      "loss": 2.9385,
      "theoretical_loss": 3.5458321240561803,
      "tokens_seen": 1364459520
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029625877632898697,
      "loss": 2.8129,
      "theoretical_loss": 3.5458168343311343,
      "tokens_seen": 1364525056
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029624874623871615,
      "loss": 2.7001,
      "theoretical_loss": 3.5458015455460155,
      "tokens_seen": 1364590592
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029623871614844533,
      "loss": 2.9792,
      "theoretical_loss": 3.545786257700723,
      "tokens_seen": 1364656128
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029622868605817457,
      "loss": 2.7858,
      "theoretical_loss": 3.545770970795153,
      "tokens_seen": 1364721664
    },
    {
      "epoch": 0.41,
      "objective/train/advantage_avg": 0.49373528361320496,
      "objective/train/docs_used": 772731,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.4841468334198,
      "objective/train/original_loss": 2.484147548675537,
      "objective/train/theoretical_loss": 3.545755684829202,
      "objective/train/tokens_used": 1385247200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24388869106769562,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050613284111023,
      "objective/train/weighted_lm_loss": 2.6103320121765137,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0304546356201172,
      "theoretical_loss": 3.545755684829202,
      "tokens_seen": 1364787200
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002962186559679037,
      "loss": 2.8262,
      "theoretical_loss": 3.545755684829202,
      "tokens_seen": 1364787200
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029620862587763293,
      "loss": 2.7448,
      "theoretical_loss": 3.5457403998027677,
      "tokens_seen": 1364852736
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029619859578736206,
      "loss": 2.9667,
      "theoretical_loss": 3.5457251157157472,
      "tokens_seen": 1364918272
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002961885656970913,
      "loss": 2.7682,
      "theoretical_loss": 3.545709832568038,
      "tokens_seen": 1364983808
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002961785356068205,
      "loss": 2.749,
      "theoretical_loss": 3.545694550359537,
      "tokens_seen": 1365049344
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029616850551654966,
      "loss": 2.8727,
      "theoretical_loss": 3.5456792690901415,
      "tokens_seen": 1365114880
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029615847542627884,
      "loss": 2.9217,
      "theoretical_loss": 3.545663988759748,
      "tokens_seen": 1365180416
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.000296148445336008,
      "loss": 2.6982,
      "theoretical_loss": 3.5456487093682547,
      "tokens_seen": 1365245952
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002961384152457372,
      "loss": 2.9154,
      "theoretical_loss": 3.545633430915559,
      "tokens_seen": 1365311488
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029612838515546643,
      "loss": 2.9739,
      "theoretical_loss": 3.5456181534015565,
      "tokens_seen": 1365377024
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029611835506519556,
      "loss": 2.7045,
      "theoretical_loss": 3.5456028768261465,
      "tokens_seen": 1365442560
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002961083249749248,
      "loss": 2.6197,
      "theoretical_loss": 3.545587601189225,
      "tokens_seen": 1365508096
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.000296098294884654,
      "loss": 2.7864,
      "theoretical_loss": 3.5455723264906895,
      "tokens_seen": 1365573632
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029608826479438316,
      "loss": 2.8252,
      "theoretical_loss": 3.545557052730438,
      "tokens_seen": 1365639168
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029607823470411234,
      "loss": 2.8076,
      "theoretical_loss": 3.545541779908367,
      "tokens_seen": 1365704704
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002960682046138415,
      "loss": 2.9782,
      "theoretical_loss": 3.5455265080243743,
      "tokens_seen": 1365770240
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002960581745235707,
      "loss": 2.6069,
      "theoretical_loss": 3.5455112370783572,
      "tokens_seen": 1365835776
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029604814443329994,
      "loss": 2.6258,
      "theoretical_loss": 3.545495967070213,
      "tokens_seen": 1365901312
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029603811434302906,
      "loss": 2.7793,
      "theoretical_loss": 3.5454806979998392,
      "tokens_seen": 1365966848
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002960280842527583,
      "loss": 2.7237,
      "theoretical_loss": 3.545465429867133,
      "tokens_seen": 1366032384
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002960180541624874,
      "loss": 2.7834,
      "theoretical_loss": 3.5454501626719925,
      "tokens_seen": 1366097920
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029600802407221666,
      "loss": 2.771,
      "theoretical_loss": 3.5454348964143145,
      "tokens_seen": 1366163456
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029599799398194584,
      "loss": 3.1038,
      "theoretical_loss": 3.5454196310939965,
      "tokens_seen": 1366228992
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.000295987963891675,
      "loss": 2.6395,
      "theoretical_loss": 3.5454043667109367,
      "tokens_seen": 1366294528
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002959779338014042,
      "loss": 2.7526,
      "theoretical_loss": 3.5453891032650313,
      "tokens_seen": 1366360064
    },
    {
      "epoch": 0.41,
      "objective/train/advantage_avg": 0.4954284727573395,
      "objective/train/docs_used": 773461,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.180568218231201,
      "objective/train/original_loss": 2.180567979812622,
      "objective/train/theoretical_loss": 3.545373840756179,
      "objective/train/tokens_used": 1386885600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24549172818660736,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050790786743164,
      "objective/train/weighted_lm_loss": 2.2912471294403076,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0425469875335693,
      "theoretical_loss": 3.545373840756179,
      "tokens_seen": 1366425600
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002959679037111334,
      "loss": 2.5161,
      "theoretical_loss": 3.545373840756179,
      "tokens_seen": 1366425600
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029595787362086257,
      "loss": 2.8303,
      "theoretical_loss": 3.5453585791842768,
      "tokens_seen": 1366491136
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002959478435305918,
      "loss": 2.8328,
      "theoretical_loss": 3.5453433185492225,
      "tokens_seen": 1366556672
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029593781344032093,
      "loss": 2.9988,
      "theoretical_loss": 3.545328058850913,
      "tokens_seen": 1366622208
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029592778335005016,
      "loss": 2.8058,
      "theoretical_loss": 3.5453128000892473,
      "tokens_seen": 1366687744
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029591775325977935,
      "loss": 2.8255,
      "theoretical_loss": 3.5452975422641213,
      "tokens_seen": 1366753280
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029590772316950853,
      "loss": 2.7383,
      "theoretical_loss": 3.5452822853754338,
      "tokens_seen": 1366818816
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002958976930792377,
      "loss": 2.8685,
      "theoretical_loss": 3.5452670294230817,
      "tokens_seen": 1366884352
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002958876629889669,
      "loss": 2.8083,
      "theoretical_loss": 3.5452517744069634,
      "tokens_seen": 1366949888
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029587763289869607,
      "loss": 2.8774,
      "theoretical_loss": 3.545236520326976,
      "tokens_seen": 1367015424
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002958676028084253,
      "loss": 2.9419,
      "theoretical_loss": 3.5452212671830168,
      "tokens_seen": 1367080960
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029585757271815443,
      "loss": 3.0326,
      "theoretical_loss": 3.545206014974985,
      "tokens_seen": 1367146496
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029584754262788367,
      "loss": 2.7074,
      "theoretical_loss": 3.5451907637027764,
      "tokens_seen": 1367212032
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002958375125376128,
      "loss": 2.8123,
      "theoretical_loss": 3.5451755133662903,
      "tokens_seen": 1367277568
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029582748244734203,
      "loss": 2.8203,
      "theoretical_loss": 3.545160263965424,
      "tokens_seen": 1367343104
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002958174523570712,
      "loss": 2.8091,
      "theoretical_loss": 3.5451450155000748,
      "tokens_seen": 1367408640
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002958074222668004,
      "loss": 2.7847,
      "theoretical_loss": 3.5451297679701406,
      "tokens_seen": 1367474176
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002957973921765296,
      "loss": 2.7601,
      "theoretical_loss": 3.545114521375519,
      "tokens_seen": 1367539712
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002957873620862588,
      "loss": 3.0892,
      "theoretical_loss": 3.545099275716109,
      "tokens_seen": 1367605248
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.000295777331995988,
      "loss": 2.8509,
      "theoretical_loss": 3.545084030991807,
      "tokens_seen": 1367670784
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029576730190571717,
      "loss": 2.7746,
      "theoretical_loss": 3.5450687872025113,
      "tokens_seen": 1367736320
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029575727181544635,
      "loss": 2.9301,
      "theoretical_loss": 3.54505354434812,
      "tokens_seen": 1367801856
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029574724172517553,
      "loss": 2.6818,
      "theoretical_loss": 3.545038302428531,
      "tokens_seen": 1367867392
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029573721163490477,
      "loss": 2.8009,
      "theoretical_loss": 3.5450230614436418,
      "tokens_seen": 1367932928
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002957271815446339,
      "loss": 2.9113,
      "theoretical_loss": 3.5450078213933507,
      "tokens_seen": 1367998464
    },
    {
      "epoch": 0.41,
      "objective/train/advantage_avg": 0.4943248927593231,
      "objective/train/docs_used": 774789,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.8447489738464355,
      "objective/train/original_loss": 2.8447489738464355,
      "objective/train/theoretical_loss": 3.5449925822775556,
      "objective/train/tokens_used": 1388524000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24442246556282043,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506750345230103,
      "objective/train/weighted_lm_loss": 2.9887642860412598,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0168063640594482,
      "theoretical_loss": 3.5449925822775556,
      "tokens_seen": 1368064000
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029571715145436313,
      "loss": 2.9761,
      "theoretical_loss": 3.5449925822775556,
      "tokens_seen": 1368064000
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029570712136409226,
      "loss": 2.5534,
      "theoretical_loss": 3.544977344096154,
      "tokens_seen": 1368129536
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002956970912738215,
      "loss": 2.7285,
      "theoretical_loss": 3.5449621068490442,
      "tokens_seen": 1368195072
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002956870611835507,
      "loss": 2.7352,
      "theoretical_loss": 3.5449468705361245,
      "tokens_seen": 1368260608
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029567703109327986,
      "loss": 2.8182,
      "theoretical_loss": 3.5449316351572917,
      "tokens_seen": 1368326144
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029566700100300904,
      "loss": 2.9147,
      "theoretical_loss": 3.544916400712445,
      "tokens_seen": 1368391680
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002956569709127382,
      "loss": 2.7798,
      "theoretical_loss": 3.5449011672014823,
      "tokens_seen": 1368457216
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002956469408224674,
      "loss": 2.7425,
      "theoretical_loss": 3.544885934624301,
      "tokens_seen": 1368522752
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029563691073219663,
      "loss": 3.005,
      "theoretical_loss": 3.5448707029807993,
      "tokens_seen": 1368588288
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029562688064192576,
      "loss": 2.8032,
      "theoretical_loss": 3.544855472270876,
      "tokens_seen": 1368653824
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.000295616850551655,
      "loss": 2.9435,
      "theoretical_loss": 3.544840242494428,
      "tokens_seen": 1368719360
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002956068204613842,
      "loss": 2.9579,
      "theoretical_loss": 3.544825013651354,
      "tokens_seen": 1368784896
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029559679037111336,
      "loss": 2.7383,
      "theoretical_loss": 3.544809785741553,
      "tokens_seen": 1368850432
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029558676028084254,
      "loss": 2.891,
      "theoretical_loss": 3.544794558764922,
      "tokens_seen": 1368915968
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002955767301905717,
      "loss": 2.9162,
      "theoretical_loss": 3.5447793327213586,
      "tokens_seen": 1368981504
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002955667001003009,
      "loss": 2.9493,
      "theoretical_loss": 3.5447641076107623,
      "tokens_seen": 1369047040
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029555667001003014,
      "loss": 2.7529,
      "theoretical_loss": 3.544748883433031,
      "tokens_seen": 1369112576
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029554663991975926,
      "loss": 2.7787,
      "theoretical_loss": 3.5447336601880624,
      "tokens_seen": 1369178112
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002955366098294885,
      "loss": 2.5168,
      "theoretical_loss": 3.544718437875755,
      "tokens_seen": 1369243648
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.0002955265797392176,
      "loss": 2.663,
      "theoretical_loss": 3.5447032164960066,
      "tokens_seen": 1369309184
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029551654964894686,
      "loss": 2.8629,
      "theoretical_loss": 3.544687996048716,
      "tokens_seen": 1369374720
    },
    {
      "epoch": 0.41,
      "learning_rate": 0.00029550651955867604,
      "loss": 2.7258,
      "theoretical_loss": 3.5446727765337815,
      "tokens_seen": 1369440256
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002954964894684052,
      "loss": 2.8501,
      "theoretical_loss": 3.5446575579511013,
      "tokens_seen": 1369505792
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002954864593781344,
      "loss": 2.8769,
      "theoretical_loss": 3.544642340300573,
      "tokens_seen": 1369571328
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002954764292878636,
      "loss": 2.6935,
      "theoretical_loss": 3.544627123582096,
      "tokens_seen": 1369636864
    },
    {
      "epoch": 0.42,
      "objective/train/advantage_avg": 0.49293550848960876,
      "objective/train/docs_used": 775516,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.590879201889038,
      "objective/train/original_loss": 2.590879201889038,
      "objective/train/theoretical_loss": 3.5446119077955673,
      "objective/train/tokens_used": 1390162400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2430829256772995,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505292415618896,
      "objective/train/weighted_lm_loss": 2.7216525077819824,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0417574644088745,
      "theoretical_loss": 3.5446119077955673,
      "tokens_seen": 1369702400
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029546639919759277,
      "loss": 2.9395,
      "theoretical_loss": 3.5446119077955673,
      "tokens_seen": 1369702400
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.000295456369107322,
      "loss": 2.7074,
      "theoretical_loss": 3.5445966929408863,
      "tokens_seen": 1369767936
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029544633901705113,
      "loss": 2.6117,
      "theoretical_loss": 3.5445814790179515,
      "tokens_seen": 1369833472
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029543630892678037,
      "loss": 2.7803,
      "theoretical_loss": 3.5445662660266604,
      "tokens_seen": 1369899008
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029542627883650955,
      "loss": 2.8766,
      "theoretical_loss": 3.544551053966912,
      "tokens_seen": 1369964544
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029541624874623873,
      "loss": 2.8498,
      "theoretical_loss": 3.544535842838604,
      "tokens_seen": 1370030080
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002954062186559679,
      "loss": 3.0026,
      "theoretical_loss": 3.544520632641636,
      "tokens_seen": 1370095616
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002953961885656971,
      "loss": 2.9456,
      "theoretical_loss": 3.544505423375905,
      "tokens_seen": 1370161152
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029538615847542627,
      "loss": 3.0097,
      "theoretical_loss": 3.5444902150413107,
      "tokens_seen": 1370226688
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002953761283851555,
      "loss": 2.7968,
      "theoretical_loss": 3.544475007637751,
      "tokens_seen": 1370292224
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029536609829488463,
      "loss": 2.8812,
      "theoretical_loss": 3.5444598011651243,
      "tokens_seen": 1370357760
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029535606820461387,
      "loss": 2.9702,
      "theoretical_loss": 3.5444445956233293,
      "tokens_seen": 1370423296
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.000295346038114343,
      "loss": 2.7063,
      "theoretical_loss": 3.5444293910122644,
      "tokens_seen": 1370488832
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029533600802407223,
      "loss": 2.7976,
      "theoretical_loss": 3.544414187331828,
      "tokens_seen": 1370554368
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002953259779338014,
      "loss": 2.4972,
      "theoretical_loss": 3.5443989845819193,
      "tokens_seen": 1370619904
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002953159478435306,
      "loss": 2.7975,
      "theoretical_loss": 3.544383782762436,
      "tokens_seen": 1370685440
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002953059177532598,
      "loss": 2.8549,
      "theoretical_loss": 3.544368581873277,
      "tokens_seen": 1370750976
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.000295295887662989,
      "loss": 2.5664,
      "theoretical_loss": 3.544353381914341,
      "tokens_seen": 1370816512
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029528585757271814,
      "loss": 2.831,
      "theoretical_loss": 3.5443381828855265,
      "tokens_seen": 1370882048
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029527582748244737,
      "loss": 2.7673,
      "theoretical_loss": 3.5443229847867324,
      "tokens_seen": 1370947584
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002952657973921765,
      "loss": 2.7548,
      "theoretical_loss": 3.5443077876178566,
      "tokens_seen": 1371013120
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029525576730190573,
      "loss": 2.8026,
      "theoretical_loss": 3.544292591378799,
      "tokens_seen": 1371078656
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002952457372116349,
      "loss": 2.7458,
      "theoretical_loss": 3.5442773960694565,
      "tokens_seen": 1371144192
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002952357071213641,
      "loss": 2.948,
      "theoretical_loss": 3.5442622016897296,
      "tokens_seen": 1371209728
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002952256770310933,
      "loss": 2.966,
      "theoretical_loss": 3.5442470082395157,
      "tokens_seen": 1371275264
    },
    {
      "epoch": 0.42,
      "objective/train/advantage_avg": 0.49345752596855164,
      "objective/train/docs_used": 776338,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.2788174152374268,
      "objective/train/original_loss": 3.2788169384002686,
      "objective/train/theoretical_loss": 3.5442318157187147,
      "objective/train/tokens_used": 1391800800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24364914000034332,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505841970443726,
      "objective/train/weighted_lm_loss": 3.4446685314178467,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 0.951229453086853,
      "theoretical_loss": 3.5442318157187147,
      "tokens_seen": 1371340800
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029521564694082246,
      "loss": 2.9255,
      "theoretical_loss": 3.5442318157187147,
      "tokens_seen": 1371340800
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029520561685055164,
      "loss": 2.8326,
      "theoretical_loss": 3.544216624127224,
      "tokens_seen": 1371406336
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002951955867602809,
      "loss": 2.6002,
      "theoretical_loss": 3.544201433464943,
      "tokens_seen": 1371471872
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029518555667001,
      "loss": 2.6414,
      "theoretical_loss": 3.544186243731771,
      "tokens_seen": 1371537408
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029517552657973924,
      "loss": 2.7372,
      "theoretical_loss": 3.544171054927606,
      "tokens_seen": 1371602944
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029516549648946836,
      "loss": 2.7653,
      "theoretical_loss": 3.544155867052347,
      "tokens_seen": 1371668480
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002951554663991976,
      "loss": 3.0426,
      "theoretical_loss": 3.544140680105893,
      "tokens_seen": 1371734016
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002951454363089268,
      "loss": 2.762,
      "theoretical_loss": 3.5441254940881426,
      "tokens_seen": 1371799552
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029513540621865596,
      "loss": 2.7442,
      "theoretical_loss": 3.5441103089989947,
      "tokens_seen": 1371865088
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029512537612838514,
      "loss": 2.8579,
      "theoretical_loss": 3.5440951248383485,
      "tokens_seen": 1371930624
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002951153460381144,
      "loss": 2.8846,
      "theoretical_loss": 3.5440799416061024,
      "tokens_seen": 1371996160
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002951053159478435,
      "loss": 2.9172,
      "theoretical_loss": 3.544064759302155,
      "tokens_seen": 1372061696
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029509528585757274,
      "loss": 2.9261,
      "theoretical_loss": 3.5440495779264065,
      "tokens_seen": 1372127232
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029508525576730187,
      "loss": 2.8849,
      "theoretical_loss": 3.5440343974787547,
      "tokens_seen": 1372192768
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002950752256770311,
      "loss": 3.2415,
      "theoretical_loss": 3.544019217959099,
      "tokens_seen": 1372258304
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002950651955867603,
      "loss": 2.9096,
      "theoretical_loss": 3.5440040393673375,
      "tokens_seen": 1372323840
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029505516549648946,
      "loss": 2.6519,
      "theoretical_loss": 3.5439888617033706,
      "tokens_seen": 1372389376
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029504513540621865,
      "loss": 2.8595,
      "theoretical_loss": 3.543973684967096,
      "tokens_seen": 1372454912
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002950351053159478,
      "loss": 2.884,
      "theoretical_loss": 3.5439585091584136,
      "tokens_seen": 1372520448
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029502507522567706,
      "loss": 2.6665,
      "theoretical_loss": 3.5439433342772224,
      "tokens_seen": 1372585984
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029501504513540624,
      "loss": 2.7568,
      "theoretical_loss": 3.5439281603234205,
      "tokens_seen": 1372651520
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002950050150451354,
      "loss": 2.7529,
      "theoretical_loss": 3.543912987296908,
      "tokens_seen": 1372717056
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002949949849548646,
      "loss": 2.9802,
      "theoretical_loss": 3.5438978151975835,
      "tokens_seen": 1372782592
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002949849548645938,
      "loss": 2.963,
      "theoretical_loss": 3.5438826440253455,
      "tokens_seen": 1372848128
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029497492477432297,
      "loss": 2.8687,
      "theoretical_loss": 3.543867473780094,
      "tokens_seen": 1372913664
    },
    {
      "epoch": 0.42,
      "objective/train/advantage_avg": 0.4950534999370575,
      "objective/train/docs_used": 777606,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.5431559085845947,
      "objective/train/original_loss": 2.5431559085845947,
      "objective/train/theoretical_loss": 3.543852304461728,
      "objective/train/tokens_used": 1393439200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24512939155101776,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507514476776123,
      "objective/train/weighted_lm_loss": 2.672102689743042,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0437887907028198,
      "theoretical_loss": 3.543852304461728,
      "tokens_seen": 1372979200
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002949648946840522,
      "loss": 2.9753,
      "theoretical_loss": 3.543852304461728,
      "tokens_seen": 1372979200
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029495486459378133,
      "loss": 2.9707,
      "theoretical_loss": 3.543837136070147,
      "tokens_seen": 1373044736
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029494483450351057,
      "loss": 2.7979,
      "theoretical_loss": 3.543821968605249,
      "tokens_seen": 1373110272
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029493480441323975,
      "loss": 2.9996,
      "theoretical_loss": 3.5438068020669338,
      "tokens_seen": 1373175808
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029492477432296893,
      "loss": 2.8947,
      "theoretical_loss": 3.5437916364551008,
      "tokens_seen": 1373241344
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002949147442326981,
      "loss": 2.7895,
      "theoretical_loss": 3.5437764717696485,
      "tokens_seen": 1373306880
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002949047141424273,
      "loss": 2.6867,
      "theoretical_loss": 3.5437613080104766,
      "tokens_seen": 1373372416
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029489468405215647,
      "loss": 2.8429,
      "theoretical_loss": 3.5437461451774843,
      "tokens_seen": 1373437952
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002948846539618857,
      "loss": 2.9241,
      "theoretical_loss": 3.543730983270571,
      "tokens_seen": 1373503488
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029487462387161483,
      "loss": 2.8589,
      "theoretical_loss": 3.5437158222896357,
      "tokens_seen": 1373569024
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029486459378134407,
      "loss": 2.8214,
      "theoretical_loss": 3.5437006622345777,
      "tokens_seen": 1373634560
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002948545636910732,
      "loss": 2.7863,
      "theoretical_loss": 3.543685503105296,
      "tokens_seen": 1373700096
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029484453360080243,
      "loss": 2.741,
      "theoretical_loss": 3.543670344901691,
      "tokens_seen": 1373765632
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002948345035105316,
      "loss": 2.6165,
      "theoretical_loss": 3.543655187623661,
      "tokens_seen": 1373831168
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002948244734202608,
      "loss": 2.9564,
      "theoretical_loss": 3.543640031271105,
      "tokens_seen": 1373896704
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029481444332999,
      "loss": 2.8826,
      "theoretical_loss": 3.5436248758439235,
      "tokens_seen": 1373962240
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002948044132397192,
      "loss": 2.9704,
      "theoretical_loss": 3.543609721342015,
      "tokens_seen": 1374027776
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029479438314944834,
      "loss": 2.8233,
      "theoretical_loss": 3.5435945677652794,
      "tokens_seen": 1374093312
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029478435305917757,
      "loss": 2.7629,
      "theoretical_loss": 3.5435794151136157,
      "tokens_seen": 1374158848
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002947743229689067,
      "loss": 2.749,
      "theoretical_loss": 3.5435642633869238,
      "tokens_seen": 1374224384
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029476429287863593,
      "loss": 2.9775,
      "theoretical_loss": 3.543549112585102,
      "tokens_seen": 1374289920
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002947542627883651,
      "loss": 2.8122,
      "theoretical_loss": 3.5435339627080515,
      "tokens_seen": 1374355456
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002947442326980943,
      "loss": 2.5195,
      "theoretical_loss": 3.5435188137556706,
      "tokens_seen": 1374420992
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002947342026078235,
      "loss": 2.9581,
      "theoretical_loss": 3.543503665727859,
      "tokens_seen": 1374486528
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029472417251755266,
      "loss": 2.8338,
      "theoretical_loss": 3.5434885186245157,
      "tokens_seen": 1374552064
    },
    {
      "epoch": 0.42,
      "objective/train/advantage_avg": 0.4929393231868744,
      "objective/train/docs_used": 778164,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.675414562225342,
      "objective/train/original_loss": 2.6754136085510254,
      "objective/train/theoretical_loss": 3.5434733724455407,
      "objective/train/tokens_used": 1395077600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2431073635816574,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505297183990479,
      "objective/train/weighted_lm_loss": 2.810957908630371,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0442860126495361,
      "theoretical_loss": 3.5434733724455407,
      "tokens_seen": 1374617600
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029471414242728184,
      "loss": 2.8801,
      "theoretical_loss": 3.5434733724455407,
      "tokens_seen": 1374617600
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002947041123370111,
      "loss": 2.8627,
      "theoretical_loss": 3.5434582271908344,
      "tokens_seen": 1374683136
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002946940822467402,
      "loss": 2.8734,
      "theoretical_loss": 3.5434430828602945,
      "tokens_seen": 1374748672
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029468405215646944,
      "loss": 2.9086,
      "theoretical_loss": 3.5434279394538217,
      "tokens_seen": 1374814208
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029467402206619856,
      "loss": 2.8456,
      "theoretical_loss": 3.543412796971316,
      "tokens_seen": 1374879744
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002946639919759278,
      "loss": 2.8594,
      "theoretical_loss": 3.543397655412676,
      "tokens_seen": 1374945280
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.000294653961885657,
      "loss": 2.6489,
      "theoretical_loss": 3.5433825147778015,
      "tokens_seen": 1375010816
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029464393179538616,
      "loss": 2.8487,
      "theoretical_loss": 3.543367375066593,
      "tokens_seen": 1375076352
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029463390170511534,
      "loss": 2.7959,
      "theoretical_loss": 3.5433522362789485,
      "tokens_seen": 1375141888
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002946238716148446,
      "loss": 2.9158,
      "theoretical_loss": 3.543337098414769,
      "tokens_seen": 1375207424
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002946138415245737,
      "loss": 2.6242,
      "theoretical_loss": 3.5433219614739535,
      "tokens_seen": 1375272960
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029460381143430294,
      "loss": 2.8378,
      "theoretical_loss": 3.543306825456402,
      "tokens_seen": 1375338496
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029459378134403207,
      "loss": 2.9769,
      "theoretical_loss": 3.5432916903620146,
      "tokens_seen": 1375404032
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002945837512537613,
      "loss": 2.6339,
      "theoretical_loss": 3.54327655619069,
      "tokens_seen": 1375469568
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002945737211634905,
      "loss": 2.9141,
      "theoretical_loss": 3.5432614229423285,
      "tokens_seen": 1375535104
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029456369107321966,
      "loss": 2.8059,
      "theoretical_loss": 3.54324629061683,
      "tokens_seen": 1375600640
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029455366098294885,
      "loss": 2.8913,
      "theoretical_loss": 3.543231159214094,
      "tokens_seen": 1375666176
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.000294543630892678,
      "loss": 2.8595,
      "theoretical_loss": 3.5432160287340206,
      "tokens_seen": 1375731712
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002945336008024072,
      "loss": 2.8078,
      "theoretical_loss": 3.543200899176509,
      "tokens_seen": 1375797248
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029452357071213644,
      "loss": 2.8657,
      "theoretical_loss": 3.5431857705414593,
      "tokens_seen": 1375862784
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029451354062186557,
      "loss": 2.932,
      "theoretical_loss": 3.5431706428287715,
      "tokens_seen": 1375928320
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002945035105315948,
      "loss": 2.9123,
      "theoretical_loss": 3.543155516038345,
      "tokens_seen": 1375993856
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029449348044132393,
      "loss": 2.8094,
      "theoretical_loss": 3.5431403901700804,
      "tokens_seen": 1376059392
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029448345035105317,
      "loss": 2.9152,
      "theoretical_loss": 3.5431252652238765,
      "tokens_seen": 1376124928
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029447342026078235,
      "loss": 2.7974,
      "theoretical_loss": 3.543110141199634,
      "tokens_seen": 1376190464
    },
    {
      "epoch": 0.42,
      "objective/train/advantage_avg": 0.493445485830307,
      "objective/train/docs_used": 779255,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.0422770977020264,
      "objective/train/original_loss": 3.0422773361206055,
      "objective/train/theoretical_loss": 3.5430950180972527,
      "objective/train/tokens_used": 1396716000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2436220943927765,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505828857421875,
      "objective/train/weighted_lm_loss": 3.1961898803710938,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0363596677780151,
      "theoretical_loss": 3.5430950180972527,
      "tokens_seen": 1376256000
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029446339017051153,
      "loss": 2.724,
      "theoretical_loss": 3.5430950180972527,
      "tokens_seen": 1376256000
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002944533600802407,
      "loss": 2.6149,
      "theoretical_loss": 3.543079895916633,
      "tokens_seen": 1376321536
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029444332998996995,
      "loss": 2.809,
      "theoretical_loss": 3.5430647746576733,
      "tokens_seen": 1376387072
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002944332998996991,
      "loss": 2.701,
      "theoretical_loss": 3.543049654320275,
      "tokens_seen": 1376452608
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002944232698094283,
      "loss": 2.9203,
      "theoretical_loss": 3.5430345349043373,
      "tokens_seen": 1376518144
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029441323971915744,
      "loss": 2.8934,
      "theoretical_loss": 3.54301941640976,
      "tokens_seen": 1376583680
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029440320962888667,
      "loss": 2.6543,
      "theoretical_loss": 3.5430042988364443,
      "tokens_seen": 1376649216
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029439317953861585,
      "loss": 2.6192,
      "theoretical_loss": 3.5429891821842894,
      "tokens_seen": 1376714752
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029438314944834503,
      "loss": 2.8753,
      "theoretical_loss": 3.542974066453195,
      "tokens_seen": 1376780288
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002943731193580742,
      "loss": 2.7322,
      "theoretical_loss": 3.5429589516430617,
      "tokens_seen": 1376845824
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002943630892678034,
      "loss": 2.7194,
      "theoretical_loss": 3.542943837753789,
      "tokens_seen": 1376911360
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002943530591775326,
      "loss": 2.6228,
      "theoretical_loss": 3.542928724785278,
      "tokens_seen": 1376976896
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002943430290872618,
      "loss": 3.0807,
      "theoretical_loss": 3.5429136127374274,
      "tokens_seen": 1377042432
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029433299899699094,
      "loss": 2.7227,
      "theoretical_loss": 3.5428985016101384,
      "tokens_seen": 1377107968
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002943229689067202,
      "loss": 2.7735,
      "theoretical_loss": 3.5428833914033104,
      "tokens_seen": 1377173504
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002943129388164493,
      "loss": 2.7874,
      "theoretical_loss": 3.542868282116844,
      "tokens_seen": 1377239040
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029430290872617854,
      "loss": 2.8794,
      "theoretical_loss": 3.5428531737506397,
      "tokens_seen": 1377304576
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029429287863590777,
      "loss": 2.7314,
      "theoretical_loss": 3.542838066304596,
      "tokens_seen": 1377370112
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002942828485456369,
      "loss": 2.8231,
      "theoretical_loss": 3.5428229597786154,
      "tokens_seen": 1377435648
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029427281845536613,
      "loss": 2.9574,
      "theoretical_loss": 3.5428078541725965,
      "tokens_seen": 1377501184
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002942627883650953,
      "loss": 3.0612,
      "theoretical_loss": 3.54279274948644,
      "tokens_seen": 1377566720
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002942527582748245,
      "loss": 2.8111,
      "theoretical_loss": 3.5427776457200464,
      "tokens_seen": 1377632256
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002942427281845537,
      "loss": 2.8218,
      "theoretical_loss": 3.542762542873315,
      "tokens_seen": 1377697792
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029423269809428286,
      "loss": 2.5955,
      "theoretical_loss": 3.5427474409461475,
      "tokens_seen": 1377763328
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029422266800401204,
      "loss": 2.901,
      "theoretical_loss": 3.5427323399384427,
      "tokens_seen": 1377828864
    },
    {
      "epoch": 0.42,
      "objective/train/advantage_avg": 0.4866192936897278,
      "objective/train/docs_used": 780014,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.328575611114502,
      "objective/train/original_loss": 2.328575372695923,
      "objective/train/theoretical_loss": 3.5427172398501012,
      "objective/train/tokens_used": 1398354400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.23748168349266052,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0498689413070679,
      "objective/train/weighted_lm_loss": 2.4443747997283936,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0369682312011719,
      "theoretical_loss": 3.5427172398501012,
      "tokens_seen": 1377894400
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002942126379137413,
      "loss": 2.685,
      "theoretical_loss": 3.5427172398501012,
      "tokens_seen": 1377894400
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002942026078234704,
      "loss": 3.0596,
      "theoretical_loss": 3.5427021406810244,
      "tokens_seen": 1377959936
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029419257773319964,
      "loss": 2.6728,
      "theoretical_loss": 3.542687042431112,
      "tokens_seen": 1378025472
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029418254764292876,
      "loss": 3.1268,
      "theoretical_loss": 3.5426719451002637,
      "tokens_seen": 1378091008
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.000294172517552658,
      "loss": 2.743,
      "theoretical_loss": 3.5426568486883805,
      "tokens_seen": 1378156544
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002941624874623872,
      "loss": 2.8777,
      "theoretical_loss": 3.5426417531953627,
      "tokens_seen": 1378222080
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029415245737211636,
      "loss": 2.7725,
      "theoretical_loss": 3.542626658621111,
      "tokens_seen": 1378287616
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029414242728184554,
      "loss": 2.9152,
      "theoretical_loss": 3.542611564965525,
      "tokens_seen": 1378353152
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002941323971915748,
      "loss": 2.9339,
      "theoretical_loss": 3.5425964722285053,
      "tokens_seen": 1378418688
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002941223671013039,
      "loss": 2.6803,
      "theoretical_loss": 3.5425813804099526,
      "tokens_seen": 1378484224
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029411233701103314,
      "loss": 2.882,
      "theoretical_loss": 3.542566289509768,
      "tokens_seen": 1378549760
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029410230692076227,
      "loss": 2.9726,
      "theoretical_loss": 3.5425511995278507,
      "tokens_seen": 1378615296
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002940922768304915,
      "loss": 2.6168,
      "theoretical_loss": 3.542536110464102,
      "tokens_seen": 1378680832
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002940822467402207,
      "loss": 2.7552,
      "theoretical_loss": 3.5425210223184216,
      "tokens_seen": 1378746368
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029407221664994986,
      "loss": 2.4115,
      "theoretical_loss": 3.542505935090711,
      "tokens_seen": 1378811904
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029406218655967905,
      "loss": 2.9444,
      "theoretical_loss": 3.5424908487808704,
      "tokens_seen": 1378877440
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002940521564694082,
      "loss": 2.8657,
      "theoretical_loss": 3.5424757633887998,
      "tokens_seen": 1378942976
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002940421263791374,
      "loss": 2.9471,
      "theoretical_loss": 3.5424606789144004,
      "tokens_seen": 1379008512
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029403209628886664,
      "loss": 3.1621,
      "theoretical_loss": 3.5424455953575724,
      "tokens_seen": 1379074048
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029402206619859577,
      "loss": 2.9332,
      "theoretical_loss": 3.5424305127182167,
      "tokens_seen": 1379139584
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.000294012036108325,
      "loss": 2.8906,
      "theoretical_loss": 3.5424154309962335,
      "tokens_seen": 1379205120
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029400200601805413,
      "loss": 2.7819,
      "theoretical_loss": 3.5424003501915236,
      "tokens_seen": 1379270656
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029399197592778337,
      "loss": 2.6775,
      "theoretical_loss": 3.542385270303988,
      "tokens_seen": 1379336192
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029398194583751255,
      "loss": 2.8761,
      "theoretical_loss": 3.542370191333527,
      "tokens_seen": 1379401728
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029397191574724173,
      "loss": 2.8773,
      "theoretical_loss": 3.5423551132800406,
      "tokens_seen": 1379467264
    },
    {
      "epoch": 0.42,
      "objective/train/advantage_avg": 0.4954702854156494,
      "objective/train/docs_used": 781246,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 1.8974475860595703,
      "objective/train/original_loss": 1.8974477052688599,
      "objective/train/theoretical_loss": 3.5423400361434307,
      "objective/train/tokens_used": 1399992800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24553343653678894,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050795078277588,
      "objective/train/weighted_lm_loss": 1.9936498403549194,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0437887907028198,
      "theoretical_loss": 3.5423400361434307,
      "tokens_seen": 1379532800
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002939618856569709,
      "loss": 2.4811,
      "theoretical_loss": 3.5423400361434307,
      "tokens_seen": 1379532800
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029395185556670015,
      "loss": 2.6882,
      "theoretical_loss": 3.5423249599235973,
      "tokens_seen": 1379598336
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002939418254764293,
      "loss": 2.9801,
      "theoretical_loss": 3.5423098846204413,
      "tokens_seen": 1379663872
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002939317953861585,
      "loss": 2.6991,
      "theoretical_loss": 3.542294810233863,
      "tokens_seen": 1379729408
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029392176529588764,
      "loss": 2.8462,
      "theoretical_loss": 3.5422797367637635,
      "tokens_seen": 1379794944
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029391173520561687,
      "loss": 2.9431,
      "theoretical_loss": 3.542264664210044,
      "tokens_seen": 1379860480
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029390170511534605,
      "loss": 2.6769,
      "theoretical_loss": 3.542249592572605,
      "tokens_seen": 1379926016
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029389167502507523,
      "loss": 2.8886,
      "theoretical_loss": 3.5422345218513467,
      "tokens_seen": 1379991552
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002938816449348044,
      "loss": 2.8281,
      "theoretical_loss": 3.5422194520461705,
      "tokens_seen": 1380057088
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002938716148445336,
      "loss": 2.7882,
      "theoretical_loss": 3.5422043831569767,
      "tokens_seen": 1380122624
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002938615847542628,
      "loss": 2.8111,
      "theoretical_loss": 3.542189315183667,
      "tokens_seen": 1380188160
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.000293851554663992,
      "loss": 2.8981,
      "theoretical_loss": 3.542174248126141,
      "tokens_seen": 1380253696
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029384152457372114,
      "loss": 2.4815,
      "theoretical_loss": 3.5421591819843012,
      "tokens_seen": 1380319232
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002938314944834504,
      "loss": 2.7138,
      "theoretical_loss": 3.542144116758047,
      "tokens_seen": 1380384768
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002938214643931795,
      "loss": 2.7097,
      "theoretical_loss": 3.5421290524472795,
      "tokens_seen": 1380450304
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029381143430290874,
      "loss": 2.8602,
      "theoretical_loss": 3.5421139890519004,
      "tokens_seen": 1380515840
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002938014042126379,
      "loss": 2.5304,
      "theoretical_loss": 3.5420989265718106,
      "tokens_seen": 1380581376
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002937913741223671,
      "loss": 2.992,
      "theoretical_loss": 3.54208386500691,
      "tokens_seen": 1380646912
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002937813440320963,
      "loss": 2.7765,
      "theoretical_loss": 3.5420688043571005,
      "tokens_seen": 1380712448
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002937713139418255,
      "loss": 2.779,
      "theoretical_loss": 3.542053744622282,
      "tokens_seen": 1380777984
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029376128385155464,
      "loss": 2.6868,
      "theoretical_loss": 3.5420386858023574,
      "tokens_seen": 1380843520
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002937512537612839,
      "loss": 2.8787,
      "theoretical_loss": 3.542023627897226,
      "tokens_seen": 1380909056
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.000293741223671013,
      "loss": 2.8535,
      "theoretical_loss": 3.5420085709067894,
      "tokens_seen": 1380974592
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029373119358074224,
      "loss": 2.8017,
      "theoretical_loss": 3.5419935148309483,
      "tokens_seen": 1381040128
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002937211634904714,
      "loss": 2.6842,
      "theoretical_loss": 3.5419784596696045,
      "tokens_seen": 1381105664
    },
    {
      "epoch": 0.42,
      "objective/train/advantage_avg": 0.4936811625957489,
      "objective/train/docs_used": 781914,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.4484012126922607,
      "objective/train/original_loss": 2.44840145111084,
      "objective/train/theoretical_loss": 3.541963405422658,
      "objective/train/tokens_used": 1401631200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24383720755577087,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506075620651245,
      "objective/train/weighted_lm_loss": 2.5726068019866943,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0408108234405518,
      "theoretical_loss": 3.541963405422658,
      "tokens_seen": 1381171200
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002937111334002006,
      "loss": 2.7299,
      "theoretical_loss": 3.541963405422658,
      "tokens_seen": 1381171200
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002937011033099298,
      "loss": 2.8135,
      "theoretical_loss": 3.541948352090011,
      "tokens_seen": 1381236736
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029369107321965896,
      "loss": 2.8614,
      "theoretical_loss": 3.5419332996715642,
      "tokens_seen": 1381302272
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029368104312938815,
      "loss": 2.8292,
      "theoretical_loss": 3.5419182481672182,
      "tokens_seen": 1381367808
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002936710130391174,
      "loss": 2.7851,
      "theoretical_loss": 3.5419031975768744,
      "tokens_seen": 1381433344
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002936609829488465,
      "loss": 2.9255,
      "theoretical_loss": 3.5418881479004343,
      "tokens_seen": 1381498880
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029365095285857574,
      "loss": 2.5888,
      "theoretical_loss": 3.5418730991377987,
      "tokens_seen": 1381564416
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029364092276830487,
      "loss": 2.8217,
      "theoretical_loss": 3.5418580512888687,
      "tokens_seen": 1381629952
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002936308926780341,
      "loss": 2.9308,
      "theoretical_loss": 3.541843004353546,
      "tokens_seen": 1381695488
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002936208625877633,
      "loss": 2.8081,
      "theoretical_loss": 3.5418279583317314,
      "tokens_seen": 1381761024
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029361083249749247,
      "loss": 2.9608,
      "theoretical_loss": 3.541812913223326,
      "tokens_seen": 1381826560
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029360080240722165,
      "loss": 2.7944,
      "theoretical_loss": 3.5417978690282315,
      "tokens_seen": 1381892096
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002935907723169509,
      "loss": 2.5773,
      "theoretical_loss": 3.5417828257463486,
      "tokens_seen": 1381957632
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029358074222668,
      "loss": 2.6673,
      "theoretical_loss": 3.541767783377579,
      "tokens_seen": 1382023168
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029357071213640925,
      "loss": 3.014,
      "theoretical_loss": 3.5417527419218233,
      "tokens_seen": 1382088704
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002935606820461384,
      "loss": 2.9756,
      "theoretical_loss": 3.5417377013789837,
      "tokens_seen": 1382154240
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002935506519558676,
      "loss": 2.6598,
      "theoretical_loss": 3.5417226617489614,
      "tokens_seen": 1382219776
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029354062186559684,
      "loss": 2.6333,
      "theoretical_loss": 3.541707623031657,
      "tokens_seen": 1382285312
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029353059177532597,
      "loss": 2.6318,
      "theoretical_loss": 3.5416925852269725,
      "tokens_seen": 1382350848
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002935205616850552,
      "loss": 2.8523,
      "theoretical_loss": 3.5416775483348086,
      "tokens_seen": 1382416384
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029351053159478433,
      "loss": 2.8657,
      "theoretical_loss": 3.541662512355068,
      "tokens_seen": 1382481920
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029350050150451357,
      "loss": 2.723,
      "theoretical_loss": 3.5416474772876505,
      "tokens_seen": 1382547456
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029349047141424275,
      "loss": 2.9225,
      "theoretical_loss": 3.541632443132458,
      "tokens_seen": 1382612992
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029348044132397193,
      "loss": 2.9034,
      "theoretical_loss": 3.541617409889392,
      "tokens_seen": 1382678528
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002934704112337011,
      "loss": 2.6366,
      "theoretical_loss": 3.541602377558355,
      "tokens_seen": 1382744064
    },
    {
      "epoch": 0.42,
      "objective/train/advantage_avg": 0.4937203824520111,
      "objective/train/docs_used": 783265,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.6116487979888916,
      "objective/train/original_loss": 2.61164927482605,
      "objective/train/theoretical_loss": 3.541587346139247,
      "objective/train/tokens_used": 1403269600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24384501576423645,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506116151809692,
      "objective/train/weighted_lm_loss": 2.743726968765259,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0338950157165527,
      "theoretical_loss": 3.541587346139247,
      "tokens_seen": 1382809600
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029346038114343035,
      "loss": 3.0164,
      "theoretical_loss": 3.541587346139247,
      "tokens_seen": 1382809600
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002934503510531595,
      "loss": 3.1055,
      "theoretical_loss": 3.54157231563197,
      "tokens_seen": 1382875136
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002934403209628887,
      "loss": 2.8219,
      "theoretical_loss": 3.541557286036425,
      "tokens_seen": 1382940672
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029343029087261784,
      "loss": 2.5917,
      "theoretical_loss": 3.541542257352514,
      "tokens_seen": 1383006208
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029342026078234707,
      "loss": 2.8966,
      "theoretical_loss": 3.5415272295801388,
      "tokens_seen": 1383071744
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029341023069207625,
      "loss": 2.8576,
      "theoretical_loss": 3.5415122027192005,
      "tokens_seen": 1383137280
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029340020060180543,
      "loss": 2.8644,
      "theoretical_loss": 3.5414971767696004,
      "tokens_seen": 1383202816
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002933901705115346,
      "loss": 2.9373,
      "theoretical_loss": 3.5414821517312403,
      "tokens_seen": 1383268352
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002933801404212638,
      "loss": 2.8631,
      "theoretical_loss": 3.5414671276040224,
      "tokens_seen": 1383333888
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.000293370110330993,
      "loss": 2.8768,
      "theoretical_loss": 3.5414521043878473,
      "tokens_seen": 1383399424
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002933600802407222,
      "loss": 2.8726,
      "theoretical_loss": 3.5414370820826164,
      "tokens_seen": 1383464960
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029335005015045134,
      "loss": 2.7353,
      "theoretical_loss": 3.541422060688233,
      "tokens_seen": 1383530496
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002933400200601806,
      "loss": 2.8605,
      "theoretical_loss": 3.541407040204597,
      "tokens_seen": 1383596032
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002933299899699097,
      "loss": 2.7404,
      "theoretical_loss": 3.5413920206316107,
      "tokens_seen": 1383661568
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029331995987963894,
      "loss": 2.7712,
      "theoretical_loss": 3.5413770019691757,
      "tokens_seen": 1383727104
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002933099297893681,
      "loss": 2.8906,
      "theoretical_loss": 3.541361984217194,
      "tokens_seen": 1383792640
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002932998996990973,
      "loss": 2.7364,
      "theoretical_loss": 3.541346967375567,
      "tokens_seen": 1383858176
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002932898696088265,
      "loss": 2.7692,
      "theoretical_loss": 3.5413319514441963,
      "tokens_seen": 1383923712
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002932798395185557,
      "loss": 2.8514,
      "theoretical_loss": 3.5413169364229837,
      "tokens_seen": 1383989248
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029326980942828484,
      "loss": 2.8555,
      "theoretical_loss": 3.541301922311831,
      "tokens_seen": 1384054784
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002932597793380141,
      "loss": 2.7163,
      "theoretical_loss": 3.5412869091106405,
      "tokens_seen": 1384120320
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002932497492477432,
      "loss": 2.6944,
      "theoretical_loss": 3.541271896819313,
      "tokens_seen": 1384185856
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029323971915747244,
      "loss": 2.8779,
      "theoretical_loss": 3.5412568854377504,
      "tokens_seen": 1384251392
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002932296890672016,
      "loss": 2.8042,
      "theoretical_loss": 3.5412418749658547,
      "tokens_seen": 1384316928
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002932196589769308,
      "loss": 2.7731,
      "theoretical_loss": 3.5412268654035284,
      "tokens_seen": 1384382464
    },
    {
      "epoch": 0.42,
      "objective/train/advantage_avg": 0.49306532740592957,
      "objective/train/docs_used": 783958,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.9920337200164795,
      "objective/train/original_loss": 2.9920334815979004,
      "objective/train/theoretical_loss": 3.5412118567506727,
      "objective/train/tokens_used": 1404908000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24320973455905914,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505428314208984,
      "objective/train/weighted_lm_loss": 3.143172264099121,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0338950157165527,
      "theoretical_loss": 3.5412118567506727,
      "tokens_seen": 1384448000
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029320962888666,
      "loss": 3.0081,
      "theoretical_loss": 3.5412118567506727,
      "tokens_seen": 1384448000
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029319959879638916,
      "loss": 2.8556,
      "theoretical_loss": 3.5411968490071892,
      "tokens_seen": 1384513536
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029318956870611835,
      "loss": 2.6867,
      "theoretical_loss": 3.54118184217298,
      "tokens_seen": 1384579072
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002931795386158476,
      "loss": 2.6619,
      "theoretical_loss": 3.541166836247947,
      "tokens_seen": 1384644608
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002931695085255767,
      "loss": 2.6942,
      "theoretical_loss": 3.5411518312319923,
      "tokens_seen": 1384710144
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029315947843530594,
      "loss": 2.7714,
      "theoretical_loss": 3.541136827125017,
      "tokens_seen": 1384775680
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029314944834503507,
      "loss": 2.7175,
      "theoretical_loss": 3.541121823926925,
      "tokens_seen": 1384841216
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002931394182547643,
      "loss": 2.8044,
      "theoretical_loss": 3.541106821637616,
      "tokens_seen": 1384906752
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002931293881644935,
      "loss": 2.8077,
      "theoretical_loss": 3.5410918202569928,
      "tokens_seen": 1384972288
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029311935807422267,
      "loss": 2.7052,
      "theoretical_loss": 3.541076819784957,
      "tokens_seen": 1385037824
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029310932798395185,
      "loss": 2.8462,
      "theoretical_loss": 3.5410618202214117,
      "tokens_seen": 1385103360
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002930992978936811,
      "loss": 2.8457,
      "theoretical_loss": 3.541046821566258,
      "tokens_seen": 1385168896
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002930892678034102,
      "loss": 2.9727,
      "theoretical_loss": 3.541031823819398,
      "tokens_seen": 1385234432
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029307923771313945,
      "loss": 2.8288,
      "theoretical_loss": 3.5410168269807336,
      "tokens_seen": 1385299968
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002930692076228686,
      "loss": 2.7229,
      "theoretical_loss": 3.541001831050167,
      "tokens_seen": 1385365504
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002930591775325978,
      "loss": 3.0031,
      "theoretical_loss": 3.5409868360276007,
      "tokens_seen": 1385431040
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.000293049147442327,
      "loss": 2.9906,
      "theoretical_loss": 3.540971841912936,
      "tokens_seen": 1385496576
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029303911735205617,
      "loss": 2.7238,
      "theoretical_loss": 3.5409568487060756,
      "tokens_seen": 1385562112
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029302908726178535,
      "loss": 2.5653,
      "theoretical_loss": 3.5409418564069215,
      "tokens_seen": 1385627648
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029301905717151453,
      "loss": 2.965,
      "theoretical_loss": 3.5409268650153756,
      "tokens_seen": 1385693184
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002930090270812437,
      "loss": 2.682,
      "theoretical_loss": 3.54091187453134,
      "tokens_seen": 1385758720
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029299899699097295,
      "loss": 2.8436,
      "theoretical_loss": 3.540896884954717,
      "tokens_seen": 1385824256
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002929889669007021,
      "loss": 2.9542,
      "theoretical_loss": 3.5408818962854087,
      "tokens_seen": 1385889792
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002929789368104313,
      "loss": 2.7603,
      "theoretical_loss": 3.5408669085233173,
      "tokens_seen": 1385955328
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002929689067201605,
      "loss": 2.9444,
      "theoretical_loss": 3.540851921668345,
      "tokens_seen": 1386020864
    },
    {
      "epoch": 0.42,
      "objective/train/advantage_avg": 0.49529561400413513,
      "objective/train/docs_used": 785103,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.584195375442505,
      "objective/train/original_loss": 2.584195137023926,
      "objective/train/theoretical_loss": 3.540836935720394,
      "objective/train/tokens_used": 1406546400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24536167085170746,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507769584655762,
      "objective/train/weighted_lm_loss": 2.7153372764587402,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0396549701690674,
      "theoretical_loss": 3.540836935720394,
      "tokens_seen": 1386086400
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002929588766298897,
      "loss": 2.6987,
      "theoretical_loss": 3.540836935720394,
      "tokens_seen": 1386086400
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029294884653961885,
      "loss": 2.7297,
      "theoretical_loss": 3.5408219506793666,
      "tokens_seen": 1386151936
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029293881644934804,
      "loss": 2.7638,
      "theoretical_loss": 3.5408069665451656,
      "tokens_seen": 1386217472
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002929287863590772,
      "loss": 2.7147,
      "theoretical_loss": 3.5407919833176917,
      "tokens_seen": 1386283008
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029291875626880645,
      "loss": 2.9287,
      "theoretical_loss": 3.5407770009968487,
      "tokens_seen": 1386348544
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002929087261785356,
      "loss": 2.6965,
      "theoretical_loss": 3.5407620195825382,
      "tokens_seen": 1386414080
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002928986960882648,
      "loss": 2.8145,
      "theoretical_loss": 3.5407470390746623,
      "tokens_seen": 1386479616
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029288866599799394,
      "loss": 2.8685,
      "theoretical_loss": 3.5407320594731235,
      "tokens_seen": 1386545152
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002928786359077232,
      "loss": 2.9217,
      "theoretical_loss": 3.5407170807778243,
      "tokens_seen": 1386610688
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029286860581745236,
      "loss": 2.925,
      "theoretical_loss": 3.5407021029886674,
      "tokens_seen": 1386676224
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029285857572718154,
      "loss": 2.5656,
      "theoretical_loss": 3.5406871261055546,
      "tokens_seen": 1386741760
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002928485456369107,
      "loss": 2.6352,
      "theoretical_loss": 3.5406721501283887,
      "tokens_seen": 1386807296
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002928385155466399,
      "loss": 2.876,
      "theoretical_loss": 3.5406571750570714,
      "tokens_seen": 1386872832
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002928284854563691,
      "loss": 2.8911,
      "theoretical_loss": 3.5406422008915053,
      "tokens_seen": 1386938368
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002928184553660983,
      "loss": 2.785,
      "theoretical_loss": 3.5406272276315933,
      "tokens_seen": 1387003904
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029280842527582744,
      "loss": 2.8895,
      "theoretical_loss": 3.540612255277238,
      "tokens_seen": 1387069440
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002927983951855567,
      "loss": 2.8875,
      "theoretical_loss": 3.5405972838283413,
      "tokens_seen": 1387134976
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002927883650952859,
      "loss": 2.8725,
      "theoretical_loss": 3.540582313284805,
      "tokens_seen": 1387200512
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029277833500501504,
      "loss": 2.863,
      "theoretical_loss": 3.5405673436465332,
      "tokens_seen": 1387266048
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002927683049147443,
      "loss": 2.7625,
      "theoretical_loss": 3.5405523749134273,
      "tokens_seen": 1387331584
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002927582748244734,
      "loss": 2.8612,
      "theoretical_loss": 3.54053740708539,
      "tokens_seen": 1387397120
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029274824473420264,
      "loss": 2.9352,
      "theoretical_loss": 3.540522440162324,
      "tokens_seen": 1387462656
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002927382146439318,
      "loss": 2.6557,
      "theoretical_loss": 3.540507474144132,
      "tokens_seen": 1387528192
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.000292728184553661,
      "loss": 2.9673,
      "theoretical_loss": 3.5404925090307158,
      "tokens_seen": 1387593728
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002927181544633902,
      "loss": 2.8269,
      "theoretical_loss": 3.5404775448219787,
      "tokens_seen": 1387659264
    },
    {
      "epoch": 0.42,
      "objective/train/advantage_avg": 0.49503663182258606,
      "objective/train/docs_used": 785794,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.953033208847046,
      "objective/train/original_loss": 2.953033447265625,
      "objective/train/theoretical_loss": 3.540462581517824,
      "objective/train/tokens_used": 1408184800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24513673782348633,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507497787475586,
      "objective/train/weighted_lm_loss": 3.1030325889587402,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0363596677780151,
      "theoretical_loss": 3.540462581517824,
      "tokens_seen": 1387724800
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029270812437311936,
      "loss": 2.8167,
      "theoretical_loss": 3.540462581517824,
      "tokens_seen": 1387724800
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029269809428284855,
      "loss": 2.8741,
      "theoretical_loss": 3.540447619118152,
      "tokens_seen": 1387790336
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002926880641925778,
      "loss": 2.6563,
      "theoretical_loss": 3.5404326576228673,
      "tokens_seen": 1387855872
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002926780341023069,
      "loss": 2.7695,
      "theoretical_loss": 3.540417697031872,
      "tokens_seen": 1387921408
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029266800401203614,
      "loss": 2.7483,
      "theoretical_loss": 3.540402737345069,
      "tokens_seen": 1387986944
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029265797392176527,
      "loss": 3.0715,
      "theoretical_loss": 3.54038777856236,
      "tokens_seen": 1388052480
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002926479438314945,
      "loss": 2.9328,
      "theoretical_loss": 3.5403728206836487,
      "tokens_seen": 1388118016
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002926379137412237,
      "loss": 2.7936,
      "theoretical_loss": 3.540357863708837,
      "tokens_seen": 1388183552
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029262788365095287,
      "loss": 2.975,
      "theoretical_loss": 3.5403429076378283,
      "tokens_seen": 1388249088
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029261785356068205,
      "loss": 2.7345,
      "theoretical_loss": 3.540327952470525,
      "tokens_seen": 1388314624
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002926078234704113,
      "loss": 2.8737,
      "theoretical_loss": 3.54031299820683,
      "tokens_seen": 1388380160
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002925977933801404,
      "loss": 2.9241,
      "theoretical_loss": 3.540298044846646,
      "tokens_seen": 1388445696
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029258776328986965,
      "loss": 2.6304,
      "theoretical_loss": 3.540283092389876,
      "tokens_seen": 1388511232
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002925777331995988,
      "loss": 2.8889,
      "theoretical_loss": 3.540268140836422,
      "tokens_seen": 1388576768
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.000292567703109328,
      "loss": 2.7903,
      "theoretical_loss": 3.540253190186187,
      "tokens_seen": 1388642304
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002925576730190572,
      "loss": 2.6957,
      "theoretical_loss": 3.5402382404390744,
      "tokens_seen": 1388707840
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029254764292878637,
      "loss": 2.7702,
      "theoretical_loss": 3.540223291594987,
      "tokens_seen": 1388773376
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029253761283851555,
      "loss": 2.7832,
      "theoretical_loss": 3.540208343653827,
      "tokens_seen": 1388838912
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029252758274824473,
      "loss": 2.9142,
      "theoretical_loss": 3.540193396615498,
      "tokens_seen": 1388904448
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002925175526579739,
      "loss": 2.8014,
      "theoretical_loss": 3.5401784504799023,
      "tokens_seen": 1388969984
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029250752256770315,
      "loss": 2.8326,
      "theoretical_loss": 3.540163505246943,
      "tokens_seen": 1389035520
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002924974924774323,
      "loss": 2.9884,
      "theoretical_loss": 3.5401485609165233,
      "tokens_seen": 1389101056
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002924874623871615,
      "loss": 2.6222,
      "theoretical_loss": 3.5401336174885456,
      "tokens_seen": 1389166592
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002924774322968907,
      "loss": 2.8628,
      "theoretical_loss": 3.5401186749629128,
      "tokens_seen": 1389232128
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002924674022066199,
      "loss": 2.9283,
      "theoretical_loss": 3.540103733339528,
      "tokens_seen": 1389297664
    },
    {
      "epoch": 0.42,
      "objective/train/advantage_avg": 0.4955155551433563,
      "objective/train/docs_used": 786491,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.450939655303955,
      "objective/train/original_loss": 2.450939178466797,
      "objective/train/theoretical_loss": 3.5400887926182953,
      "objective/train/tokens_used": 1409823200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24559111893177032,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507999658584595,
      "objective/train/weighted_lm_loss": 2.575413465499878,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0304546356201172,
      "theoretical_loss": 3.5400887926182953,
      "tokens_seen": 1389363200
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029245737211634906,
      "loss": 2.9204,
      "theoretical_loss": 3.5400887926182953,
      "tokens_seen": 1389363200
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029244734202607824,
      "loss": 2.8248,
      "theoretical_loss": 3.5400738527991154,
      "tokens_seen": 1389428736
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002924373119358074,
      "loss": 2.8118,
      "theoretical_loss": 3.5400589138818934,
      "tokens_seen": 1389494272
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029242728184553665,
      "loss": 2.92,
      "theoretical_loss": 3.540043975866531,
      "tokens_seen": 1389559808
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002924172517552658,
      "loss": 2.9401,
      "theoretical_loss": 3.540029038752932,
      "tokens_seen": 1389625344
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.000292407221664995,
      "loss": 2.8671,
      "theoretical_loss": 3.540014102540999,
      "tokens_seen": 1389690880
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029239719157472414,
      "loss": 2.7506,
      "theoretical_loss": 3.539999167230635,
      "tokens_seen": 1389756416
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002923871614844534,
      "loss": 2.8343,
      "theoretical_loss": 3.5399842328217437,
      "tokens_seen": 1389821952
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029237713139418256,
      "loss": 2.8555,
      "theoretical_loss": 3.539969299314227,
      "tokens_seen": 1389887488
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029236710130391174,
      "loss": 2.9096,
      "theoretical_loss": 3.539954366707989,
      "tokens_seen": 1389953024
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002923570712136409,
      "loss": 2.9678,
      "theoretical_loss": 3.5399394350029327,
      "tokens_seen": 1390018560
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002923470411233701,
      "loss": 2.5666,
      "theoretical_loss": 3.5399245041989613,
      "tokens_seen": 1390084096
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002923370110330993,
      "loss": 3.2362,
      "theoretical_loss": 3.5399095742959776,
      "tokens_seen": 1390149632
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002923269809428285,
      "loss": 2.956,
      "theoretical_loss": 3.5398946452938844,
      "tokens_seen": 1390215168
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029231695085255765,
      "loss": 2.9945,
      "theoretical_loss": 3.5398797171925858,
      "tokens_seen": 1390280704
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002923069207622869,
      "loss": 2.866,
      "theoretical_loss": 3.5398647899919844,
      "tokens_seen": 1390346240
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029229689067201606,
      "loss": 3.1338,
      "theoretical_loss": 3.5398498636919835,
      "tokens_seen": 1390411776
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029228686058174524,
      "loss": 2.7254,
      "theoretical_loss": 3.5398349382924863,
      "tokens_seen": 1390477312
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002922768304914744,
      "loss": 2.9953,
      "theoretical_loss": 3.5398200137933964,
      "tokens_seen": 1390542848
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002922668004012036,
      "loss": 3.0238,
      "theoretical_loss": 3.5398050901946165,
      "tokens_seen": 1390608384
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002922567703109328,
      "loss": 2.8779,
      "theoretical_loss": 3.53979016749605,
      "tokens_seen": 1390673920
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.000292246740220662,
      "loss": 3.0798,
      "theoretical_loss": 3.5397752456976006,
      "tokens_seen": 1390739456
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029223671013039115,
      "loss": 3.0855,
      "theoretical_loss": 3.539760324799171,
      "tokens_seen": 1390804992
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002922266800401204,
      "loss": 2.8662,
      "theoretical_loss": 3.539745404800665,
      "tokens_seen": 1390870528
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002922166499498495,
      "loss": 2.9021,
      "theoretical_loss": 3.539730485701985,
      "tokens_seen": 1390936064
    },
    {
      "epoch": 0.42,
      "objective/train/advantage_avg": 0.49611392617225647,
      "objective/train/docs_used": 787676,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.397355556488037,
      "objective/train/original_loss": 2.397355794906616,
      "objective/train/theoretical_loss": 3.539715567503036,
      "objective/train/tokens_used": 1411461600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24617546796798706,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0508629083633423,
      "objective/train/weighted_lm_loss": 2.5192408561706543,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 3.539715567503036,
      "tokens_seen": 1391001600
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029220661985957875,
      "loss": 2.7146,
      "theoretical_loss": 3.539715567503036,
      "tokens_seen": 1391001600
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002921965897693079,
      "loss": 2.8321,
      "theoretical_loss": 3.53970065020372,
      "tokens_seen": 1391067136
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002921865596790371,
      "loss": 2.8748,
      "theoretical_loss": 3.5396857338039407,
      "tokens_seen": 1391132672
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002921765295887663,
      "loss": 2.7563,
      "theoretical_loss": 3.5396708183036014,
      "tokens_seen": 1391198208
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029216649949849547,
      "loss": 2.8855,
      "theoretical_loss": 3.539655903702606,
      "tokens_seen": 1391263744
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029215646940822465,
      "loss": 2.9073,
      "theoretical_loss": 3.5396409900008576,
      "tokens_seen": 1391329280
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002921464393179539,
      "loss": 2.4927,
      "theoretical_loss": 3.539626077198259,
      "tokens_seen": 1391394816
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.000292136409227683,
      "loss": 2.5524,
      "theoretical_loss": 3.5396111652947146,
      "tokens_seen": 1391460352
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029212637913741225,
      "loss": 2.8087,
      "theoretical_loss": 3.539596254290128,
      "tokens_seen": 1391525888
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029211634904714143,
      "loss": 2.8712,
      "theoretical_loss": 3.5395813441844015,
      "tokens_seen": 1391591424
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002921063189568706,
      "loss": 2.924,
      "theoretical_loss": 3.5395664349774396,
      "tokens_seen": 1391656960
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002920962888665998,
      "loss": 3.0373,
      "theoretical_loss": 3.539551526669145,
      "tokens_seen": 1391722496
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.000292086258776329,
      "loss": 2.6894,
      "theoretical_loss": 3.5395366192594224,
      "tokens_seen": 1391788032
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029207622868605815,
      "loss": 2.6583,
      "theoretical_loss": 3.539521712748174,
      "tokens_seen": 1391853568
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002920661985957874,
      "loss": 2.657,
      "theoretical_loss": 3.5395068071353037,
      "tokens_seen": 1391919104
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002920561685055165,
      "loss": 2.8445,
      "theoretical_loss": 3.539491902420716,
      "tokens_seen": 1391984640
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029204613841524575,
      "loss": 3.1189,
      "theoretical_loss": 3.5394769986043135,
      "tokens_seen": 1392050176
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029203610832497493,
      "loss": 2.8527,
      "theoretical_loss": 3.5394620956860003,
      "tokens_seen": 1392115712
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002920260782347041,
      "loss": 2.8931,
      "theoretical_loss": 3.5394471936656795,
      "tokens_seen": 1392181248
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029201604814443335,
      "loss": 2.8473,
      "theoretical_loss": 3.539432292543255,
      "tokens_seen": 1392246784
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002920060180541625,
      "loss": 2.7646,
      "theoretical_loss": 3.5394173923186307,
      "tokens_seen": 1392312320
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002919959879638917,
      "loss": 2.5428,
      "theoretical_loss": 3.53940249299171,
      "tokens_seen": 1392377856
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002919859578736209,
      "loss": 2.6077,
      "theoretical_loss": 3.5393875945623963,
      "tokens_seen": 1392443392
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002919759277833501,
      "loss": 2.9557,
      "theoretical_loss": 3.539372697030594,
      "tokens_seen": 1392508928
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029196589769307926,
      "loss": 2.7327,
      "theoretical_loss": 3.539357800396206,
      "tokens_seen": 1392574464
    },
    {
      "epoch": 0.42,
      "objective/train/advantage_avg": 0.4949609935283661,
      "objective/train/docs_used": 788318,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.0876190662384033,
      "objective/train/original_loss": 2.0876190662384033,
      "objective/train/theoretical_loss": 3.5393429046591365,
      "objective/train/tokens_used": 1413100000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24504633247852325,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507416725158691,
      "objective/train/weighted_lm_loss": 2.193493127822876,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0304546356201172,
      "theoretical_loss": 3.5393429046591365,
      "tokens_seen": 1392640000
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029195586760280844,
      "loss": 2.7484,
      "theoretical_loss": 3.5393429046591365,
      "tokens_seen": 1392640000
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002919458375125376,
      "loss": 2.7471,
      "theoretical_loss": 3.5393280098192887,
      "tokens_seen": 1392705536
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029193580742226685,
      "loss": 2.6881,
      "theoretical_loss": 3.539313115876567,
      "tokens_seen": 1392771072
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.000291925777331996,
      "loss": 2.8342,
      "theoretical_loss": 3.539298222830875,
      "tokens_seen": 1392836608
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002919157472417252,
      "loss": 2.8901,
      "theoretical_loss": 3.5392833306821165,
      "tokens_seen": 1392902144
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029190571715145434,
      "loss": 2.8158,
      "theoretical_loss": 3.5392684394301943,
      "tokens_seen": 1392967680
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002918956870611836,
      "loss": 2.7715,
      "theoretical_loss": 3.539253549075014,
      "tokens_seen": 1393033216
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029188565697091276,
      "loss": 2.8478,
      "theoretical_loss": 3.539238659616478,
      "tokens_seen": 1393098752
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029187562688064194,
      "loss": 2.952,
      "theoretical_loss": 3.5392237710544907,
      "tokens_seen": 1393164288
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002918655967903711,
      "loss": 2.9743,
      "theoretical_loss": 3.5392088833889557,
      "tokens_seen": 1393229824
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002918555667001003,
      "loss": 3.2228,
      "theoretical_loss": 3.539193996619777,
      "tokens_seen": 1393295360
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002918455366098295,
      "loss": 2.9525,
      "theoretical_loss": 3.5391791107468586,
      "tokens_seen": 1393360896
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002918355065195587,
      "loss": 2.6346,
      "theoretical_loss": 3.539164225770104,
      "tokens_seen": 1393426432
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029182547642928785,
      "loss": 2.9103,
      "theoretical_loss": 3.5391493416894173,
      "tokens_seen": 1393491968
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002918154463390171,
      "loss": 2.9243,
      "theoretical_loss": 3.5391344585047024,
      "tokens_seen": 1393557504
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029180541624874626,
      "loss": 2.7935,
      "theoretical_loss": 3.539119576215864,
      "tokens_seen": 1393623040
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029179538615847544,
      "loss": 2.8024,
      "theoretical_loss": 3.5391046948228047,
      "tokens_seen": 1393688576
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002917853560682046,
      "loss": 2.7446,
      "theoretical_loss": 3.53908981432543,
      "tokens_seen": 1393754112
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002917753259779338,
      "loss": 3.019,
      "theoretical_loss": 3.539074934723642,
      "tokens_seen": 1393819648
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.000291765295887663,
      "loss": 2.7818,
      "theoretical_loss": 3.5390600560173455,
      "tokens_seen": 1393885184
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002917552657973922,
      "loss": 2.9264,
      "theoretical_loss": 3.539045178206446,
      "tokens_seen": 1393950720
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029174523570712135,
      "loss": 2.7413,
      "theoretical_loss": 3.539030301290845,
      "tokens_seen": 1394016256
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002917352056168506,
      "loss": 2.7229,
      "theoretical_loss": 3.539015425270448,
      "tokens_seen": 1394081792
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002917251755265797,
      "loss": 2.8381,
      "theoretical_loss": 3.5390005501451594,
      "tokens_seen": 1394147328
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029171514543630895,
      "loss": 2.7573,
      "theoretical_loss": 3.538985675914882,
      "tokens_seen": 1394212864
    },
    {
      "epoch": 0.42,
      "objective/train/advantage_avg": 0.4930134117603302,
      "objective/train/docs_used": 789787,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.5204784870147705,
      "objective/train/original_loss": 2.5204782485961914,
      "objective/train/theoretical_loss": 3.538970802579521,
      "objective/train/tokens_used": 1414738400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24314343929290771,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050537347793579,
      "objective/train/weighted_lm_loss": 2.6477088928222656,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0408108234405518,
      "theoretical_loss": 3.538970802579521,
      "tokens_seen": 1394278400
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002917051153460381,
      "loss": 2.6923,
      "theoretical_loss": 3.538970802579521,
      "tokens_seen": 1394278400
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002916950852557673,
      "loss": 2.6828,
      "theoretical_loss": 3.5389559301389797,
      "tokens_seen": 1394343936
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002916850551654965,
      "loss": 2.92,
      "theoretical_loss": 3.5389410585931627,
      "tokens_seen": 1394409472
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029167502507522567,
      "loss": 2.8595,
      "theoretical_loss": 3.538926187941974,
      "tokens_seen": 1394475008
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029166499498495485,
      "loss": 3.0607,
      "theoretical_loss": 3.538911318185318,
      "tokens_seen": 1394540544
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002916549648946841,
      "loss": 3.0509,
      "theoretical_loss": 3.538896449323098,
      "tokens_seen": 1394606080
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002916449348044132,
      "loss": 2.7645,
      "theoretical_loss": 3.538881581355219,
      "tokens_seen": 1394671616
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029163490471414245,
      "loss": 2.7525,
      "theoretical_loss": 3.5388667142815855,
      "tokens_seen": 1394737152
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029162487462387163,
      "loss": 2.8637,
      "theoretical_loss": 3.5388518481021003,
      "tokens_seen": 1394802688
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002916148445336008,
      "loss": 3.0167,
      "theoretical_loss": 3.538836982816669,
      "tokens_seen": 1394868224
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029160481444333,
      "loss": 2.7247,
      "theoretical_loss": 3.5388221184251947,
      "tokens_seen": 1394933760
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002915947843530592,
      "loss": 2.768,
      "theoretical_loss": 3.538807254927583,
      "tokens_seen": 1394999296
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029158475426278835,
      "loss": 2.7939,
      "theoretical_loss": 3.538792392323737,
      "tokens_seen": 1395064832
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002915747241725176,
      "loss": 2.669,
      "theoretical_loss": 3.538777530613561,
      "tokens_seen": 1395130368
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002915646940822467,
      "loss": 2.8572,
      "theoretical_loss": 3.5387626697969603,
      "tokens_seen": 1395195904
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029155466399197595,
      "loss": 3.0727,
      "theoretical_loss": 3.538747809873838,
      "tokens_seen": 1395261440
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002915446339017051,
      "loss": 3.0295,
      "theoretical_loss": 3.5387329508440994,
      "tokens_seen": 1395326976
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002915346038114343,
      "loss": 2.6833,
      "theoretical_loss": 3.538718092707648,
      "tokens_seen": 1395392512
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002915245737211635,
      "loss": 2.7966,
      "theoretical_loss": 3.5387032354643893,
      "tokens_seen": 1395458048
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002915145436308927,
      "loss": 3.0466,
      "theoretical_loss": 3.538688379114226,
      "tokens_seen": 1395523584
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029150451354062186,
      "loss": 3.0259,
      "theoretical_loss": 3.538673523657064,
      "tokens_seen": 1395589120
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002914944834503511,
      "loss": 2.7594,
      "theoretical_loss": 3.538658669092807,
      "tokens_seen": 1395654656
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002914844533600802,
      "loss": 3.1148,
      "theoretical_loss": 3.5386438154213593,
      "tokens_seen": 1395720192
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029147442326980946,
      "loss": 2.7661,
      "theoretical_loss": 3.5386289626426253,
      "tokens_seen": 1395785728
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002914643931795386,
      "loss": 2.9832,
      "theoretical_loss": 3.53861411075651,
      "tokens_seen": 1395851264
    },
    {
      "epoch": 0.42,
      "objective/train/advantage_avg": 0.4933207929134369,
      "objective/train/docs_used": 790377,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.0117387771606445,
      "objective/train/original_loss": 3.0117383003234863,
      "objective/train/theoretical_loss": 3.5385992597629174,
      "objective/train/tokens_used": 1416376800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24343624711036682,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505695343017578,
      "objective/train/weighted_lm_loss": 3.164088726043701,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0417574644088745,
      "theoretical_loss": 3.5385992597629174,
      "tokens_seen": 1395916800
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002914543630892678,
      "loss": 2.8338,
      "theoretical_loss": 3.5385992597629174,
      "tokens_seen": 1395916800
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.000291444332998997,
      "loss": 2.6547,
      "theoretical_loss": 3.5385844096617523,
      "tokens_seen": 1395982336
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002914343029087262,
      "loss": 2.8255,
      "theoretical_loss": 3.5385695604529186,
      "tokens_seen": 1396047872
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029142427281845536,
      "loss": 2.9574,
      "theoretical_loss": 3.538554712136321,
      "tokens_seen": 1396113408
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029141424272818454,
      "loss": 2.895,
      "theoretical_loss": 3.538539864711865,
      "tokens_seen": 1396178944
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002914042126379137,
      "loss": 2.9075,
      "theoretical_loss": 3.5385250181794534,
      "tokens_seen": 1396244480
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029139418254764296,
      "loss": 2.9496,
      "theoretical_loss": 3.5385101725389916,
      "tokens_seen": 1396310016
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002913841524573721,
      "loss": 2.7123,
      "theoretical_loss": 3.5384953277903843,
      "tokens_seen": 1396375552
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002913741223671013,
      "loss": 2.6613,
      "theoretical_loss": 3.5384804839335366,
      "tokens_seen": 1396441088
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029136409227683045,
      "loss": 2.7763,
      "theoretical_loss": 3.5384656409683517,
      "tokens_seen": 1396506624
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002913540621865597,
      "loss": 2.8029,
      "theoretical_loss": 3.5384507988947353,
      "tokens_seen": 1396572160
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029134403209628886,
      "loss": 2.7211,
      "theoretical_loss": 3.5384359577125912,
      "tokens_seen": 1396637696
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029133400200601805,
      "loss": 2.7766,
      "theoretical_loss": 3.5384211174218247,
      "tokens_seen": 1396703232
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002913239719157472,
      "loss": 2.8524,
      "theoretical_loss": 3.5384062780223404,
      "tokens_seen": 1396768768
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029131394182547646,
      "loss": 2.7221,
      "theoretical_loss": 3.538391439514042,
      "tokens_seen": 1396834304
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002913039117352056,
      "loss": 2.8602,
      "theoretical_loss": 3.538376601896836,
      "tokens_seen": 1396899840
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002912938816449348,
      "loss": 2.8975,
      "theoretical_loss": 3.538361765170625,
      "tokens_seen": 1396965376
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.000291283851554664,
      "loss": 2.7609,
      "theoretical_loss": 3.538346929335315,
      "tokens_seen": 1397030912
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002912738214643932,
      "loss": 2.8935,
      "theoretical_loss": 3.5383320943908108,
      "tokens_seen": 1397096448
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002912637913741224,
      "loss": 2.8619,
      "theoretical_loss": 3.5383172603370165,
      "tokens_seen": 1397161984
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029125376128385155,
      "loss": 3.0581,
      "theoretical_loss": 3.538302427173837,
      "tokens_seen": 1397227520
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002912437311935808,
      "loss": 2.6392,
      "theoretical_loss": 3.538287594901177,
      "tokens_seen": 1397293056
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002912337011033099,
      "loss": 3.022,
      "theoretical_loss": 3.538272763518942,
      "tokens_seen": 1397358592
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029122367101303915,
      "loss": 2.9554,
      "theoretical_loss": 3.538257933027036,
      "tokens_seen": 1397424128
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029121364092276833,
      "loss": 2.7182,
      "theoretical_loss": 3.538243103425364,
      "tokens_seen": 1397489664
    },
    {
      "epoch": 0.42,
      "objective/train/advantage_avg": 0.49490436911582947,
      "objective/train/docs_used": 791674,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.2765185832977295,
      "objective/train/original_loss": 2.2765181064605713,
      "objective/train/theoretical_loss": 3.5382282747138305,
      "objective/train/tokens_used": 1418015200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24499768018722534,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507359504699707,
      "objective/train/weighted_lm_loss": 2.391915798187256,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0396549701690674,
      "theoretical_loss": 3.5382282747138305,
      "tokens_seen": 1397555200
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002912036108324975,
      "loss": 2.7106,
      "theoretical_loss": 3.5382282747138305,
      "tokens_seen": 1397555200
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002911935807422267,
      "loss": 2.7945,
      "theoretical_loss": 3.5382134468923407,
      "tokens_seen": 1397620736
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029118355065195587,
      "loss": 2.7252,
      "theoretical_loss": 3.5381986199607995,
      "tokens_seen": 1397686272
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029117352056168505,
      "loss": 3.0532,
      "theoretical_loss": 3.538183793919112,
      "tokens_seen": 1397751808
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002911634904714143,
      "loss": 2.8263,
      "theoretical_loss": 3.5381689687671822,
      "tokens_seen": 1397817344
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002911534603811434,
      "loss": 2.8299,
      "theoretical_loss": 3.538154144504916,
      "tokens_seen": 1397882880
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029114343029087265,
      "loss": 2.7222,
      "theoretical_loss": 3.538139321132218,
      "tokens_seen": 1397948416
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029113340020060183,
      "loss": 2.6103,
      "theoretical_loss": 3.538124498648992,
      "tokens_seen": 1398013952
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.000291123370110331,
      "loss": 2.6139,
      "theoretical_loss": 3.538109677055145,
      "tokens_seen": 1398079488
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002911133400200602,
      "loss": 2.8344,
      "theoretical_loss": 3.5380948563505803,
      "tokens_seen": 1398145024
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002911033099297894,
      "loss": 2.9784,
      "theoretical_loss": 3.538080036535203,
      "tokens_seen": 1398210560
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029109327983951855,
      "loss": 2.7618,
      "theoretical_loss": 3.5380652176089193,
      "tokens_seen": 1398276096
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002910832497492478,
      "loss": 2.7505,
      "theoretical_loss": 3.5380503995716333,
      "tokens_seen": 1398341632
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002910732196589769,
      "loss": 2.9253,
      "theoretical_loss": 3.53803558242325,
      "tokens_seen": 1398407168
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029106318956870615,
      "loss": 2.6358,
      "theoretical_loss": 3.5380207661636742,
      "tokens_seen": 1398472704
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002910531594784353,
      "loss": 3.0897,
      "theoretical_loss": 3.538005950792811,
      "tokens_seen": 1398538240
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002910431293881645,
      "loss": 2.6811,
      "theoretical_loss": 3.5379911363105663,
      "tokens_seen": 1398603776
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002910330992978937,
      "loss": 2.7271,
      "theoretical_loss": 3.537976322716845,
      "tokens_seen": 1398669312
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002910230692076229,
      "loss": 2.6183,
      "theoretical_loss": 3.5379615100115505,
      "tokens_seen": 1398734848
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029101303911735206,
      "loss": 2.914,
      "theoretical_loss": 3.53794669819459,
      "tokens_seen": 1398800384
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002910030090270813,
      "loss": 2.9171,
      "theoretical_loss": 3.5379318872658674,
      "tokens_seen": 1398865920
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002909929789368104,
      "loss": 2.9165,
      "theoretical_loss": 3.5379170772252886,
      "tokens_seen": 1398931456
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029098294884653966,
      "loss": 2.9534,
      "theoretical_loss": 3.5379022680727576,
      "tokens_seen": 1398996992
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002909729187562688,
      "loss": 2.8827,
      "theoretical_loss": 3.537887459808181,
      "tokens_seen": 1399062528
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.000290962888665998,
      "loss": 2.9916,
      "theoretical_loss": 3.5378726524314628,
      "tokens_seen": 1399128064
    },
    {
      "epoch": 0.42,
      "objective/train/advantage_avg": 0.49427294731140137,
      "objective/train/docs_used": 792315,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.4622414112091064,
      "objective/train/original_loss": 2.4622416496276855,
      "objective/train/theoretical_loss": 3.5378578459425087,
      "objective/train/tokens_used": 1419653600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24436157941818237,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050669550895691,
      "objective/train/weighted_lm_loss": 2.5869109630584717,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0462769269943237,
      "theoretical_loss": 3.5378578459425087,
      "tokens_seen": 1399193600
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002909528585757272,
      "loss": 2.9938,
      "theoretical_loss": 3.5378578459425087,
      "tokens_seen": 1399193600
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002909428284854564,
      "loss": 2.9111,
      "theoretical_loss": 3.537843040341224,
      "tokens_seen": 1399259136
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029093279839518556,
      "loss": 2.8641,
      "theoretical_loss": 3.5378282356275137,
      "tokens_seen": 1399324672
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029092276830491474,
      "loss": 2.8684,
      "theoretical_loss": 3.5378134318012826,
      "tokens_seen": 1399390208
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002909127382146439,
      "loss": 2.753,
      "theoretical_loss": 3.5377986288624372,
      "tokens_seen": 1399455744
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029090270812437316,
      "loss": 2.8235,
      "theoretical_loss": 3.537783826810881,
      "tokens_seen": 1399521280
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002908926780341023,
      "loss": 3.0236,
      "theoretical_loss": 3.5377690256465213,
      "tokens_seen": 1399586816
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002908826479438315,
      "loss": 3.0097,
      "theoretical_loss": 3.5377542253692615,
      "tokens_seen": 1399652352
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029087261785356065,
      "loss": 2.9591,
      "theoretical_loss": 3.537739425979008,
      "tokens_seen": 1399717888
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002908625877632899,
      "loss": 2.8121,
      "theoretical_loss": 3.537724627475666,
      "tokens_seen": 1399783424
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029085255767301906,
      "loss": 2.7422,
      "theoretical_loss": 3.53770982985914,
      "tokens_seen": 1399848960
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029084252758274825,
      "loss": 2.7957,
      "theoretical_loss": 3.537695033129336,
      "tokens_seen": 1399914496
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002908324974924774,
      "loss": 2.7962,
      "theoretical_loss": 3.5376802372861604,
      "tokens_seen": 1399980032
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029082246740220666,
      "loss": 2.6896,
      "theoretical_loss": 3.5376654423295166,
      "tokens_seen": 1400045568
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002908124373119358,
      "loss": 2.7408,
      "theoretical_loss": 3.5376506482593113,
      "tokens_seen": 1400111104
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.000290802407221665,
      "loss": 2.5946,
      "theoretical_loss": 3.537635855075449,
      "tokens_seen": 1400176640
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029079237713139415,
      "loss": 2.6121,
      "theoretical_loss": 3.537621062777836,
      "tokens_seen": 1400242176
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002907823470411234,
      "loss": 2.5544,
      "theoretical_loss": 3.5376062713663776,
      "tokens_seen": 1400307712
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029077231695085257,
      "loss": 2.7853,
      "theoretical_loss": 3.5375914808409785,
      "tokens_seen": 1400373248
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029076228686058175,
      "loss": 2.8763,
      "theoretical_loss": 3.5375766912015445,
      "tokens_seen": 1400438784
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029075225677031093,
      "loss": 2.7524,
      "theoretical_loss": 3.537561902447982,
      "tokens_seen": 1400504320
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002907422266800401,
      "loss": 2.7442,
      "theoretical_loss": 3.537547114580195,
      "tokens_seen": 1400569856
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002907321965897693,
      "loss": 2.7178,
      "theoretical_loss": 3.53753232759809,
      "tokens_seen": 1400635392
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029072216649949853,
      "loss": 2.8985,
      "theoretical_loss": 3.5375175415015723,
      "tokens_seen": 1400700928
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029071213640922765,
      "loss": 3.0048,
      "theoretical_loss": 3.5375027562905474,
      "tokens_seen": 1400766464
    },
    {
      "epoch": 0.42,
      "objective/train/advantage_avg": 0.4941922128200531,
      "objective/train/docs_used": 793814,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.69189453125,
      "objective/train/original_loss": 2.691894292831421,
      "objective/train/theoretical_loss": 3.5374879719649206,
      "objective/train/tokens_used": 1421292000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24431660771369934,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506612062454224,
      "objective/train/weighted_lm_loss": 2.828240394592285,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0304546356201172,
      "theoretical_loss": 3.5374879719649206,
      "tokens_seen": 1400832000
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002907021063189569,
      "loss": 3.1014,
      "theoretical_loss": 3.5374879719649206,
      "tokens_seen": 1400832000
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.000290692076228686,
      "loss": 2.7383,
      "theoretical_loss": 3.537473188524598,
      "tokens_seen": 1400897536
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029068204613841525,
      "loss": 2.8701,
      "theoretical_loss": 3.5374584059694847,
      "tokens_seen": 1400963072
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029067201604814443,
      "loss": 2.9465,
      "theoretical_loss": 3.5374436242994864,
      "tokens_seen": 1401028608
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002906619859578736,
      "loss": 2.9,
      "theoretical_loss": 3.5374288435145083,
      "tokens_seen": 1401094144
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002906519558676028,
      "loss": 2.7259,
      "theoretical_loss": 3.537414063614457,
      "tokens_seen": 1401159680
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029064192577733203,
      "loss": 2.669,
      "theoretical_loss": 3.5373992845992372,
      "tokens_seen": 1401225216
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029063189568706116,
      "loss": 2.9183,
      "theoretical_loss": 3.5373845064687552,
      "tokens_seen": 1401290752
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002906218655967904,
      "loss": 2.9259,
      "theoretical_loss": 3.5373697292229167,
      "tokens_seen": 1401356288
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002906118355065195,
      "loss": 2.8811,
      "theoretical_loss": 3.5373549528616266,
      "tokens_seen": 1401421824
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029060180541624876,
      "loss": 2.917,
      "theoretical_loss": 3.5373401773847912,
      "tokens_seen": 1401487360
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029059177532597794,
      "loss": 2.6834,
      "theoretical_loss": 3.537325402792316,
      "tokens_seen": 1401552896
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002905817452357071,
      "loss": 2.8247,
      "theoretical_loss": 3.5373106290841068,
      "tokens_seen": 1401618432
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002905717151454363,
      "loss": 2.7888,
      "theoretical_loss": 3.5372958562600694,
      "tokens_seen": 1401683968
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002905616850551655,
      "loss": 2.8282,
      "theoretical_loss": 3.5372810843201097,
      "tokens_seen": 1401749504
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029055165496489466,
      "loss": 2.934,
      "theoretical_loss": 3.5372663132641327,
      "tokens_seen": 1401815040
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002905416248746239,
      "loss": 2.6243,
      "theoretical_loss": 3.537251543092045,
      "tokens_seen": 1401880576
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002905315947843531,
      "loss": 3.0759,
      "theoretical_loss": 3.5372367738037527,
      "tokens_seen": 1401946112
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029052156469408226,
      "loss": 2.7535,
      "theoretical_loss": 3.53722200539916,
      "tokens_seen": 1402011648
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002905115346038115,
      "loss": 2.9324,
      "theoretical_loss": 3.537207237878174,
      "tokens_seen": 1402077184
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002905015045135406,
      "loss": 2.8706,
      "theoretical_loss": 3.537192471240701,
      "tokens_seen": 1402142720
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.00029049147442326986,
      "loss": 2.5917,
      "theoretical_loss": 3.537177705486645,
      "tokens_seen": 1402208256
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.000290481444332999,
      "loss": 2.9931,
      "theoretical_loss": 3.5371629406159135,
      "tokens_seen": 1402273792
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002904714142427282,
      "loss": 3.0551,
      "theoretical_loss": 3.537148176628412,
      "tokens_seen": 1402339328
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002904613841524574,
      "loss": 2.8484,
      "theoretical_loss": 3.537133413524046,
      "tokens_seen": 1402404864
    },
    {
      "epoch": 0.42,
      "objective/train/advantage_avg": 0.4968591332435608,
      "objective/train/docs_used": 794381,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.824244737625122,
      "objective/train/original_loss": 2.824244737625122,
      "objective/train/theoretical_loss": 3.537118651302722,
      "objective/train/tokens_used": 1422930400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24688902497291565,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0509411096572876,
      "objective/train/weighted_lm_loss": 2.9680871963500977,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.048884630203247,
      "theoretical_loss": 3.537118651302722,
      "tokens_seen": 1402470400
    },
    {
      "epoch": 0.42,
      "learning_rate": 0.0002904513540621866,
      "loss": 2.6974,
      "theoretical_loss": 3.537118651302722,
      "tokens_seen": 1402470400
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00029044132397191576,
      "loss": 2.6338,
      "theoretical_loss": 3.537103889964345,
      "tokens_seen": 1402535936
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00029043129388164494,
      "loss": 2.8226,
      "theoretical_loss": 3.5370891295088223,
      "tokens_seen": 1402601472
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002904212637913741,
      "loss": 2.6321,
      "theoretical_loss": 3.5370743699360583,
      "tokens_seen": 1402667008
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00029041123370110336,
      "loss": 2.9243,
      "theoretical_loss": 3.5370596112459607,
      "tokens_seen": 1402732544
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002904012036108325,
      "loss": 2.8365,
      "theoretical_loss": 3.5370448534384336,
      "tokens_seen": 1402798080
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002903911735205617,
      "loss": 2.707,
      "theoretical_loss": 3.537030096513384,
      "tokens_seen": 1402863616
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00029038114343029085,
      "loss": 2.7918,
      "theoretical_loss": 3.537015340470718,
      "tokens_seen": 1402929152
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002903711133400201,
      "loss": 2.3793,
      "theoretical_loss": 3.5370005853103414,
      "tokens_seen": 1402994688
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00029036108324974926,
      "loss": 2.7778,
      "theoretical_loss": 3.5369858310321605,
      "tokens_seen": 1403060224
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00029035105315947845,
      "loss": 2.7134,
      "theoretical_loss": 3.536971077636081,
      "tokens_seen": 1403125760
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002903410230692076,
      "loss": 2.711,
      "theoretical_loss": 3.536956325122009,
      "tokens_seen": 1403191296
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00029033099297893686,
      "loss": 2.8674,
      "theoretical_loss": 3.536941573489851,
      "tokens_seen": 1403256832
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.000290320962888666,
      "loss": 2.6965,
      "theoretical_loss": 3.5369268227395123,
      "tokens_seen": 1403322368
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002903109327983952,
      "loss": 2.8964,
      "theoretical_loss": 3.5369120728708996,
      "tokens_seen": 1403387904
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00029030090270812435,
      "loss": 2.8001,
      "theoretical_loss": 3.5368973238839185,
      "tokens_seen": 1403453440
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002902908726178536,
      "loss": 2.6707,
      "theoretical_loss": 3.5368825757784763,
      "tokens_seen": 1403518976
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00029028084252758277,
      "loss": 2.7021,
      "theoretical_loss": 3.536867828554478,
      "tokens_seen": 1403584512
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00029027081243731195,
      "loss": 2.8518,
      "theoretical_loss": 3.53685308221183,
      "tokens_seen": 1403650048
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00029026078234704113,
      "loss": 2.604,
      "theoretical_loss": 3.536838336750439,
      "tokens_seen": 1403715584
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002902507522567703,
      "loss": 2.852,
      "theoretical_loss": 3.5368235921702107,
      "tokens_seen": 1403781120
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002902407221664995,
      "loss": 2.8181,
      "theoretical_loss": 3.5368088484710514,
      "tokens_seen": 1403846656
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00029023069207622873,
      "loss": 2.6345,
      "theoretical_loss": 3.5367941056528673,
      "tokens_seen": 1403912192
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00029022066198595785,
      "loss": 2.8288,
      "theoretical_loss": 3.536779363715565,
      "tokens_seen": 1403977728
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002902106318956871,
      "loss": 2.7953,
      "theoretical_loss": 3.5367646226590503,
      "tokens_seen": 1404043264
    },
    {
      "epoch": 0.43,
      "objective/train/advantage_avg": 0.49440479278564453,
      "objective/train/docs_used": 795506,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.7760870456695557,
      "objective/train/original_loss": 2.7760868072509766,
      "objective/train/theoretical_loss": 3.536749882483229,
      "objective/train/tokens_used": 1424568800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24448619782924652,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506832599639893,
      "objective/train/weighted_lm_loss": 2.9166972637176514,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.038975715637207,
      "theoretical_loss": 3.536749882483229,
      "tokens_seen": 1404108800
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002902006018054162,
      "loss": 2.8983,
      "theoretical_loss": 3.536749882483229,
      "tokens_seen": 1404108800
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00029019057171514545,
      "loss": 2.7925,
      "theoretical_loss": 3.536735143188009,
      "tokens_seen": 1404174336
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00029018054162487463,
      "loss": 2.7641,
      "theoretical_loss": 3.536720404773295,
      "tokens_seen": 1404239872
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002901705115346038,
      "loss": 2.699,
      "theoretical_loss": 3.5367056672389943,
      "tokens_seen": 1404305408
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.000290160481444333,
      "loss": 2.7563,
      "theoretical_loss": 3.5366909305850127,
      "tokens_seen": 1404370944
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00029015045135406223,
      "loss": 2.9293,
      "theoretical_loss": 3.5366761948112564,
      "tokens_seen": 1404436480
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00029014042126379136,
      "loss": 2.9243,
      "theoretical_loss": 3.536661459917632,
      "tokens_seen": 1404502016
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002901303911735206,
      "loss": 2.6744,
      "theoretical_loss": 3.5366467259040464,
      "tokens_seen": 1404567552
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002901203610832497,
      "loss": 2.9936,
      "theoretical_loss": 3.536631992770405,
      "tokens_seen": 1404633088
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00029011033099297896,
      "loss": 2.8527,
      "theoretical_loss": 3.536617260516615,
      "tokens_seen": 1404698624
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00029010030090270814,
      "loss": 2.75,
      "theoretical_loss": 3.536602529142582,
      "tokens_seen": 1404764160
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002900902708124373,
      "loss": 2.827,
      "theoretical_loss": 3.5365877986482133,
      "tokens_seen": 1404829696
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002900802407221665,
      "loss": 3.0311,
      "theoretical_loss": 3.5365730690334147,
      "tokens_seen": 1404895232
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002900702106318957,
      "loss": 2.8282,
      "theoretical_loss": 3.536558340298093,
      "tokens_seen": 1404960768
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00029006018054162486,
      "loss": 2.8144,
      "theoretical_loss": 3.5365436124421548,
      "tokens_seen": 1405026304
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002900501504513541,
      "loss": 2.6039,
      "theoretical_loss": 3.536528885465506,
      "tokens_seen": 1405091840
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002900401203610832,
      "loss": 2.5237,
      "theoretical_loss": 3.5365141593680534,
      "tokens_seen": 1405157376
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00029003009027081246,
      "loss": 2.5379,
      "theoretical_loss": 3.536499434149704,
      "tokens_seen": 1405222912
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002900200601805416,
      "loss": 2.685,
      "theoretical_loss": 3.536484709810364,
      "tokens_seen": 1405288448
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002900100300902708,
      "loss": 2.8908,
      "theoretical_loss": 3.536469986349939,
      "tokens_seen": 1405353984
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00029,
      "loss": 2.7409,
      "theoretical_loss": 3.536455263768337,
      "tokens_seen": 1405419520
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002899899699097292,
      "loss": 2.7363,
      "theoretical_loss": 3.5364405420654634,
      "tokens_seen": 1405485056
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028997993981945836,
      "loss": 2.8422,
      "theoretical_loss": 3.5364258212412256,
      "tokens_seen": 1405550592
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002899699097291876,
      "loss": 2.8058,
      "theoretical_loss": 3.5364111012955304,
      "tokens_seen": 1405616128
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002899598796389167,
      "loss": 3.0359,
      "theoretical_loss": 3.5363963822282836,
      "tokens_seen": 1405681664
    },
    {
      "epoch": 0.43,
      "objective/train/advantage_avg": 0.49454155564308167,
      "objective/train/docs_used": 796159,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.5895416736602783,
      "objective/train/original_loss": 2.5895419120788574,
      "objective/train/theoretical_loss": 3.536381664039392,
      "objective/train/tokens_used": 1426207200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24463027715682983,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050697684288025,
      "objective/train/weighted_lm_loss": 2.7209513187408447,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0428944826126099,
      "theoretical_loss": 3.536381664039392,
      "tokens_seen": 1405747200
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028994984954864596,
      "loss": 2.652,
      "theoretical_loss": 3.536381664039392,
      "tokens_seen": 1405747200
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002899398194583751,
      "loss": 2.6667,
      "theoretical_loss": 3.536366946728762,
      "tokens_seen": 1405812736
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002899297893681043,
      "loss": 2.9421,
      "theoretical_loss": 3.5363522302963015,
      "tokens_seen": 1405878272
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002899197592778335,
      "loss": 2.6507,
      "theoretical_loss": 3.536337514741916,
      "tokens_seen": 1405943808
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002899097291875627,
      "loss": 2.7856,
      "theoretical_loss": 3.5363228000655123,
      "tokens_seen": 1406009344
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028989969909729187,
      "loss": 2.7754,
      "theoretical_loss": 3.536308086266997,
      "tokens_seen": 1406074880
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028988966900702105,
      "loss": 2.6793,
      "theoretical_loss": 3.536293373346278,
      "tokens_seen": 1406140416
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028987963891675023,
      "loss": 2.8051,
      "theoretical_loss": 3.5362786613032604,
      "tokens_seen": 1406205952
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028986960882647946,
      "loss": 2.7925,
      "theoretical_loss": 3.536263950137852,
      "tokens_seen": 1406271488
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002898595787362086,
      "loss": 2.8711,
      "theoretical_loss": 3.536249239849959,
      "tokens_seen": 1406337024
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002898495486459378,
      "loss": 2.7901,
      "theoretical_loss": 3.536234530439488,
      "tokens_seen": 1406402560
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.000289839518555667,
      "loss": 2.6874,
      "theoretical_loss": 3.536219821906347,
      "tokens_seen": 1406468096
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002898294884653962,
      "loss": 2.861,
      "theoretical_loss": 3.536205114250442,
      "tokens_seen": 1406533632
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028981945837512537,
      "loss": 2.7876,
      "theoretical_loss": 3.536190407471679,
      "tokens_seen": 1406599168
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028980942828485455,
      "loss": 2.8901,
      "theoretical_loss": 3.536175701569966,
      "tokens_seen": 1406664704
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028979939819458373,
      "loss": 2.731,
      "theoretical_loss": 3.5361609965452097,
      "tokens_seen": 1406730240
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028978936810431297,
      "loss": 3.0311,
      "theoretical_loss": 3.5361462923973166,
      "tokens_seen": 1406795776
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028977933801404215,
      "loss": 2.8508,
      "theoretical_loss": 3.5361315891261933,
      "tokens_seen": 1406861312
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028976930792377133,
      "loss": 2.9458,
      "theoretical_loss": 3.5361168867317474,
      "tokens_seen": 1406926848
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002897592778335005,
      "loss": 2.8587,
      "theoretical_loss": 3.5361021852138848,
      "tokens_seen": 1406992384
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002897492477432297,
      "loss": 2.8915,
      "theoretical_loss": 3.5360874845725134,
      "tokens_seen": 1407057920
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028973921765295893,
      "loss": 2.8164,
      "theoretical_loss": 3.5360727848075397,
      "tokens_seen": 1407123456
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028972918756268805,
      "loss": 2.9354,
      "theoretical_loss": 3.5360580859188713,
      "tokens_seen": 1407188992
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002897191574724173,
      "loss": 2.7607,
      "theoretical_loss": 3.536043387906414,
      "tokens_seen": 1407254528
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002897091273821464,
      "loss": 2.9883,
      "theoretical_loss": 3.5360286907700758,
      "tokens_seen": 1407320064
    },
    {
      "epoch": 0.43,
      "objective/train/advantage_avg": 0.49356696009635925,
      "objective/train/docs_used": 797613,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.8455982208251953,
      "objective/train/original_loss": 2.8455982208251953,
      "objective/train/theoretical_loss": 3.536013994509762,
      "objective/train/tokens_used": 1427845600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2436605989933014,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505952835083008,
      "objective/train/weighted_lm_loss": 2.989692211151123,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0467102527618408,
      "theoretical_loss": 3.536013994509762,
      "tokens_seen": 1407385600
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028969909729187565,
      "loss": 3.0343,
      "theoretical_loss": 3.536013994509762,
      "tokens_seen": 1407385600
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028968906720160483,
      "loss": 3.0174,
      "theoretical_loss": 3.5359992991253817,
      "tokens_seen": 1407451136
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.000289679037111334,
      "loss": 2.9067,
      "theoretical_loss": 3.535984604616841,
      "tokens_seen": 1407516672
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002896690070210632,
      "loss": 2.8301,
      "theoretical_loss": 3.535969910984047,
      "tokens_seen": 1407582208
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028965897693079243,
      "loss": 3.0604,
      "theoretical_loss": 3.535955218226907,
      "tokens_seen": 1407647744
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028964894684052156,
      "loss": 2.7303,
      "theoretical_loss": 3.5359405263453274,
      "tokens_seen": 1407713280
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002896389167502508,
      "loss": 2.9476,
      "theoretical_loss": 3.5359258353392153,
      "tokens_seen": 1407778816
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002896288866599799,
      "loss": 2.8074,
      "theoretical_loss": 3.535911145208478,
      "tokens_seen": 1407844352
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028961885656970916,
      "loss": 2.778,
      "theoretical_loss": 3.535896455953023,
      "tokens_seen": 1407909888
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028960882647943834,
      "loss": 2.7495,
      "theoretical_loss": 3.535881767572757,
      "tokens_seen": 1407975424
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002895987963891675,
      "loss": 2.7056,
      "theoretical_loss": 3.535867080067587,
      "tokens_seen": 1408040960
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002895887662988967,
      "loss": 2.886,
      "theoretical_loss": 3.5358523934374206,
      "tokens_seen": 1408106496
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002895787362086259,
      "loss": 2.7516,
      "theoretical_loss": 3.5358377076821643,
      "tokens_seen": 1408172032
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028956870611835506,
      "loss": 2.8185,
      "theoretical_loss": 3.5358230228017264,
      "tokens_seen": 1408237568
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002895586760280843,
      "loss": 2.6495,
      "theoretical_loss": 3.535808338796013,
      "tokens_seen": 1408303104
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002895486459378134,
      "loss": 2.8154,
      "theoretical_loss": 3.535793655664931,
      "tokens_seen": 1408368640
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028953861584754266,
      "loss": 2.9779,
      "theoretical_loss": 3.535778973408389,
      "tokens_seen": 1408434176
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002895285857572718,
      "loss": 2.8678,
      "theoretical_loss": 3.5357642920262933,
      "tokens_seen": 1408499712
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.000289518555667001,
      "loss": 2.9124,
      "theoretical_loss": 3.535749611518551,
      "tokens_seen": 1408565248
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002895085255767302,
      "loss": 2.7596,
      "theoretical_loss": 3.53573493188507,
      "tokens_seen": 1408630784
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002894984954864594,
      "loss": 2.6088,
      "theoretical_loss": 3.535720253125757,
      "tokens_seen": 1408696320
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028948846539618856,
      "loss": 2.9603,
      "theoretical_loss": 3.5357055752405193,
      "tokens_seen": 1408761856
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002894784353059178,
      "loss": 2.9322,
      "theoretical_loss": 3.535690898229265,
      "tokens_seen": 1408827392
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002894684052156469,
      "loss": 2.9129,
      "theoretical_loss": 3.5356762220919,
      "tokens_seen": 1408892928
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028945837512537616,
      "loss": 2.7814,
      "theoretical_loss": 3.535661546828333,
      "tokens_seen": 1408958464
    },
    {
      "epoch": 0.43,
      "objective/train/advantage_avg": 0.49365124106407166,
      "objective/train/docs_used": 798401,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.6763062477111816,
      "objective/train/original_loss": 2.6763064861297607,
      "objective/train/theoretical_loss": 3.5356468724384706,
      "objective/train/tokens_used": 1429484000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24379035830497742,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506044626235962,
      "objective/train/weighted_lm_loss": 2.8115828037261963,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0304546356201172,
      "theoretical_loss": 3.5356468724384706,
      "tokens_seen": 1409024000
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002894483450351053,
      "loss": 2.8619,
      "theoretical_loss": 3.5356468724384706,
      "tokens_seen": 1409024000
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002894383149448345,
      "loss": 2.8547,
      "theoretical_loss": 3.5356321989222197,
      "tokens_seen": 1409089536
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002894282848545637,
      "loss": 2.9183,
      "theoretical_loss": 3.535617526279489,
      "tokens_seen": 1409155072
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002894182547642929,
      "loss": 2.8563,
      "theoretical_loss": 3.5356028545101847,
      "tokens_seen": 1409220608
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028940822467402207,
      "loss": 2.8455,
      "theoretical_loss": 3.535588183614215,
      "tokens_seen": 1409286144
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028939819458375125,
      "loss": 3.0639,
      "theoretical_loss": 3.535573513591487,
      "tokens_seen": 1409351680
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028938816449348043,
      "loss": 2.9461,
      "theoretical_loss": 3.5355588444419075,
      "tokens_seen": 1409417216
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028937813440320966,
      "loss": 2.8265,
      "theoretical_loss": 3.5355441761653847,
      "tokens_seen": 1409482752
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002893681043129388,
      "loss": 2.7579,
      "theoretical_loss": 3.535529508761826,
      "tokens_seen": 1409548288
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.000289358074222668,
      "loss": 2.9032,
      "theoretical_loss": 3.535514842231138,
      "tokens_seen": 1409613824
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002893480441323972,
      "loss": 3.0327,
      "theoretical_loss": 3.53550017657323,
      "tokens_seen": 1409679360
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002893380140421264,
      "loss": 2.9224,
      "theoretical_loss": 3.535485511788008,
      "tokens_seen": 1409744896
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028932798395185557,
      "loss": 3.0156,
      "theoretical_loss": 3.53547084787538,
      "tokens_seen": 1409810432
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028931795386158475,
      "loss": 2.8023,
      "theoretical_loss": 3.535456184835253,
      "tokens_seen": 1409875968
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028930792377131393,
      "loss": 3.2131,
      "theoretical_loss": 3.5354415226675346,
      "tokens_seen": 1409941504
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028929789368104317,
      "loss": 2.8677,
      "theoretical_loss": 3.535426861372133,
      "tokens_seen": 1410007040
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002892878635907723,
      "loss": 2.8016,
      "theoretical_loss": 3.5354122009489553,
      "tokens_seen": 1410072576
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028927783350050153,
      "loss": 2.7964,
      "theoretical_loss": 3.5353975413979093,
      "tokens_seen": 1410138112
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028926780341023066,
      "loss": 2.7139,
      "theoretical_loss": 3.5353828827189027,
      "tokens_seen": 1410203648
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002892577733199599,
      "loss": 2.7979,
      "theoretical_loss": 3.535368224911843,
      "tokens_seen": 1410269184
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002892477432296891,
      "loss": 2.9329,
      "theoretical_loss": 3.535353567976637,
      "tokens_seen": 1410334720
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028923771313941825,
      "loss": 2.9766,
      "theoretical_loss": 3.5353389119131937,
      "tokens_seen": 1410400256
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028922768304914744,
      "loss": 2.5584,
      "theoretical_loss": 3.5353242567214203,
      "tokens_seen": 1410465792
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002892176529588766,
      "loss": 2.8307,
      "theoretical_loss": 3.535309602401224,
      "tokens_seen": 1410531328
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002892076228686058,
      "loss": 2.8865,
      "theoretical_loss": 3.535294948952512,
      "tokens_seen": 1410596864
    },
    {
      "epoch": 0.43,
      "objective/train/advantage_avg": 0.49515268206596375,
      "objective/train/docs_used": 799700,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.9228596687316895,
      "objective/train/original_loss": 2.9228596687316895,
      "objective/train/theoretical_loss": 3.5352802963751935,
      "objective/train/tokens_used": 1431122400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24525710940361023,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507620573043823,
      "objective/train/weighted_lm_loss": 3.071223020553589,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0363596677780151,
      "theoretical_loss": 3.5352802963751935,
      "tokens_seen": 1410662400
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028919759277833503,
      "loss": 3.0635,
      "theoretical_loss": 3.5352802963751935,
      "tokens_seen": 1410662400
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028918756268806416,
      "loss": 2.6923,
      "theoretical_loss": 3.535265644669175,
      "tokens_seen": 1410727936
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002891775325977934,
      "loss": 2.8776,
      "theoretical_loss": 3.5352509938343646,
      "tokens_seen": 1410793472
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002891675025075226,
      "loss": 2.6818,
      "theoretical_loss": 3.535236343870671,
      "tokens_seen": 1410859008
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028915747241725176,
      "loss": 3.0342,
      "theoretical_loss": 3.5352216947779995,
      "tokens_seen": 1410924544
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028914744232698094,
      "loss": 2.8231,
      "theoretical_loss": 3.5352070465562604,
      "tokens_seen": 1410990080
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002891374122367101,
      "loss": 2.7632,
      "theoretical_loss": 3.5351923992053598,
      "tokens_seen": 1411055616
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002891273821464393,
      "loss": 2.8806,
      "theoretical_loss": 3.5351777527252066,
      "tokens_seen": 1411121152
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028911735205616854,
      "loss": 2.7985,
      "theoretical_loss": 3.535163107115708,
      "tokens_seen": 1411186688
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028910732196589766,
      "loss": 2.9068,
      "theoretical_loss": 3.535148462376772,
      "tokens_seen": 1411252224
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002890972918756269,
      "loss": 2.9158,
      "theoretical_loss": 3.5351338185083065,
      "tokens_seen": 1411317760
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.000289087261785356,
      "loss": 2.8617,
      "theoretical_loss": 3.5351191755102187,
      "tokens_seen": 1411383296
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028907723169508526,
      "loss": 3.0095,
      "theoretical_loss": 3.5351045333824174,
      "tokens_seen": 1411448832
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028906720160481444,
      "loss": 2.706,
      "theoretical_loss": 3.5350898921248097,
      "tokens_seen": 1411514368
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002890571715145436,
      "loss": 2.7923,
      "theoretical_loss": 3.5350752517373043,
      "tokens_seen": 1411579904
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002890471414242728,
      "loss": 2.9786,
      "theoretical_loss": 3.535060612219808,
      "tokens_seen": 1411645440
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.000289037111334002,
      "loss": 2.7004,
      "theoretical_loss": 3.53504597357223,
      "tokens_seen": 1411710976
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002890270812437312,
      "loss": 2.653,
      "theoretical_loss": 3.535031335794477,
      "tokens_seen": 1411776512
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002890170511534604,
      "loss": 2.8377,
      "theoretical_loss": 3.5350166988864578,
      "tokens_seen": 1411842048
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002890070210631896,
      "loss": 2.7274,
      "theoretical_loss": 3.53500206284808,
      "tokens_seen": 1411907584
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028899699097291876,
      "loss": 3.1001,
      "theoretical_loss": 3.5349874276792512,
      "tokens_seen": 1411973120
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.000288986960882648,
      "loss": 2.8972,
      "theoretical_loss": 3.5349727933798807,
      "tokens_seen": 1412038656
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002889769307923771,
      "loss": 2.8647,
      "theoretical_loss": 3.5349581599498747,
      "tokens_seen": 1412104192
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028896690070210636,
      "loss": 2.7454,
      "theoretical_loss": 3.5349435273891423,
      "tokens_seen": 1412169728
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002889568706118355,
      "loss": 2.6984,
      "theoretical_loss": 3.5349288956975915,
      "tokens_seen": 1412235264
    },
    {
      "epoch": 0.43,
      "objective/train/advantage_avg": 0.49382442235946655,
      "objective/train/docs_used": 800754,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.058685064315796,
      "objective/train/original_loss": 2.058685064315796,
      "objective/train/theoretical_loss": 3.5349142648751304,
      "objective/train/tokens_used": 1432760800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24391134083271027,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506223440170288,
      "objective/train/weighted_lm_loss": 2.162693977355957,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.044721245765686,
      "theoretical_loss": 3.5349142648751304,
      "tokens_seen": 1412300800
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002889468405215647,
      "loss": 2.595,
      "theoretical_loss": 3.5349142648751304,
      "tokens_seen": 1412300800
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002889368104312939,
      "loss": 2.845,
      "theoretical_loss": 3.5348996349216666,
      "tokens_seen": 1412366336
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002889267803410231,
      "loss": 2.7625,
      "theoretical_loss": 3.534885005837108,
      "tokens_seen": 1412431872
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028891675025075227,
      "loss": 2.8069,
      "theoretical_loss": 3.5348703776213632,
      "tokens_seen": 1412497408
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028890672016048145,
      "loss": 2.851,
      "theoretical_loss": 3.5348557502743407,
      "tokens_seen": 1412562944
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028889669007021063,
      "loss": 3.056,
      "theoretical_loss": 3.534841123795948,
      "tokens_seen": 1412628480
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028888665997993986,
      "loss": 2.8656,
      "theoretical_loss": 3.534826498186093,
      "tokens_seen": 1412694016
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.000288876629889669,
      "loss": 2.8486,
      "theoretical_loss": 3.534811873444684,
      "tokens_seen": 1412759552
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028886659979939823,
      "loss": 2.9873,
      "theoretical_loss": 3.5347972495716293,
      "tokens_seen": 1412825088
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002888565697091274,
      "loss": 3.1424,
      "theoretical_loss": 3.5347826265668374,
      "tokens_seen": 1412890624
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002888465396188566,
      "loss": 2.9321,
      "theoretical_loss": 3.534768004430216,
      "tokens_seen": 1412956160
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028883650952858577,
      "loss": 2.7859,
      "theoretical_loss": 3.5347533831616733,
      "tokens_seen": 1413021696
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028882647943831495,
      "loss": 2.9837,
      "theoretical_loss": 3.5347387627611173,
      "tokens_seen": 1413087232
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028881644934804413,
      "loss": 2.6273,
      "theoretical_loss": 3.5347241432284573,
      "tokens_seen": 1413152768
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028880641925777337,
      "loss": 2.7663,
      "theoretical_loss": 3.5347095245636,
      "tokens_seen": 1413218304
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002887963891675025,
      "loss": 2.7595,
      "theoretical_loss": 3.534694906766455,
      "tokens_seen": 1413283840
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028878635907723173,
      "loss": 2.8357,
      "theoretical_loss": 3.5346802898369303,
      "tokens_seen": 1413349376
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028877632898696086,
      "loss": 2.7602,
      "theoretical_loss": 3.534665673774933,
      "tokens_seen": 1413414912
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002887662988966901,
      "loss": 2.7931,
      "theoretical_loss": 3.5346510585803728,
      "tokens_seen": 1413480448
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002887562688064193,
      "loss": 2.9974,
      "theoretical_loss": 3.534636444253157,
      "tokens_seen": 1413545984
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028874623871614845,
      "loss": 2.6171,
      "theoretical_loss": 3.5346218307931947,
      "tokens_seen": 1413611520
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028873620862587764,
      "loss": 2.8964,
      "theoretical_loss": 3.534607218200394,
      "tokens_seen": 1413677056
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002887261785356068,
      "loss": 2.8581,
      "theoretical_loss": 3.5345926064746633,
      "tokens_seen": 1413742592
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.000288716148445336,
      "loss": 3.0728,
      "theoretical_loss": 3.5345779956159102,
      "tokens_seen": 1413808128
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028870611835506523,
      "loss": 2.8258,
      "theoretical_loss": 3.534563385624044,
      "tokens_seen": 1413873664
    },
    {
      "epoch": 0.43,
      "objective/train/advantage_avg": 0.4945061206817627,
      "objective/train/docs_used": 801653,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.9477858543395996,
      "objective/train/original_loss": 2.9477853775024414,
      "objective/train/theoretical_loss": 3.5345487764989727,
      "objective/train/tokens_used": 1434399200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24461787939071655,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506941080093384,
      "objective/train/weighted_lm_loss": 3.0974953174591064,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0253151655197144,
      "theoretical_loss": 3.5345487764989727,
      "tokens_seen": 1413939200
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028869608826479436,
      "loss": 3.001,
      "theoretical_loss": 3.5345487764989727,
      "tokens_seen": 1413939200
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002886860581745236,
      "loss": 2.9038,
      "theoretical_loss": 3.5345341682406053,
      "tokens_seen": 1414004736
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002886760280842528,
      "loss": 3.0695,
      "theoretical_loss": 3.534519560848849,
      "tokens_seen": 1414070272
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028866599799398196,
      "loss": 2.6993,
      "theoretical_loss": 3.534504954323613,
      "tokens_seen": 1414135808
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028865596790371114,
      "loss": 2.5754,
      "theoretical_loss": 3.5344903486648063,
      "tokens_seen": 1414201344
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002886459378134403,
      "loss": 2.8784,
      "theoretical_loss": 3.534475743872336,
      "tokens_seen": 1414266880
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002886359077231695,
      "loss": 2.8471,
      "theoretical_loss": 3.534461139946112,
      "tokens_seen": 1414332416
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028862587763289874,
      "loss": 2.9413,
      "theoretical_loss": 3.5344465368860414,
      "tokens_seen": 1414397952
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028861584754262786,
      "loss": 2.9662,
      "theoretical_loss": 3.5344319346920336,
      "tokens_seen": 1414463488
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002886058174523571,
      "loss": 2.9169,
      "theoretical_loss": 3.534417333363997,
      "tokens_seen": 1414529024
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002885957873620862,
      "loss": 2.7354,
      "theoretical_loss": 3.53440273290184,
      "tokens_seen": 1414594560
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028858575727181546,
      "loss": 2.8981,
      "theoretical_loss": 3.5343881333054714,
      "tokens_seen": 1414660096
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028857572718154464,
      "loss": 2.9514,
      "theoretical_loss": 3.534373534574799,
      "tokens_seen": 1414725632
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002885656970912738,
      "loss": 2.9234,
      "theoretical_loss": 3.534358936709732,
      "tokens_seen": 1414791168
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.000288555667001003,
      "loss": 2.9039,
      "theoretical_loss": 3.5343443397101795,
      "tokens_seen": 1414856704
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002885456369107322,
      "loss": 2.9209,
      "theoretical_loss": 3.5343297435760492,
      "tokens_seen": 1414922240
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028853560682046137,
      "loss": 2.7802,
      "theoretical_loss": 3.5343151483072495,
      "tokens_seen": 1414987776
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002885255767301906,
      "loss": 2.875,
      "theoretical_loss": 3.5343005539036896,
      "tokens_seen": 1415053312
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028851554663991973,
      "loss": 2.7336,
      "theoretical_loss": 3.5342859603652785,
      "tokens_seen": 1415118848
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028850551654964896,
      "loss": 2.785,
      "theoretical_loss": 3.5342713676919235,
      "tokens_seen": 1415184384
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028849548645937815,
      "loss": 3.0188,
      "theoretical_loss": 3.5342567758835353,
      "tokens_seen": 1415249920
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002884854563691073,
      "loss": 2.7176,
      "theoretical_loss": 3.5342421849400205,
      "tokens_seen": 1415315456
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002884754262788365,
      "loss": 2.8722,
      "theoretical_loss": 3.534227594861289,
      "tokens_seen": 1415380992
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002884653961885657,
      "loss": 2.7862,
      "theoretical_loss": 3.5342130056472496,
      "tokens_seen": 1415446528
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028845536609829487,
      "loss": 2.9479,
      "theoretical_loss": 3.5341984172978105,
      "tokens_seen": 1415512064
    },
    {
      "epoch": 0.43,
      "objective/train/advantage_avg": 0.4935374855995178,
      "objective/train/docs_used": 802505,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.0536317825317383,
      "objective/train/original_loss": 3.053631544113159,
      "objective/train/theoretical_loss": 3.5341838298128803,
      "objective/train/tokens_used": 1436037600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2436642199754715,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050592303276062,
      "objective/train/weighted_lm_loss": 3.2081761360168457,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0442860126495361,
      "theoretical_loss": 3.5341838298128803,
      "tokens_seen": 1415577600
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002884453360080241,
      "loss": 2.7759,
      "theoretical_loss": 3.5341838298128803,
      "tokens_seen": 1415577600
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028843530591775323,
      "loss": 2.9557,
      "theoretical_loss": 3.534169243192368,
      "tokens_seen": 1415643136
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028842527582748247,
      "loss": 2.8018,
      "theoretical_loss": 3.534154657436183,
      "tokens_seen": 1415708672
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002884152457372116,
      "loss": 2.9256,
      "theoretical_loss": 3.5341400725442327,
      "tokens_seen": 1415774208
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028840521564694083,
      "loss": 2.8629,
      "theoretical_loss": 3.5341254885164273,
      "tokens_seen": 1415839744
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028839518555667,
      "loss": 2.8246,
      "theoretical_loss": 3.5341109053526747,
      "tokens_seen": 1415905280
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002883851554663992,
      "loss": 2.984,
      "theoretical_loss": 3.534096323052884,
      "tokens_seen": 1415970816
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002883751253761284,
      "loss": 2.6656,
      "theoretical_loss": 3.534081741616964,
      "tokens_seen": 1416036352
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002883650952858576,
      "loss": 2.7161,
      "theoretical_loss": 3.5340671610448235,
      "tokens_seen": 1416101888
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028835506519558674,
      "loss": 2.8342,
      "theoretical_loss": 3.534052581336372,
      "tokens_seen": 1416167424
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028834503510531597,
      "loss": 2.8861,
      "theoretical_loss": 3.534038002491517,
      "tokens_seen": 1416232960
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002883350050150451,
      "loss": 3.1827,
      "theoretical_loss": 3.5340234245101687,
      "tokens_seen": 1416298496
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028832497492477433,
      "loss": 2.9255,
      "theoretical_loss": 3.5340088473922355,
      "tokens_seen": 1416364032
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002883149448345035,
      "loss": 2.8348,
      "theoretical_loss": 3.533994271137626,
      "tokens_seen": 1416429568
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002883049147442327,
      "loss": 2.8075,
      "theoretical_loss": 3.5339796957462495,
      "tokens_seen": 1416495104
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002882948846539619,
      "loss": 2.8069,
      "theoretical_loss": 3.5339651212180145,
      "tokens_seen": 1416560640
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028828485456369106,
      "loss": 2.9327,
      "theoretical_loss": 3.5339505475528314,
      "tokens_seen": 1416626176
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002882748244734203,
      "loss": 2.9005,
      "theoretical_loss": 3.533935974750607,
      "tokens_seen": 1416691712
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002882647943831495,
      "loss": 2.7099,
      "theoretical_loss": 3.533921402811252,
      "tokens_seen": 1416757248
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028825476429287866,
      "loss": 2.792,
      "theoretical_loss": 3.533906831734675,
      "tokens_seen": 1416822784
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028824473420260784,
      "loss": 2.9298,
      "theoretical_loss": 3.5338922615207844,
      "tokens_seen": 1416888320
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.000288234704112337,
      "loss": 2.9355,
      "theoretical_loss": 3.533877692169489,
      "tokens_seen": 1416953856
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002882246740220662,
      "loss": 2.9205,
      "theoretical_loss": 3.5338631236806997,
      "tokens_seen": 1417019392
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028821464393179543,
      "loss": 2.8488,
      "theoretical_loss": 3.533848556054324,
      "tokens_seen": 1417084928
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028820461384152456,
      "loss": 2.8303,
      "theoretical_loss": 3.5338339892902706,
      "tokens_seen": 1417150464
    },
    {
      "epoch": 0.43,
      "objective/train/advantage_avg": 0.49562254548072815,
      "objective/train/docs_used": 803171,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.8683207035064697,
      "objective/train/original_loss": 2.868321180343628,
      "objective/train/theoretical_loss": 3.53381942338845,
      "objective/train/tokens_used": 1437676000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24570146203041077,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0508112907409668,
      "objective/train/weighted_lm_loss": 3.014037847518921,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0304546356201172,
      "theoretical_loss": 3.53381942338845,
      "tokens_seen": 1417216000
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002881945837512538,
      "loss": 2.943,
      "theoretical_loss": 3.53381942338845,
      "tokens_seen": 1417216000
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.000288184553660983,
      "loss": 2.8807,
      "theoretical_loss": 3.5338048583487702,
      "tokens_seen": 1417281536
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028817452357071216,
      "loss": 2.8815,
      "theoretical_loss": 3.5337902941711405,
      "tokens_seen": 1417347072
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028816449348044134,
      "loss": 2.5444,
      "theoretical_loss": 3.5337757308554707,
      "tokens_seen": 1417412608
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002881544633901705,
      "loss": 2.9116,
      "theoretical_loss": 3.533761168401669,
      "tokens_seen": 1417478144
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002881444332998997,
      "loss": 2.7497,
      "theoretical_loss": 3.533746606809645,
      "tokens_seen": 1417543680
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028813440320962894,
      "loss": 2.9444,
      "theoretical_loss": 3.533732046079308,
      "tokens_seen": 1417609216
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028812437311935806,
      "loss": 2.7933,
      "theoretical_loss": 3.533717486210567,
      "tokens_seen": 1417674752
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002881143430290873,
      "loss": 2.8987,
      "theoretical_loss": 3.533702927203332,
      "tokens_seen": 1417740288
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002881043129388164,
      "loss": 3.0598,
      "theoretical_loss": 3.5336883690575105,
      "tokens_seen": 1417805824
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028809428284854566,
      "loss": 2.9514,
      "theoretical_loss": 3.533673811773012,
      "tokens_seen": 1417871360
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028808425275827484,
      "loss": 2.7808,
      "theoretical_loss": 3.5336592553497477,
      "tokens_seen": 1417936896
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.000288074222668004,
      "loss": 2.8435,
      "theoretical_loss": 3.533644699787625,
      "tokens_seen": 1418002432
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002880641925777332,
      "loss": 2.561,
      "theoretical_loss": 3.533630145086553,
      "tokens_seen": 1418067968
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002880541624874624,
      "loss": 2.7728,
      "theoretical_loss": 3.533615591246442,
      "tokens_seen": 1418133504
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028804413239719157,
      "loss": 2.7365,
      "theoretical_loss": 3.533601038267202,
      "tokens_seen": 1418199040
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002880341023069208,
      "loss": 2.8102,
      "theoretical_loss": 3.53358648614874,
      "tokens_seen": 1418264576
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028802407221664993,
      "loss": 2.934,
      "theoretical_loss": 3.5335719348909667,
      "tokens_seen": 1418330112
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028801404212637916,
      "loss": 2.8799,
      "theoretical_loss": 3.5335573844937915,
      "tokens_seen": 1418395648
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028800401203610835,
      "loss": 2.8085,
      "theoretical_loss": 3.5335428349571236,
      "tokens_seen": 1418461184
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002879939819458375,
      "loss": 2.8395,
      "theoretical_loss": 3.5335282862808723,
      "tokens_seen": 1418526720
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002879839518555667,
      "loss": 2.842,
      "theoretical_loss": 3.533513738464947,
      "tokens_seen": 1418592256
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002879739217652959,
      "loss": 2.9035,
      "theoretical_loss": 3.5334991915092564,
      "tokens_seen": 1418657792
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028796389167502507,
      "loss": 2.8805,
      "theoretical_loss": 3.5334846454137114,
      "tokens_seen": 1418723328
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002879538615847543,
      "loss": 2.6648,
      "theoretical_loss": 3.53347010017822,
      "tokens_seen": 1418788864
    },
    {
      "epoch": 0.43,
      "objective/train/advantage_avg": 0.4923365116119385,
      "objective/train/docs_used": 804514,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.5292587280273438,
      "objective/train/original_loss": 2.529259204864502,
      "objective/train/theoretical_loss": 3.533455555802692,
      "objective/train/tokens_used": 1439314400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24259129166603088,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0504666566848755,
      "objective/train/weighted_lm_loss": 2.657050848007202,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0338950157165527,
      "theoretical_loss": 3.533455555802692,
      "tokens_seen": 1418854400
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028794383149448343,
      "loss": 2.9411,
      "theoretical_loss": 3.533455555802692,
      "tokens_seen": 1418854400
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028793380140421267,
      "loss": 2.8854,
      "theoretical_loss": 3.5334410122870374,
      "tokens_seen": 1418919936
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002879237713139418,
      "loss": 2.8402,
      "theoretical_loss": 3.5334264696311655,
      "tokens_seen": 1418985472
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028791374122367103,
      "loss": 2.6397,
      "theoretical_loss": 3.533411927834985,
      "tokens_seen": 1419051008
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002879037111334002,
      "loss": 2.7458,
      "theoretical_loss": 3.5333973868984057,
      "tokens_seen": 1419116544
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002878936810431294,
      "loss": 3.0717,
      "theoretical_loss": 3.533382846821338,
      "tokens_seen": 1419182080
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002878836509528586,
      "loss": 2.9765,
      "theoretical_loss": 3.5333683076036904,
      "tokens_seen": 1419247616
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002878736208625878,
      "loss": 2.6758,
      "theoretical_loss": 3.5333537692453723,
      "tokens_seen": 1419313152
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028786359077231694,
      "loss": 2.7484,
      "theoretical_loss": 3.533339231746294,
      "tokens_seen": 1419378688
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028785356068204617,
      "loss": 2.9813,
      "theoretical_loss": 3.5333246951063644,
      "tokens_seen": 1419444224
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002878435305917753,
      "loss": 2.9242,
      "theoretical_loss": 3.533310159325494,
      "tokens_seen": 1419509760
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028783350050150453,
      "loss": 2.8519,
      "theoretical_loss": 3.533295624403591,
      "tokens_seen": 1419575296
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002878234704112337,
      "loss": 2.8621,
      "theoretical_loss": 3.5332810903405663,
      "tokens_seen": 1419640832
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002878134403209629,
      "loss": 2.7531,
      "theoretical_loss": 3.5332665571363284,
      "tokens_seen": 1419706368
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002878034102306921,
      "loss": 2.975,
      "theoretical_loss": 3.533252024790788,
      "tokens_seen": 1419771904
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028779338014042126,
      "loss": 2.9299,
      "theoretical_loss": 3.5332374933038535,
      "tokens_seen": 1419837440
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028778335005015044,
      "loss": 2.6752,
      "theoretical_loss": 3.5332229626754357,
      "tokens_seen": 1419902976
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002877733199598797,
      "loss": 2.9733,
      "theoretical_loss": 3.533208432905444,
      "tokens_seen": 1419968512
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002877632898696088,
      "loss": 2.5476,
      "theoretical_loss": 3.533193903993787,
      "tokens_seen": 1420034048
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028775325977933804,
      "loss": 2.927,
      "theoretical_loss": 3.5331793759403762,
      "tokens_seen": 1420099584
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028774322968906716,
      "loss": 2.6308,
      "theoretical_loss": 3.5331648487451197,
      "tokens_seen": 1420165120
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002877331995987964,
      "loss": 2.7589,
      "theoretical_loss": 3.5331503224079284,
      "tokens_seen": 1420230656
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002877231695085256,
      "loss": 2.8739,
      "theoretical_loss": 3.533135796928711,
      "tokens_seen": 1420296192
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028771313941825476,
      "loss": 2.5997,
      "theoretical_loss": 3.533121272307378,
      "tokens_seen": 1420361728
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028770310932798394,
      "loss": 2.7716,
      "theoretical_loss": 3.5331067485438385,
      "tokens_seen": 1420427264
    },
    {
      "epoch": 0.43,
      "objective/train/advantage_avg": 0.49502548575401306,
      "objective/train/docs_used": 804990,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.898453712463379,
      "objective/train/original_loss": 2.898453950881958,
      "objective/train/theoretical_loss": 3.5330922256380024,
      "objective/train/tokens_used": 1440952800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24512560665607452,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050748586654663,
      "objective/train/weighted_lm_loss": 3.0455362796783447,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0363596677780151,
      "theoretical_loss": 3.5330922256380024,
      "tokens_seen": 1420492800
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002876930792377132,
      "loss": 2.8198,
      "theoretical_loss": 3.5330922256380024,
      "tokens_seen": 1420492800
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002876830491474423,
      "loss": 2.8875,
      "theoretical_loss": 3.5330777035897807,
      "tokens_seen": 1420558336
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028767301905717154,
      "loss": 2.7016,
      "theoretical_loss": 3.5330631823990815,
      "tokens_seen": 1420623872
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028766298896690067,
      "loss": 2.5861,
      "theoretical_loss": 3.5330486620658155,
      "tokens_seen": 1420689408
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002876529588766299,
      "loss": 2.6473,
      "theoretical_loss": 3.533034142589892,
      "tokens_seen": 1420754944
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002876429287863591,
      "loss": 2.9279,
      "theoretical_loss": 3.5330196239712217,
      "tokens_seen": 1420820480
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028763289869608826,
      "loss": 2.8041,
      "theoretical_loss": 3.533005106209713,
      "tokens_seen": 1420886016
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028762286860581745,
      "loss": 2.9694,
      "theoretical_loss": 3.5329905893052778,
      "tokens_seen": 1420951552
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002876128385155466,
      "loss": 2.7139,
      "theoretical_loss": 3.5329760732578244,
      "tokens_seen": 1421017088
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002876028084252758,
      "loss": 2.8695,
      "theoretical_loss": 3.532961558067263,
      "tokens_seen": 1421082624
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028759277833500504,
      "loss": 2.9216,
      "theoretical_loss": 3.532947043733504,
      "tokens_seen": 1421148160
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028758274824473417,
      "loss": 2.6632,
      "theoretical_loss": 3.5329325302564563,
      "tokens_seen": 1421213696
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002875727181544634,
      "loss": 2.915,
      "theoretical_loss": 3.532918017636031,
      "tokens_seen": 1421279232
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028756268806419253,
      "loss": 2.6997,
      "theoretical_loss": 3.532903505872137,
      "tokens_seen": 1421344768
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028755265797392177,
      "loss": 2.9451,
      "theoretical_loss": 3.5328889949646856,
      "tokens_seen": 1421410304
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028754262788365095,
      "loss": 2.984,
      "theoretical_loss": 3.532874484913586,
      "tokens_seen": 1421475840
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028753259779338013,
      "loss": 2.8587,
      "theoretical_loss": 3.5328599757187478,
      "tokens_seen": 1421541376
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028752256770310936,
      "loss": 2.7376,
      "theoretical_loss": 3.5328454673800813,
      "tokens_seen": 1421606912
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028751253761283855,
      "loss": 2.818,
      "theoretical_loss": 3.5328309598974963,
      "tokens_seen": 1421672448
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002875025075225677,
      "loss": 2.8618,
      "theoretical_loss": 3.5328164532709034,
      "tokens_seen": 1421737984
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002874924774322969,
      "loss": 2.7567,
      "theoretical_loss": 3.5328019475002126,
      "tokens_seen": 1421803520
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002874824473420261,
      "loss": 2.9754,
      "theoretical_loss": 3.532787442585333,
      "tokens_seen": 1421869056
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028747241725175527,
      "loss": 2.7416,
      "theoretical_loss": 3.532772938526176,
      "tokens_seen": 1421934592
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002874623871614845,
      "loss": 2.8511,
      "theoretical_loss": 3.53275843532265,
      "tokens_seen": 1422000128
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028745235707121363,
      "loss": 2.9796,
      "theoretical_loss": 3.5327439329746673,
      "tokens_seen": 1422065664
    },
    {
      "epoch": 0.43,
      "objective/train/advantage_avg": 0.49424776434898376,
      "objective/train/docs_used": 806295,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.514603853225708,
      "objective/train/original_loss": 2.514604330062866,
      "objective/train/theoretical_loss": 3.5327294314821365,
      "objective/train/tokens_used": 1442591200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24438931047916412,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506670475006104,
      "objective/train/weighted_lm_loss": 2.6418697834014893,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0304546356201172,
      "theoretical_loss": 3.5327294314821365,
      "tokens_seen": 1422131200
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028744232698094287,
      "loss": 2.8673,
      "theoretical_loss": 3.5327294314821365,
      "tokens_seen": 1422131200
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.000287432296890672,
      "loss": 2.7966,
      "theoretical_loss": 3.5327149308449677,
      "tokens_seen": 1422196736
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028742226680040123,
      "loss": 2.7306,
      "theoretical_loss": 3.532700431063072,
      "tokens_seen": 1422262272
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002874122367101304,
      "loss": 2.759,
      "theoretical_loss": 3.532685932136358,
      "tokens_seen": 1422327808
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002874022066198596,
      "loss": 2.8538,
      "theoretical_loss": 3.5326714340647376,
      "tokens_seen": 1422393344
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002873921765295888,
      "loss": 2.9903,
      "theoretical_loss": 3.53265693684812,
      "tokens_seen": 1422458880
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.000287382146439318,
      "loss": 2.9662,
      "theoretical_loss": 3.5326424404864154,
      "tokens_seen": 1422524416
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028737211634904714,
      "loss": 2.9369,
      "theoretical_loss": 3.5326279449795344,
      "tokens_seen": 1422589952
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028736208625877637,
      "loss": 3.2073,
      "theoretical_loss": 3.532613450327387,
      "tokens_seen": 1422655488
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002873520561685055,
      "loss": 2.8926,
      "theoretical_loss": 3.532598956529883,
      "tokens_seen": 1422721024
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028734202607823473,
      "loss": 2.8696,
      "theoretical_loss": 3.5325844635869337,
      "tokens_seen": 1422786560
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002873319959879639,
      "loss": 2.7378,
      "theoretical_loss": 3.5325699714984484,
      "tokens_seen": 1422852096
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002873219658976931,
      "loss": 2.7212,
      "theoretical_loss": 3.5325554802643375,
      "tokens_seen": 1422917632
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002873119358074223,
      "loss": 2.8135,
      "theoretical_loss": 3.5325409898845113,
      "tokens_seen": 1422983168
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028730190571715146,
      "loss": 2.8215,
      "theoretical_loss": 3.5325265003588804,
      "tokens_seen": 1423048704
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028729187562688064,
      "loss": 2.7063,
      "theoretical_loss": 3.5325120116873556,
      "tokens_seen": 1423114240
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002872818455366099,
      "loss": 2.8365,
      "theoretical_loss": 3.5324975238698455,
      "tokens_seen": 1423179776
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.000287271815446339,
      "loss": 3.0101,
      "theoretical_loss": 3.5324830369062625,
      "tokens_seen": 1423245312
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028726178535606824,
      "loss": 2.7546,
      "theoretical_loss": 3.5324685507965157,
      "tokens_seen": 1423310848
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028725175526579736,
      "loss": 2.7449,
      "theoretical_loss": 3.5324540655405157,
      "tokens_seen": 1423376384
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002872417251755266,
      "loss": 2.7711,
      "theoretical_loss": 3.5324395811381724,
      "tokens_seen": 1423441920
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002872316950852558,
      "loss": 3.0672,
      "theoretical_loss": 3.5324250975893974,
      "tokens_seen": 1423507456
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028722166499498496,
      "loss": 2.6498,
      "theoretical_loss": 3.5324106148941,
      "tokens_seen": 1423572992
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028721163490471414,
      "loss": 3.1087,
      "theoretical_loss": 3.5323961330521914,
      "tokens_seen": 1423638528
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002872016048144434,
      "loss": 2.6856,
      "theoretical_loss": 3.532381652063582,
      "tokens_seen": 1423704064
    },
    {
      "epoch": 0.43,
      "objective/train/advantage_avg": 0.4936656057834625,
      "objective/train/docs_used": 806909,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.9505128860473633,
      "objective/train/original_loss": 2.9505128860473633,
      "objective/train/theoretical_loss": 3.532367171928181,
      "objective/train/tokens_used": 1444229600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24379603564739227,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506058931350708,
      "objective/train/weighted_lm_loss": 3.0998449325561523,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 3.532367171928181,
      "tokens_seen": 1423769600
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002871915747241725,
      "loss": 2.7799,
      "theoretical_loss": 3.532367171928181,
      "tokens_seen": 1423769600
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028718154463390174,
      "loss": 2.691,
      "theoretical_loss": 3.5323526926459,
      "tokens_seen": 1423835136
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028717151454363087,
      "loss": 2.7444,
      "theoretical_loss": 3.53233821421665,
      "tokens_seen": 1423900672
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002871614844533601,
      "loss": 2.9997,
      "theoretical_loss": 3.5323237366403397,
      "tokens_seen": 1423966208
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002871514543630893,
      "loss": 2.9263,
      "theoretical_loss": 3.5323092599168815,
      "tokens_seen": 1424031744
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028714142427281846,
      "loss": 2.6974,
      "theoretical_loss": 3.5322947840461847,
      "tokens_seen": 1424097280
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028713139418254765,
      "loss": 2.7696,
      "theoretical_loss": 3.5322803090281605,
      "tokens_seen": 1424162816
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002871213640922768,
      "loss": 2.8618,
      "theoretical_loss": 3.5322658348627187,
      "tokens_seen": 1424228352
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.000287111334002006,
      "loss": 3.0001,
      "theoretical_loss": 3.5322513615497706,
      "tokens_seen": 1424293888
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028710130391173524,
      "loss": 3.1306,
      "theoretical_loss": 3.5322368890892264,
      "tokens_seen": 1424359424
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028709127382146437,
      "loss": 2.7652,
      "theoretical_loss": 3.5322224174809964,
      "tokens_seen": 1424424960
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002870812437311936,
      "loss": 2.8419,
      "theoretical_loss": 3.532207946724992,
      "tokens_seen": 1424490496
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028707121364092273,
      "loss": 2.8907,
      "theoretical_loss": 3.532193476821123,
      "tokens_seen": 1424556032
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028706118355065197,
      "loss": 2.8753,
      "theoretical_loss": 3.5321790077693005,
      "tokens_seen": 1424621568
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028705115346038115,
      "loss": 2.8549,
      "theoretical_loss": 3.5321645395694348,
      "tokens_seen": 1424687104
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028704112337011033,
      "loss": 3.0939,
      "theoretical_loss": 3.532150072221437,
      "tokens_seen": 1424752640
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002870310932798395,
      "loss": 2.7488,
      "theoretical_loss": 3.5321356057252173,
      "tokens_seen": 1424818176
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028702106318956875,
      "loss": 2.9109,
      "theoretical_loss": 3.5321211400806867,
      "tokens_seen": 1424883712
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002870110330992979,
      "loss": 3.033,
      "theoretical_loss": 3.5321066752877557,
      "tokens_seen": 1424949248
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002870010030090271,
      "loss": 2.8499,
      "theoretical_loss": 3.532092211346335,
      "tokens_seen": 1425014784
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028699097291875624,
      "loss": 2.9421,
      "theoretical_loss": 3.5320777482563352,
      "tokens_seen": 1425080320
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028698094282848547,
      "loss": 2.8682,
      "theoretical_loss": 3.5320632860176677,
      "tokens_seen": 1425145856
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028697091273821465,
      "loss": 2.9894,
      "theoretical_loss": 3.532048824630243,
      "tokens_seen": 1425211392
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028696088264794383,
      "loss": 3.0681,
      "theoretical_loss": 3.532034364093971,
      "tokens_seen": 1425276928
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.000286950852557673,
      "loss": 2.6172,
      "theoretical_loss": 3.532019904408763,
      "tokens_seen": 1425342464
    },
    {
      "epoch": 0.43,
      "objective/train/advantage_avg": 0.4945722818374634,
      "objective/train/docs_used": 808391,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.5650441646575928,
      "objective/train/original_loss": 2.565044403076172,
      "objective/train/theoretical_loss": 3.5320054455745304,
      "objective/train/tokens_used": 1445868000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2446717768907547,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507010221481323,
      "objective/train/weighted_lm_loss": 2.695240020751953,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0417574644088745,
      "theoretical_loss": 3.5320054455745304,
      "tokens_seen": 1425408000
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002869408224674022,
      "loss": 2.5549,
      "theoretical_loss": 3.5320054455745304,
      "tokens_seen": 1425408000
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002869307923771314,
      "loss": 3.0453,
      "theoretical_loss": 3.5319909875911835,
      "tokens_seen": 1425473536
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002869207622868606,
      "loss": 2.979,
      "theoretical_loss": 3.5319765304586324,
      "tokens_seen": 1425539072
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028691073219658974,
      "loss": 2.6336,
      "theoretical_loss": 3.531962074176789,
      "tokens_seen": 1425604608
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.000286900702106319,
      "loss": 2.7809,
      "theoretical_loss": 3.531947618745564,
      "tokens_seen": 1425670144
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002868906720160481,
      "loss": 2.7301,
      "theoretical_loss": 3.5319331641648675,
      "tokens_seen": 1425735680
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028688064192577734,
      "loss": 3.0202,
      "theoretical_loss": 3.531918710434611,
      "tokens_seen": 1425801216
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002868706118355065,
      "loss": 2.6295,
      "theoretical_loss": 3.5319042575547055,
      "tokens_seen": 1425866752
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002868605817452357,
      "loss": 2.8036,
      "theoretical_loss": 3.5318898055250614,
      "tokens_seen": 1425932288
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002868505516549649,
      "loss": 2.9755,
      "theoretical_loss": 3.5318753543455896,
      "tokens_seen": 1425997824
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002868405215646941,
      "loss": 2.7761,
      "theoretical_loss": 3.5318609040162015,
      "tokens_seen": 1426063360
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028683049147442324,
      "loss": 2.8901,
      "theoretical_loss": 3.531846454536808,
      "tokens_seen": 1426128896
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002868204613841525,
      "loss": 2.6547,
      "theoretical_loss": 3.5318320059073196,
      "tokens_seen": 1426194432
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002868104312938816,
      "loss": 2.5297,
      "theoretical_loss": 3.5318175581276474,
      "tokens_seen": 1426259968
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028680040120361084,
      "loss": 2.9084,
      "theoretical_loss": 3.5318031111977026,
      "tokens_seen": 1426325504
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028679037111334,
      "loss": 2.7201,
      "theoretical_loss": 3.5317886651173955,
      "tokens_seen": 1426391040
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002867803410230692,
      "loss": 2.576,
      "theoretical_loss": 3.531774219886638,
      "tokens_seen": 1426456576
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028677031093279844,
      "loss": 3.0321,
      "theoretical_loss": 3.5317597755053414,
      "tokens_seen": 1426522112
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028676028084252756,
      "loss": 2.7375,
      "theoretical_loss": 3.5317453319734153,
      "tokens_seen": 1426587648
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002867502507522568,
      "loss": 2.8971,
      "theoretical_loss": 3.5317308892907717,
      "tokens_seen": 1426653184
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.000286740220661986,
      "loss": 2.6494,
      "theoretical_loss": 3.5317164474573217,
      "tokens_seen": 1426718720
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028673019057171516,
      "loss": 2.9237,
      "theoretical_loss": 3.531702006472976,
      "tokens_seen": 1426784256
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028672016048144434,
      "loss": 2.8004,
      "theoretical_loss": 3.5316875663376455,
      "tokens_seen": 1426849792
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002867101303911736,
      "loss": 2.925,
      "theoretical_loss": 3.531673127051242,
      "tokens_seen": 1426915328
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002867001003009027,
      "loss": 2.9376,
      "theoretical_loss": 3.5316586886136756,
      "tokens_seen": 1426980864
    },
    {
      "epoch": 0.43,
      "objective/train/advantage_avg": 0.49427106976509094,
      "objective/train/docs_used": 809012,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.6075987815856934,
      "objective/train/original_loss": 2.6075990200042725,
      "objective/train/theoretical_loss": 3.5316442510248587,
      "objective/train/tokens_used": 1447506400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24437735974788666,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506693124771118,
      "objective/train/weighted_lm_loss": 2.739879608154297,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0253151655197144,
      "theoretical_loss": 3.5316442510248587,
      "tokens_seen": 1427046400
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028669007021063194,
      "loss": 2.7842,
      "theoretical_loss": 3.5316442510248587,
      "tokens_seen": 1427046400
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028668004012036107,
      "loss": 3.0684,
      "theoretical_loss": 3.5316298142847016,
      "tokens_seen": 1427111936
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002866700100300903,
      "loss": 2.5975,
      "theoretical_loss": 3.5316153783931155,
      "tokens_seen": 1427177472
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002866599799398195,
      "loss": 3.103,
      "theoretical_loss": 3.5316009433500115,
      "tokens_seen": 1427243008
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028664994984954866,
      "loss": 3.0297,
      "theoretical_loss": 3.531586509155301,
      "tokens_seen": 1427308544
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028663991975927785,
      "loss": 3.0795,
      "theoretical_loss": 3.5315720758088953,
      "tokens_seen": 1427374080
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.000286629889669007,
      "loss": 2.9217,
      "theoretical_loss": 3.5315576433107054,
      "tokens_seen": 1427439616
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002866198595787362,
      "loss": 2.8342,
      "theoretical_loss": 3.5315432116606424,
      "tokens_seen": 1427505152
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028660982948846544,
      "loss": 2.7649,
      "theoretical_loss": 3.5315287808586175,
      "tokens_seen": 1427570688
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028659979939819457,
      "loss": 2.8206,
      "theoretical_loss": 3.5315143509045424,
      "tokens_seen": 1427636224
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002865897693079238,
      "loss": 2.9435,
      "theoretical_loss": 3.5314999217983285,
      "tokens_seen": 1427701760
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028657973921765293,
      "loss": 2.7619,
      "theoretical_loss": 3.531485493539886,
      "tokens_seen": 1427767296
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028656970912738217,
      "loss": 2.8981,
      "theoretical_loss": 3.531471066129127,
      "tokens_seen": 1427832832
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028655967903711135,
      "loss": 2.725,
      "theoretical_loss": 3.5314566395659623,
      "tokens_seen": 1427898368
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028654964894684053,
      "loss": 2.6316,
      "theoretical_loss": 3.5314422138503034,
      "tokens_seen": 1427963904
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002865396188565697,
      "loss": 2.7925,
      "theoretical_loss": 3.5314277889820618,
      "tokens_seen": 1428029440
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028652958876629895,
      "loss": 2.6558,
      "theoretical_loss": 3.531413364961149,
      "tokens_seen": 1428094976
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002865195586760281,
      "loss": 2.688,
      "theoretical_loss": 3.5313989417874763,
      "tokens_seen": 1428160512
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002865095285857573,
      "loss": 2.7153,
      "theoretical_loss": 3.5313845194609543,
      "tokens_seen": 1428226048
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028649949849548644,
      "loss": 2.7998,
      "theoretical_loss": 3.5313700979814957,
      "tokens_seen": 1428291584
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028648946840521567,
      "loss": 2.8824,
      "theoretical_loss": 3.5313556773490102,
      "tokens_seen": 1428357120
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028647943831494485,
      "loss": 2.9002,
      "theoretical_loss": 3.531341257563411,
      "tokens_seen": 1428422656
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028646940822467403,
      "loss": 2.7987,
      "theoretical_loss": 3.531326838624608,
      "tokens_seen": 1428488192
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002864593781344032,
      "loss": 2.9247,
      "theoretical_loss": 3.5313124205325135,
      "tokens_seen": 1428553728
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002864493480441324,
      "loss": 2.9784,
      "theoretical_loss": 3.5312980032870387,
      "tokens_seen": 1428619264
    },
    {
      "epoch": 0.43,
      "objective/train/advantage_avg": 0.4950830042362213,
      "objective/train/docs_used": 809683,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.827895164489746,
      "objective/train/original_loss": 2.827894926071167,
      "objective/train/theoretical_loss": 3.5312835868880947,
      "objective/train/tokens_used": 1449144800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2451726794242859,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507546663284302,
      "objective/train/weighted_lm_loss": 2.9713404178619385,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0396549701690674,
      "theoretical_loss": 3.5312835868880947,
      "tokens_seen": 1428684800
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002864393179538616,
      "loss": 2.6793,
      "theoretical_loss": 3.5312835868880947,
      "tokens_seen": 1428684800
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002864292878635908,
      "loss": 2.804,
      "theoretical_loss": 3.5312691713355937,
      "tokens_seen": 1428750336
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028641925777331994,
      "loss": 2.8312,
      "theoretical_loss": 3.5312547566294463,
      "tokens_seen": 1428815872
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002864092276830492,
      "loss": 2.8839,
      "theoretical_loss": 3.531240342769565,
      "tokens_seen": 1428881408
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002863991975927783,
      "loss": 2.7872,
      "theoretical_loss": 3.531225929755861,
      "tokens_seen": 1428946944
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028638916750250754,
      "loss": 2.8413,
      "theoretical_loss": 3.531211517588245,
      "tokens_seen": 1429012480
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002863791374122367,
      "loss": 2.9353,
      "theoretical_loss": 3.5311971062666294,
      "tokens_seen": 1429078016
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002863691073219659,
      "loss": 2.8173,
      "theoretical_loss": 3.5311826957909256,
      "tokens_seen": 1429143552
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002863590772316951,
      "loss": 2.8835,
      "theoretical_loss": 3.5311682861610447,
      "tokens_seen": 1429209088
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002863490471414243,
      "loss": 2.5228,
      "theoretical_loss": 3.531153877376899,
      "tokens_seen": 1429274624
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028633901705115344,
      "loss": 2.9068,
      "theoretical_loss": 3.5311394694383997,
      "tokens_seen": 1429340160
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002863289869608827,
      "loss": 2.8255,
      "theoretical_loss": 3.531125062345458,
      "tokens_seen": 1429405696
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002863189568706118,
      "loss": 2.7796,
      "theoretical_loss": 3.531110656097986,
      "tokens_seen": 1429471232
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028630892678034104,
      "loss": 2.777,
      "theoretical_loss": 3.531096250695896,
      "tokens_seen": 1429536768
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002862988966900702,
      "loss": 2.8368,
      "theoretical_loss": 3.531081846139098,
      "tokens_seen": 1429602304
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002862888665997994,
      "loss": 2.8528,
      "theoretical_loss": 3.5310674424275046,
      "tokens_seen": 1429667840
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002862788365095286,
      "loss": 2.6287,
      "theoretical_loss": 3.5310530395610273,
      "tokens_seen": 1429733376
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028626880641925776,
      "loss": 2.9493,
      "theoretical_loss": 3.531038637539579,
      "tokens_seen": 1429798912
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028625877632898694,
      "loss": 2.9387,
      "theoretical_loss": 3.531024236363069,
      "tokens_seen": 1429864448
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002862487462387162,
      "loss": 2.8925,
      "theoretical_loss": 3.531009836031411,
      "tokens_seen": 1429929984
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002862387161484453,
      "loss": 2.6948,
      "theoretical_loss": 3.530995436544516,
      "tokens_seen": 1429995520
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028622868605817454,
      "loss": 2.7159,
      "theoretical_loss": 3.5309810379022952,
      "tokens_seen": 1430061056
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028621865596790367,
      "loss": 2.9278,
      "theoretical_loss": 3.530966640104661,
      "tokens_seen": 1430126592
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002862086258776329,
      "loss": 2.8841,
      "theoretical_loss": 3.5309522431515257,
      "tokens_seen": 1430192128
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002861985957873621,
      "loss": 2.5872,
      "theoretical_loss": 3.5309378470427997,
      "tokens_seen": 1430257664
    },
    {
      "epoch": 0.43,
      "objective/train/advantage_avg": 0.4953283965587616,
      "objective/train/docs_used": 810937,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.747922897338867,
      "objective/train/original_loss": 2.7479231357574463,
      "objective/train/theoretical_loss": 3.5309234517783956,
      "objective/train/tokens_used": 1450783200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24540278315544128,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507802963256836,
      "objective/train/weighted_lm_loss": 2.8873636722564697,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0396549701690674,
      "theoretical_loss": 3.5309234517783956,
      "tokens_seen": 1430323200
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028618856569709127,
      "loss": 2.8208,
      "theoretical_loss": 3.5309234517783956,
      "tokens_seen": 1430323200
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028617853560682045,
      "loss": 2.6908,
      "theoretical_loss": 3.530909057358225,
      "tokens_seen": 1430388736
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002861685055165497,
      "loss": 3.1918,
      "theoretical_loss": 3.5308946637822003,
      "tokens_seen": 1430454272
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002861584754262788,
      "loss": 2.8091,
      "theoretical_loss": 3.5308802710502327,
      "tokens_seen": 1430519808
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028614844533600805,
      "loss": 2.9081,
      "theoretical_loss": 3.530865879162234,
      "tokens_seen": 1430585344
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028613841524573717,
      "loss": 2.8284,
      "theoretical_loss": 3.530851488118116,
      "tokens_seen": 1430650880
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002861283851554664,
      "loss": 2.8842,
      "theoretical_loss": 3.530837097917791,
      "tokens_seen": 1430716416
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002861183550651956,
      "loss": 2.8951,
      "theoretical_loss": 3.530822708561171,
      "tokens_seen": 1430781952
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028610832497492477,
      "loss": 3.091,
      "theoretical_loss": 3.530808320048167,
      "tokens_seen": 1430847488
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028609829488465395,
      "loss": 3.0112,
      "theoretical_loss": 3.5307939323786917,
      "tokens_seen": 1430913024
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028608826479438313,
      "loss": 2.8713,
      "theoretical_loss": 3.5307795455526567,
      "tokens_seen": 1430978560
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002860782347041123,
      "loss": 2.7478,
      "theoretical_loss": 3.5307651595699743,
      "tokens_seen": 1431044096
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028606820461384155,
      "loss": 2.6907,
      "theoretical_loss": 3.5307507744305555,
      "tokens_seen": 1431109632
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002860581745235707,
      "loss": 3.0659,
      "theoretical_loss": 3.5307363901343134,
      "tokens_seen": 1431175168
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002860481444332999,
      "loss": 2.6462,
      "theoretical_loss": 3.5307220066811595,
      "tokens_seen": 1431240704
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002860381143430291,
      "loss": 2.8529,
      "theoretical_loss": 3.530707624071006,
      "tokens_seen": 1431306240
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002860280842527583,
      "loss": 2.8199,
      "theoretical_loss": 3.530693242303764,
      "tokens_seen": 1431371776
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002860180541624875,
      "loss": 2.8899,
      "theoretical_loss": 3.5306788613793465,
      "tokens_seen": 1431437312
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028600802407221664,
      "loss": 2.9733,
      "theoretical_loss": 3.5306644812976655,
      "tokens_seen": 1431502848
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028599799398194587,
      "loss": 2.8903,
      "theoretical_loss": 3.5306501020586323,
      "tokens_seen": 1431568384
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028598796389167505,
      "loss": 2.7458,
      "theoretical_loss": 3.5306357236621597,
      "tokens_seen": 1431633920
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028597793380140423,
      "loss": 2.5951,
      "theoretical_loss": 3.530621346108159,
      "tokens_seen": 1431699456
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002859679037111334,
      "loss": 2.6421,
      "theoretical_loss": 3.5306069693965427,
      "tokens_seen": 1431764992
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002859578736208626,
      "loss": 2.9456,
      "theoretical_loss": 3.5305925935272233,
      "tokens_seen": 1431830528
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002859478435305918,
      "loss": 2.9321,
      "theoretical_loss": 3.5305782185001124,
      "tokens_seen": 1431896064
    },
    {
      "epoch": 0.43,
      "objective/train/advantage_avg": 0.49360111355781555,
      "objective/train/docs_used": 811574,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.7183427810668945,
      "objective/train/original_loss": 2.718343496322632,
      "objective/train/theoretical_loss": 3.530563844315122,
      "objective/train/tokens_used": 1452421600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24373900890350342,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050599217414856,
      "objective/train/weighted_lm_loss": 2.8558549880981445,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0363596677780151,
      "theoretical_loss": 3.530563844315122,
      "tokens_seen": 1431961600
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.000285937813440321,
      "loss": 2.743,
      "theoretical_loss": 3.530563844315122,
      "tokens_seen": 1431961600
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028592778335005014,
      "loss": 2.8744,
      "theoretical_loss": 3.530549470972165,
      "tokens_seen": 1432027136
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002859177532597794,
      "loss": 2.9221,
      "theoretical_loss": 3.5305350984711525,
      "tokens_seen": 1432092672
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002859077231695085,
      "loss": 2.8085,
      "theoretical_loss": 3.5305207268119974,
      "tokens_seen": 1432158208
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028589769307923774,
      "loss": 2.7486,
      "theoretical_loss": 3.5305063559946115,
      "tokens_seen": 1432223744
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002858876629889669,
      "loss": 3.0416,
      "theoretical_loss": 3.530491986018907,
      "tokens_seen": 1432289280
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002858776328986961,
      "loss": 2.9859,
      "theoretical_loss": 3.5304776168847964,
      "tokens_seen": 1432354816
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002858676028084253,
      "loss": 2.8928,
      "theoretical_loss": 3.5304632485921914,
      "tokens_seen": 1432420352
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002858575727181545,
      "loss": 2.8496,
      "theoretical_loss": 3.530448881141005,
      "tokens_seen": 1432485888
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028584754262788364,
      "loss": 2.7401,
      "theoretical_loss": 3.530434514531149,
      "tokens_seen": 1432551424
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002858375125376129,
      "loss": 2.776,
      "theoretical_loss": 3.530420148762535,
      "tokens_seen": 1432616960
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.000285827482447342,
      "loss": 2.6856,
      "theoretical_loss": 3.530405783835076,
      "tokens_seen": 1432682496
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028581745235707124,
      "loss": 2.9258,
      "theoretical_loss": 3.5303914197486845,
      "tokens_seen": 1432748032
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002858074222668004,
      "loss": 2.7131,
      "theoretical_loss": 3.5303770565032724,
      "tokens_seen": 1432813568
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002857973921765296,
      "loss": 2.7746,
      "theoretical_loss": 3.530362694098752,
      "tokens_seen": 1432879104
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002857873620862588,
      "loss": 2.6863,
      "theoretical_loss": 3.530348332535035,
      "tokens_seen": 1432944640
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028577733199598796,
      "loss": 2.5418,
      "theoretical_loss": 3.530333971812035,
      "tokens_seen": 1433010176
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028576730190571714,
      "loss": 2.6939,
      "theoretical_loss": 3.530319611929664,
      "tokens_seen": 1433075712
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002857572718154464,
      "loss": 2.7908,
      "theoretical_loss": 3.5303052528878336,
      "tokens_seen": 1433141248
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002857472417251755,
      "loss": 2.8852,
      "theoretical_loss": 3.530290894686457,
      "tokens_seen": 1433206784
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028573721163490474,
      "loss": 2.9554,
      "theoretical_loss": 3.530276537325446,
      "tokens_seen": 1433272320
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028572718154463387,
      "loss": 2.6254,
      "theoretical_loss": 3.530262180804713,
      "tokens_seen": 1433337856
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002857171514543631,
      "loss": 2.8955,
      "theoretical_loss": 3.530247825124171,
      "tokens_seen": 1433403392
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002857071213640923,
      "loss": 2.7192,
      "theoretical_loss": 3.530233470283732,
      "tokens_seen": 1433468928
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028569709127382147,
      "loss": 2.6106,
      "theoretical_loss": 3.530219116283308,
      "tokens_seen": 1433534464
    },
    {
      "epoch": 0.43,
      "objective/train/advantage_avg": 0.4931759536266327,
      "objective/train/docs_used": 812859,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.85764479637146,
      "objective/train/original_loss": 2.857644557952881,
      "objective/train/theoretical_loss": 3.5302047631228124,
      "objective/train/tokens_used": 1454060000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2433094084262848,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505542755126953,
      "objective/train/weighted_lm_loss": 3.0020487308502197,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 3.5302047631228124,
      "tokens_seen": 1433600000
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028568706118355065,
      "loss": 2.9624,
      "theoretical_loss": 3.5302047631228124,
      "tokens_seen": 1433600000
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002856770310932799,
      "loss": 2.7869,
      "theoretical_loss": 3.530190410802157,
      "tokens_seen": 1433665536
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.000285667001003009,
      "loss": 2.6933,
      "theoretical_loss": 3.5301760593212537,
      "tokens_seen": 1433731072
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028565697091273825,
      "loss": 3.0092,
      "theoretical_loss": 3.5301617086800166,
      "tokens_seen": 1433796608
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028564694082246737,
      "loss": 2.7632,
      "theoretical_loss": 3.530147358878357,
      "tokens_seen": 1433862144
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002856369107321966,
      "loss": 2.8057,
      "theoretical_loss": 3.530133009916188,
      "tokens_seen": 1433927680
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002856268806419258,
      "loss": 2.6219,
      "theoretical_loss": 3.5301186617934217,
      "tokens_seen": 1433993216
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028561685055165497,
      "loss": 2.8463,
      "theoretical_loss": 3.5301043145099706,
      "tokens_seen": 1434058752
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028560682046138415,
      "loss": 2.9641,
      "theoretical_loss": 3.530089968065748,
      "tokens_seen": 1434124288
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028559679037111333,
      "loss": 2.8828,
      "theoretical_loss": 3.5300756224606653,
      "tokens_seen": 1434189824
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002855867602808425,
      "loss": 2.7658,
      "theoretical_loss": 3.530061277694636,
      "tokens_seen": 1434255360
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028557673019057175,
      "loss": 2.9267,
      "theoretical_loss": 3.530046933767572,
      "tokens_seen": 1434320896
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002855667001003009,
      "loss": 2.8877,
      "theoretical_loss": 3.5300325906793866,
      "tokens_seen": 1434386432
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002855566700100301,
      "loss": 2.8155,
      "theoretical_loss": 3.530018248429992,
      "tokens_seen": 1434451968
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002855466399197593,
      "loss": 2.6762,
      "theoretical_loss": 3.530003907019301,
      "tokens_seen": 1434517504
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002855366098294885,
      "loss": 2.8626,
      "theoretical_loss": 3.5299895664472256,
      "tokens_seen": 1434583040
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028552657973921765,
      "loss": 2.6247,
      "theoretical_loss": 3.5299752267136797,
      "tokens_seen": 1434648576
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028551654964894684,
      "loss": 2.9014,
      "theoretical_loss": 3.529960887818575,
      "tokens_seen": 1434714112
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.000285506519558676,
      "loss": 2.8529,
      "theoretical_loss": 3.5299465497618243,
      "tokens_seen": 1434779648
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028549648946840525,
      "loss": 2.6904,
      "theoretical_loss": 3.5299322125433408,
      "tokens_seen": 1434845184
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002854864593781344,
      "loss": 2.7529,
      "theoretical_loss": 3.5299178761630365,
      "tokens_seen": 1434910720
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002854764292878636,
      "loss": 2.8539,
      "theoretical_loss": 3.529903540620825,
      "tokens_seen": 1434976256
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028546639919759274,
      "loss": 2.8677,
      "theoretical_loss": 3.5298892059166183,
      "tokens_seen": 1435041792
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.000285456369107322,
      "loss": 3.0226,
      "theoretical_loss": 3.5298748720503292,
      "tokens_seen": 1435107328
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028544633901705116,
      "loss": 2.6923,
      "theoretical_loss": 3.5298605390218705,
      "tokens_seen": 1435172864
    },
    {
      "epoch": 0.43,
      "objective/train/advantage_avg": 0.4948815703392029,
      "objective/train/docs_used": 813625,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.7134852409362793,
      "objective/train/original_loss": 2.7134854793548584,
      "objective/train/theoretical_loss": 3.5298462068311554,
      "objective/train/tokens_used": 1455698400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24496346712112427,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507333278656006,
      "objective/train/weighted_lm_loss": 2.85127329826355,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0417574644088745,
      "theoretical_loss": 3.5298462068311554,
      "tokens_seen": 1435238400
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.00028543630892678034,
      "loss": 2.6482,
      "theoretical_loss": 3.5298462068311554,
      "tokens_seen": 1435238400
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002854262788365095,
      "loss": 2.7243,
      "theoretical_loss": 3.529831875478096,
      "tokens_seen": 1435303936
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002854162487462387,
      "loss": 2.966,
      "theoretical_loss": 3.5298175449626057,
      "tokens_seen": 1435369472
    },
    {
      "epoch": 0.43,
      "learning_rate": 0.0002854062186559679,
      "loss": 2.8795,
      "theoretical_loss": 3.529803215284597,
      "tokens_seen": 1435435008
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002853961885656971,
      "loss": 2.8778,
      "theoretical_loss": 3.529788886443983,
      "tokens_seen": 1435500544
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028538615847542624,
      "loss": 2.8711,
      "theoretical_loss": 3.5297745584406766,
      "tokens_seen": 1435566080
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002853761283851555,
      "loss": 2.6788,
      "theoretical_loss": 3.5297602312745897,
      "tokens_seen": 1435631616
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028536609829488466,
      "loss": 3.0467,
      "theoretical_loss": 3.5297459049456363,
      "tokens_seen": 1435697152
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028535606820461384,
      "loss": 2.9347,
      "theoretical_loss": 3.5297315794537285,
      "tokens_seen": 1435762688
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.000285346038114343,
      "loss": 2.8222,
      "theoretical_loss": 3.52971725479878,
      "tokens_seen": 1435828224
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002853360080240722,
      "loss": 2.6826,
      "theoretical_loss": 3.5297029309807026,
      "tokens_seen": 1435893760
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002853259779338014,
      "loss": 2.8182,
      "theoretical_loss": 3.5296886079994105,
      "tokens_seen": 1435959296
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002853159478435306,
      "loss": 2.8226,
      "theoretical_loss": 3.529674285854816,
      "tokens_seen": 1436024832
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028530591775325975,
      "loss": 2.7514,
      "theoretical_loss": 3.5296599645468314,
      "tokens_seen": 1436090368
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.000285295887662989,
      "loss": 2.7003,
      "theoretical_loss": 3.5296456440753707,
      "tokens_seen": 1436155904
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002852858575727181,
      "loss": 3.064,
      "theoretical_loss": 3.5296313244403463,
      "tokens_seen": 1436221440
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028527582748244735,
      "loss": 2.5833,
      "theoretical_loss": 3.529617005641671,
      "tokens_seen": 1436286976
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002852657973921766,
      "loss": 2.9315,
      "theoretical_loss": 3.529602687679259,
      "tokens_seen": 1436352512
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002852557673019057,
      "loss": 2.833,
      "theoretical_loss": 3.529588370553022,
      "tokens_seen": 1436418048
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028524573721163494,
      "loss": 2.6846,
      "theoretical_loss": 3.529574054262873,
      "tokens_seen": 1436483584
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028523570712136407,
      "loss": 2.7334,
      "theoretical_loss": 3.529559738808726,
      "tokens_seen": 1436549120
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002852256770310933,
      "loss": 2.7786,
      "theoretical_loss": 3.529545424190493,
      "tokens_seen": 1436614656
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002852156469408225,
      "loss": 2.508,
      "theoretical_loss": 3.5295311104080884,
      "tokens_seen": 1436680192
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028520561685055167,
      "loss": 2.5889,
      "theoretical_loss": 3.529516797461424,
      "tokens_seen": 1436745728
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028519558676028085,
      "loss": 2.7473,
      "theoretical_loss": 3.529502485350413,
      "tokens_seen": 1436811264
    },
    {
      "epoch": 0.44,
      "objective/train/advantage_avg": 0.49449554085731506,
      "objective/train/docs_used": 814744,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.00410532951355,
      "objective/train/original_loss": 3.00410532951355,
      "objective/train/theoretical_loss": 3.5294881740749693,
      "objective/train/tokens_used": 1457336800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.244594544172287,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506929159164429,
      "objective/train/weighted_lm_loss": 3.15651273727417,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0396549701690674,
      "theoretical_loss": 3.5294881740749693,
      "tokens_seen": 1436876800
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002851855566700101,
      "loss": 2.9241,
      "theoretical_loss": 3.5294881740749693,
      "tokens_seen": 1436876800
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002851755265797392,
      "loss": 2.7526,
      "theoretical_loss": 3.5294738636350056,
      "tokens_seen": 1436942336
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028516549648946845,
      "loss": 2.6831,
      "theoretical_loss": 3.5294595540304345,
      "tokens_seen": 1437007872
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028515546639919757,
      "loss": 2.8417,
      "theoretical_loss": 3.52944524526117,
      "tokens_seen": 1437073408
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002851454363089268,
      "loss": 2.8785,
      "theoretical_loss": 3.529430937327125,
      "tokens_seen": 1437138944
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.000285135406218656,
      "loss": 3.0147,
      "theoretical_loss": 3.529416630228212,
      "tokens_seen": 1437204480
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028512537612838517,
      "loss": 3.0242,
      "theoretical_loss": 3.5294023239643453,
      "tokens_seen": 1437270016
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028511534603811435,
      "loss": 2.9446,
      "theoretical_loss": 3.529388018535437,
      "tokens_seen": 1437335552
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028510531594784353,
      "loss": 2.6622,
      "theoretical_loss": 3.5293737139414008,
      "tokens_seen": 1437401088
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002850952858575727,
      "loss": 2.8538,
      "theoretical_loss": 3.52935941018215,
      "tokens_seen": 1437466624
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028508525576730195,
      "loss": 2.7561,
      "theoretical_loss": 3.5293451072575976,
      "tokens_seen": 1437532160
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002850752256770311,
      "loss": 2.862,
      "theoretical_loss": 3.5293308051676573,
      "tokens_seen": 1437597696
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002850651955867603,
      "loss": 2.835,
      "theoretical_loss": 3.529316503912242,
      "tokens_seen": 1437663232
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002850551654964895,
      "loss": 2.7926,
      "theoretical_loss": 3.529302203491265,
      "tokens_seen": 1437728768
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002850451354062187,
      "loss": 2.6846,
      "theoretical_loss": 3.529287903904639,
      "tokens_seen": 1437794304
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028503510531594785,
      "loss": 2.7664,
      "theoretical_loss": 3.529273605152279,
      "tokens_seen": 1437859840
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028502507522567704,
      "loss": 3.0462,
      "theoretical_loss": 3.529259307234096,
      "tokens_seen": 1437925376
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002850150451354062,
      "loss": 2.8025,
      "theoretical_loss": 3.5292450101500052,
      "tokens_seen": 1437990912
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028500501504513545,
      "loss": 2.9343,
      "theoretical_loss": 3.529230713899919,
      "tokens_seen": 1438056448
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002849949849548646,
      "loss": 2.6765,
      "theoretical_loss": 3.529216418483751,
      "tokens_seen": 1438121984
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002849849548645938,
      "loss": 2.7278,
      "theoretical_loss": 3.529202123901414,
      "tokens_seen": 1438187520
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028497492477432294,
      "loss": 2.6959,
      "theoretical_loss": 3.5291878301528232,
      "tokens_seen": 1438253056
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002849648946840522,
      "loss": 2.7319,
      "theoretical_loss": 3.52917353723789,
      "tokens_seen": 1438318592
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028495486459378136,
      "loss": 2.99,
      "theoretical_loss": 3.529159245156528,
      "tokens_seen": 1438384128
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028494483450351054,
      "loss": 2.8135,
      "theoretical_loss": 3.5291449539086512,
      "tokens_seen": 1438449664
    },
    {
      "epoch": 0.44,
      "objective/train/advantage_avg": 0.4933953881263733,
      "objective/train/docs_used": 815309,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.12150239944458,
      "objective/train/original_loss": 3.121502161026001,
      "objective/train/theoretical_loss": 3.5291306634941737,
      "objective/train/tokens_used": 1458975200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24353596568107605,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505774021148682,
      "objective/train/weighted_lm_loss": 3.279345989227295,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 3.5291306634941737,
      "tokens_seen": 1438515200
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002849348044132397,
      "loss": 2.7444,
      "theoretical_loss": 3.5291306634941737,
      "tokens_seen": 1438515200
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002849247743229689,
      "loss": 2.8719,
      "theoretical_loss": 3.5291163739130074,
      "tokens_seen": 1438580736
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002849147442326981,
      "loss": 2.8311,
      "theoretical_loss": 3.529102085165067,
      "tokens_seen": 1438646272
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002849047141424273,
      "loss": 2.9263,
      "theoretical_loss": 3.5290877972502646,
      "tokens_seen": 1438711808
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028489468405215644,
      "loss": 2.8394,
      "theoretical_loss": 3.529073510168516,
      "tokens_seen": 1438777344
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002848846539618857,
      "loss": 2.4567,
      "theoretical_loss": 3.529059223919732,
      "tokens_seen": 1438842880
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028487462387161486,
      "loss": 2.7416,
      "theoretical_loss": 3.5290449385038274,
      "tokens_seen": 1438908416
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028486459378134404,
      "loss": 2.722,
      "theoretical_loss": 3.529030653920716,
      "tokens_seen": 1438973952
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002848545636910732,
      "loss": 3.0184,
      "theoretical_loss": 3.529016370170311,
      "tokens_seen": 1439039488
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002848445336008024,
      "loss": 2.5099,
      "theoretical_loss": 3.5290020872525254,
      "tokens_seen": 1439105024
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002848345035105316,
      "loss": 2.6983,
      "theoretical_loss": 3.5289878051672736,
      "tokens_seen": 1439170560
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002848244734202608,
      "loss": 2.6455,
      "theoretical_loss": 3.528973523914469,
      "tokens_seen": 1439236096
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028481444332998995,
      "loss": 2.9222,
      "theoretical_loss": 3.5289592434940245,
      "tokens_seen": 1439301632
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002848044132397192,
      "loss": 2.8558,
      "theoretical_loss": 3.5289449639058543,
      "tokens_seen": 1439367168
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002847943831494483,
      "loss": 2.991,
      "theoretical_loss": 3.5289306851498723,
      "tokens_seen": 1439432704
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028478435305917755,
      "loss": 2.8832,
      "theoretical_loss": 3.5289164072259913,
      "tokens_seen": 1439498240
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002847743229689067,
      "loss": 2.8475,
      "theoretical_loss": 3.5289021301341252,
      "tokens_seen": 1439563776
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002847642928786359,
      "loss": 2.865,
      "theoretical_loss": 3.5288878538741884,
      "tokens_seen": 1439629312
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002847542627883651,
      "loss": 2.8425,
      "theoretical_loss": 3.5288735784460936,
      "tokens_seen": 1439694848
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028474423269809427,
      "loss": 2.9112,
      "theoretical_loss": 3.5288593038497544,
      "tokens_seen": 1439760384
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028473420260782345,
      "loss": 2.8244,
      "theoretical_loss": 3.528845030085085,
      "tokens_seen": 1439825920
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002847241725175527,
      "loss": 2.7543,
      "theoretical_loss": 3.528830757152,
      "tokens_seen": 1439891456
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002847141424272818,
      "loss": 2.907,
      "theoretical_loss": 3.528816485050411,
      "tokens_seen": 1439956992
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028470411233701105,
      "loss": 2.6885,
      "theoretical_loss": 3.528802213780233,
      "tokens_seen": 1440022528
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028469408224674023,
      "loss": 2.9021,
      "theoretical_loss": 3.5287879433413796,
      "tokens_seen": 1440088064
    },
    {
      "epoch": 0.44,
      "objective/train/advantage_avg": 0.49501198530197144,
      "objective/train/docs_used": 816515,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.1472928524017334,
      "objective/train/original_loss": 3.1472930908203125,
      "objective/train/theoretical_loss": 3.528773673733764,
      "objective/train/tokens_used": 1460613600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24509774148464203,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507471561431885,
      "objective/train/weighted_lm_loss": 3.3069818019866943,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0304546356201172,
      "theoretical_loss": 3.528773673733764,
      "tokens_seen": 1440153600
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002846840521564694,
      "loss": 2.7067,
      "theoretical_loss": 3.528773673733764,
      "tokens_seen": 1440153600
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002846740220661986,
      "loss": 2.8104,
      "theoretical_loss": 3.5287594049573014,
      "tokens_seen": 1440219136
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002846639919759278,
      "loss": 2.9182,
      "theoretical_loss": 3.528745137011904,
      "tokens_seen": 1440284672
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028465396188565695,
      "loss": 2.9995,
      "theoretical_loss": 3.528730869897486,
      "tokens_seen": 1440350208
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002846439317953862,
      "loss": 2.7926,
      "theoretical_loss": 3.5287166036139617,
      "tokens_seen": 1440415744
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002846339017051153,
      "loss": 2.6801,
      "theoretical_loss": 3.528702338161245,
      "tokens_seen": 1440481280
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028462387161484455,
      "loss": 2.9922,
      "theoretical_loss": 3.5286880735392483,
      "tokens_seen": 1440546816
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002846138415245737,
      "loss": 2.9083,
      "theoretical_loss": 3.528673809747887,
      "tokens_seen": 1440612352
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002846038114343029,
      "loss": 3.1946,
      "theoretical_loss": 3.5286595467870745,
      "tokens_seen": 1440677888
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002845937813440321,
      "loss": 2.8082,
      "theoretical_loss": 3.5286452846567244,
      "tokens_seen": 1440743424
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002845837512537613,
      "loss": 2.9467,
      "theoretical_loss": 3.528631023356751,
      "tokens_seen": 1440808960
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028457372116349046,
      "loss": 2.8661,
      "theoretical_loss": 3.5286167628870677,
      "tokens_seen": 1440874496
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002845636910732197,
      "loss": 3.0653,
      "theoretical_loss": 3.5286025032475883,
      "tokens_seen": 1440940032
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002845536609829488,
      "loss": 2.8275,
      "theoretical_loss": 3.5285882444382275,
      "tokens_seen": 1441005568
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028454363089267805,
      "loss": 2.9145,
      "theoretical_loss": 3.5285739864588983,
      "tokens_seen": 1441071104
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028453360080240724,
      "loss": 2.7945,
      "theoretical_loss": 3.5285597293095154,
      "tokens_seen": 1441136640
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002845235707121364,
      "loss": 2.8518,
      "theoretical_loss": 3.5285454729899923,
      "tokens_seen": 1441202176
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028451354062186565,
      "loss": 3.0215,
      "theoretical_loss": 3.5285312175002432,
      "tokens_seen": 1441267712
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002845035105315948,
      "loss": 2.8471,
      "theoretical_loss": 3.528516962840182,
      "tokens_seen": 1441333248
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.000284493480441324,
      "loss": 2.9181,
      "theoretical_loss": 3.5285027090097225,
      "tokens_seen": 1441398784
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028448345035105314,
      "loss": 2.6769,
      "theoretical_loss": 3.5284884560087795,
      "tokens_seen": 1441464320
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002844734202607824,
      "loss": 2.6797,
      "theoretical_loss": 3.5284742038372654,
      "tokens_seen": 1441529856
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028446339017051156,
      "loss": 2.9169,
      "theoretical_loss": 3.528459952495096,
      "tokens_seen": 1441595392
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028445336008024074,
      "loss": 2.81,
      "theoretical_loss": 3.528445701982184,
      "tokens_seen": 1441660928
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002844433299899699,
      "loss": 2.6924,
      "theoretical_loss": 3.528431452298445,
      "tokens_seen": 1441726464
    },
    {
      "epoch": 0.44,
      "objective/train/advantage_avg": 0.4940248131752014,
      "objective/train/docs_used": 817059,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.842756509780884,
      "objective/train/original_loss": 2.842756986618042,
      "objective/train/theoretical_loss": 3.528417203443791,
      "objective/train/tokens_used": 1462252000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24415801465511322,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506435632705688,
      "objective/train/weighted_lm_loss": 2.9867725372314453,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 3.528417203443791,
      "tokens_seen": 1441792000
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002844332998996991,
      "loss": 2.8344,
      "theoretical_loss": 3.528417203443791,
      "tokens_seen": 1441792000
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002844232698094283,
      "loss": 2.785,
      "theoretical_loss": 3.5284029554181373,
      "tokens_seen": 1441857536
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002844132397191575,
      "loss": 2.8808,
      "theoretical_loss": 3.5283887082213985,
      "tokens_seen": 1441923072
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028440320962888664,
      "loss": 2.7459,
      "theoretical_loss": 3.5283744618534874,
      "tokens_seen": 1441988608
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002843931795386159,
      "loss": 2.8982,
      "theoretical_loss": 3.5283602163143186,
      "tokens_seen": 1442054144
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028438314944834506,
      "loss": 2.7826,
      "theoretical_loss": 3.528345971603807,
      "tokens_seen": 1442119680
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028437311935807424,
      "loss": 2.775,
      "theoretical_loss": 3.5283317277218664,
      "tokens_seen": 1442185216
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002843630892678034,
      "loss": 2.7683,
      "theoretical_loss": 3.5283174846684098,
      "tokens_seen": 1442250752
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002843530591775326,
      "loss": 2.6177,
      "theoretical_loss": 3.5283032424433527,
      "tokens_seen": 1442316288
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002843430290872618,
      "loss": 2.6486,
      "theoretical_loss": 3.528289001046609,
      "tokens_seen": 1442381824
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.000284332998996991,
      "loss": 2.9514,
      "theoretical_loss": 3.5282747604780926,
      "tokens_seen": 1442447360
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028432296890672015,
      "loss": 2.9714,
      "theoretical_loss": 3.528260520737718,
      "tokens_seen": 1442512896
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002843129388164494,
      "loss": 2.6818,
      "theoretical_loss": 3.528246281825399,
      "tokens_seen": 1442578432
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002843029087261785,
      "loss": 2.7076,
      "theoretical_loss": 3.52823204374105,
      "tokens_seen": 1442643968
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028429287863590775,
      "loss": 2.8878,
      "theoretical_loss": 3.5282178064845855,
      "tokens_seen": 1442709504
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002842828485456369,
      "loss": 2.9013,
      "theoretical_loss": 3.52820357005592,
      "tokens_seen": 1442775040
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002842728184553661,
      "loss": 2.7259,
      "theoretical_loss": 3.528189334454967,
      "tokens_seen": 1442840576
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002842627883650953,
      "loss": 2.8588,
      "theoretical_loss": 3.528175099681641,
      "tokens_seen": 1442906112
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028425275827482447,
      "loss": 2.9041,
      "theoretical_loss": 3.528160865735857,
      "tokens_seen": 1442971648
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028424272818455365,
      "loss": 2.8445,
      "theoretical_loss": 3.528146632617528,
      "tokens_seen": 1443037184
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002842326980942829,
      "loss": 2.8544,
      "theoretical_loss": 3.5281324003265695,
      "tokens_seen": 1443102720
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.000284222668004012,
      "loss": 2.7595,
      "theoretical_loss": 3.5281181688628953,
      "tokens_seen": 1443168256
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028421263791374125,
      "loss": 2.7282,
      "theoretical_loss": 3.5281039382264203,
      "tokens_seen": 1443233792
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028420260782347043,
      "loss": 2.8768,
      "theoretical_loss": 3.528089708417058,
      "tokens_seen": 1443299328
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002841925777331996,
      "loss": 2.7702,
      "theoretical_loss": 3.5280754794347233,
      "tokens_seen": 1443364864
    },
    {
      "epoch": 0.44,
      "objective/train/advantage_avg": 0.4948015809059143,
      "objective/train/docs_used": 818508,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.6497204303741455,
      "objective/train/original_loss": 2.6497206687927246,
      "objective/train/theoretical_loss": 3.528061251279331,
      "objective/train/tokens_used": 1463890400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24493205547332764,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507252216339111,
      "objective/train/weighted_lm_loss": 2.7840847969055176,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0408108234405518,
      "theoretical_loss": 3.528061251279331,
      "tokens_seen": 1443430400
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002841825476429288,
      "loss": 2.8684,
      "theoretical_loss": 3.528061251279331,
      "tokens_seen": 1443430400
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.000284172517552658,
      "loss": 2.7873,
      "theoretical_loss": 3.528047023950794,
      "tokens_seen": 1443495936
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028416248746238715,
      "loss": 2.812,
      "theoretical_loss": 3.528032797449028,
      "tokens_seen": 1443561472
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002841524573721164,
      "loss": 2.7545,
      "theoretical_loss": 3.5280185717739476,
      "tokens_seen": 1443627008
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002841424272818455,
      "loss": 2.8277,
      "theoretical_loss": 3.5280043469254663,
      "tokens_seen": 1443692544
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028413239719157475,
      "loss": 2.8455,
      "theoretical_loss": 3.5279901229034993,
      "tokens_seen": 1443758080
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002841223671013039,
      "loss": 2.8584,
      "theoretical_loss": 3.5279758997079607,
      "tokens_seen": 1443823616
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002841123370110331,
      "loss": 2.9151,
      "theoretical_loss": 3.527961677338765,
      "tokens_seen": 1443889152
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002841023069207623,
      "loss": 2.9019,
      "theoretical_loss": 3.527947455795827,
      "tokens_seen": 1443954688
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002840922768304915,
      "loss": 2.8234,
      "theoretical_loss": 3.52793323507906,
      "tokens_seen": 1444020224
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028408224674022066,
      "loss": 2.7011,
      "theoretical_loss": 3.5279190151883806,
      "tokens_seen": 1444085760
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002840722166499499,
      "loss": 2.9455,
      "theoretical_loss": 3.5279047961237016,
      "tokens_seen": 1444151296
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.000284062186559679,
      "loss": 2.8851,
      "theoretical_loss": 3.5278905778849383,
      "tokens_seen": 1444216832
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028405215646940825,
      "loss": 2.6714,
      "theoretical_loss": 3.527876360472005,
      "tokens_seen": 1444282368
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002840421263791374,
      "loss": 2.81,
      "theoretical_loss": 3.5278621438848163,
      "tokens_seen": 1444347904
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002840320962888666,
      "loss": 2.7938,
      "theoretical_loss": 3.527847928123287,
      "tokens_seen": 1444413440
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002840220661985958,
      "loss": 2.8151,
      "theoretical_loss": 3.5278337131873307,
      "tokens_seen": 1444478976
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.000284012036108325,
      "loss": 2.6459,
      "theoretical_loss": 3.5278194990768634,
      "tokens_seen": 1444544512
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028400200601805416,
      "loss": 3.0418,
      "theoretical_loss": 3.5278052857917994,
      "tokens_seen": 1444610048
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028399197592778334,
      "loss": 2.8033,
      "theoretical_loss": 3.527791073332053,
      "tokens_seen": 1444675584
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002839819458375125,
      "loss": 2.7546,
      "theoretical_loss": 3.5277768616975385,
      "tokens_seen": 1444741120
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028397191574724176,
      "loss": 3.0555,
      "theoretical_loss": 3.527762650888171,
      "tokens_seen": 1444806656
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002839618856569709,
      "loss": 2.747,
      "theoretical_loss": 3.5277484409038653,
      "tokens_seen": 1444872192
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002839518555667001,
      "loss": 2.8689,
      "theoretical_loss": 3.5277342317445353,
      "tokens_seen": 1444937728
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028394182547642925,
      "loss": 2.7576,
      "theoretical_loss": 3.5277200234100965,
      "tokens_seen": 1445003264
    },
    {
      "epoch": 0.44,
      "objective/train/advantage_avg": 0.49384886026382446,
      "objective/train/docs_used": 819330,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.7524194717407227,
      "objective/train/original_loss": 2.7524194717407227,
      "objective/train/theoretical_loss": 3.5277058159004637,
      "objective/train/tokens_used": 1465528800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2439562976360321,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050624966621399,
      "objective/train/weighted_lm_loss": 2.8915939331054688,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0440459251403809,
      "theoretical_loss": 3.5277058159004637,
      "tokens_seen": 1445068800
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002839317953861585,
      "loss": 2.8181,
      "theoretical_loss": 3.5277058159004637,
      "tokens_seen": 1445068800
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028392176529588766,
      "loss": 2.9124,
      "theoretical_loss": 3.5276916092155512,
      "tokens_seen": 1445134336
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028391173520561684,
      "loss": 2.8346,
      "theoretical_loss": 3.527677403355274,
      "tokens_seen": 1445199872
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.000283901705115346,
      "loss": 3.0053,
      "theoretical_loss": 3.527663198319546,
      "tokens_seen": 1445265408
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028389167502507526,
      "loss": 2.7769,
      "theoretical_loss": 3.527648994108283,
      "tokens_seen": 1445330944
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002838816449348044,
      "loss": 2.9778,
      "theoretical_loss": 3.5276347907213994,
      "tokens_seen": 1445396480
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002838716148445336,
      "loss": 2.8124,
      "theoretical_loss": 3.5276205881588103,
      "tokens_seen": 1445462016
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028386158475426275,
      "loss": 2.704,
      "theoretical_loss": 3.5276063864204295,
      "tokens_seen": 1445527552
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.000283851554663992,
      "loss": 3.0821,
      "theoretical_loss": 3.527592185506173,
      "tokens_seen": 1445593088
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028384152457372117,
      "loss": 2.9642,
      "theoretical_loss": 3.527577985415955,
      "tokens_seen": 1445658624
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028383149448345035,
      "loss": 2.7358,
      "theoretical_loss": 3.52756378614969,
      "tokens_seen": 1445724160
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028382146439317953,
      "loss": 2.7827,
      "theoretical_loss": 3.5275495877072935,
      "tokens_seen": 1445789696
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002838114343029087,
      "loss": 2.8388,
      "theoretical_loss": 3.52753539008868,
      "tokens_seen": 1445855232
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002838014042126379,
      "loss": 2.6901,
      "theoretical_loss": 3.5275211932937647,
      "tokens_seen": 1445920768
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002837913741223671,
      "loss": 2.8559,
      "theoretical_loss": 3.5275069973224618,
      "tokens_seen": 1445986304
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002837813440320963,
      "loss": 2.8092,
      "theoretical_loss": 3.527492802174687,
      "tokens_seen": 1446051840
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002837713139418255,
      "loss": 2.8761,
      "theoretical_loss": 3.5274786078503553,
      "tokens_seen": 1446117376
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028376128385155467,
      "loss": 2.8823,
      "theoretical_loss": 3.5274644143493807,
      "tokens_seen": 1446182912
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028375125376128385,
      "loss": 2.6396,
      "theoretical_loss": 3.5274502216716783,
      "tokens_seen": 1446248448
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002837412236710131,
      "loss": 2.9006,
      "theoretical_loss": 3.5274360298171636,
      "tokens_seen": 1446313984
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002837311935807422,
      "loss": 2.8022,
      "theoretical_loss": 3.5274218387857514,
      "tokens_seen": 1446379520
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028372116349047145,
      "loss": 2.9852,
      "theoretical_loss": 3.5274076485773564,
      "tokens_seen": 1446445056
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028371113340020063,
      "loss": 2.8084,
      "theoretical_loss": 3.5273934591918943,
      "tokens_seen": 1446510592
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002837011033099298,
      "loss": 2.8067,
      "theoretical_loss": 3.527379270629279,
      "tokens_seen": 1446576128
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.000283691073219659,
      "loss": 2.8198,
      "theoretical_loss": 3.527365082889426,
      "tokens_seen": 1446641664
    },
    {
      "epoch": 0.44,
      "objective/train/advantage_avg": 0.4948766231536865,
      "objective/train/docs_used": 820636,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.583678722381592,
      "objective/train/original_loss": 2.5836784839630127,
      "objective/train/theoretical_loss": 3.5273508959722504,
      "objective/train/tokens_used": 1467167200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24495743215084076,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507328510284424,
      "objective/train/weighted_lm_loss": 2.7144978046417236,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0396549701690674,
      "theoretical_loss": 3.5273508959722504,
      "tokens_seen": 1446707200
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002836810431293882,
      "loss": 2.9045,
      "theoretical_loss": 3.5273508959722504,
      "tokens_seen": 1446707200
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028367101303911735,
      "loss": 2.7167,
      "theoretical_loss": 3.527336709877668,
      "tokens_seen": 1446772736
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002836609829488466,
      "loss": 3.0308,
      "theoretical_loss": 3.527322524605592,
      "tokens_seen": 1446838272
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002836509528585757,
      "loss": 2.8738,
      "theoretical_loss": 3.527308340155939,
      "tokens_seen": 1446903808
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028364092276830495,
      "loss": 2.8783,
      "theoretical_loss": 3.5272941565286238,
      "tokens_seen": 1446969344
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002836308926780341,
      "loss": 2.7898,
      "theoretical_loss": 3.527279973723561,
      "tokens_seen": 1447034880
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002836208625877633,
      "loss": 2.5524,
      "theoretical_loss": 3.527265791740666,
      "tokens_seen": 1447100416
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002836108324974925,
      "loss": 2.9396,
      "theoretical_loss": 3.527251610579854,
      "tokens_seen": 1447165952
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002836008024072217,
      "loss": 2.793,
      "theoretical_loss": 3.5272374302410396,
      "tokens_seen": 1447231488
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028359077231695086,
      "loss": 2.8119,
      "theoretical_loss": 3.5272232507241386,
      "tokens_seen": 1447297024
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002835807422266801,
      "loss": 2.7619,
      "theoretical_loss": 3.527209072029066,
      "tokens_seen": 1447362560
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002835707121364092,
      "loss": 2.7983,
      "theoretical_loss": 3.5271948941557367,
      "tokens_seen": 1447428096
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028356068204613846,
      "loss": 2.877,
      "theoretical_loss": 3.527180717104066,
      "tokens_seen": 1447493632
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002835506519558676,
      "loss": 2.8067,
      "theoretical_loss": 3.527166540873969,
      "tokens_seen": 1447559168
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002835406218655968,
      "loss": 2.9761,
      "theoretical_loss": 3.527152365465361,
      "tokens_seen": 1447624704
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.000283530591775326,
      "loss": 2.8787,
      "theoretical_loss": 3.5271381908781576,
      "tokens_seen": 1447690240
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002835205616850552,
      "loss": 2.9605,
      "theoretical_loss": 3.527124017112273,
      "tokens_seen": 1447755776
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028351053159478436,
      "loss": 2.7471,
      "theoretical_loss": 3.527109844167623,
      "tokens_seen": 1447821312
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028350050150451354,
      "loss": 2.9258,
      "theoretical_loss": 3.5270956720441236,
      "tokens_seen": 1447886848
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002834904714142427,
      "loss": 2.8209,
      "theoretical_loss": 3.527081500741689,
      "tokens_seen": 1447952384
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028348044132397196,
      "loss": 2.8052,
      "theoretical_loss": 3.527067330260235,
      "tokens_seen": 1448017920
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002834704112337011,
      "loss": 2.8429,
      "theoretical_loss": 3.5270531605996767,
      "tokens_seen": 1448083456
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002834603811434303,
      "loss": 2.7941,
      "theoretical_loss": 3.5270389917599294,
      "tokens_seen": 1448148992
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028345035105315945,
      "loss": 2.8632,
      "theoretical_loss": 3.5270248237409083,
      "tokens_seen": 1448214528
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002834403209628887,
      "loss": 2.9112,
      "theoretical_loss": 3.5270106565425285,
      "tokens_seen": 1448280064
    },
    {
      "epoch": 0.44,
      "objective/train/advantage_avg": 0.4928238093852997,
      "objective/train/docs_used": 821297,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.902559757232666,
      "objective/train/original_loss": 2.902559995651245,
      "objective/train/theoretical_loss": 3.5269964901647066,
      "objective/train/tokens_used": 1468805600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24301569163799286,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505177974700928,
      "objective/train/weighted_lm_loss": 3.049269676208496,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0363596677780151,
      "theoretical_loss": 3.5269964901647066,
      "tokens_seen": 1448345600
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028343029087261786,
      "loss": 2.8267,
      "theoretical_loss": 3.5269964901647066,
      "tokens_seen": 1448345600
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028342026078234705,
      "loss": 2.7434,
      "theoretical_loss": 3.5269823246073564,
      "tokens_seen": 1448411136
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002834102306920762,
      "loss": 2.9648,
      "theoretical_loss": 3.5269681598703935,
      "tokens_seen": 1448476672
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028340020060180546,
      "loss": 2.9248,
      "theoretical_loss": 3.5269539959537344,
      "tokens_seen": 1448542208
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002833901705115346,
      "loss": 2.7089,
      "theoretical_loss": 3.526939832857294,
      "tokens_seen": 1448607744
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002833801404212638,
      "loss": 2.845,
      "theoretical_loss": 3.5269256705809866,
      "tokens_seen": 1448673280
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028337011033099295,
      "loss": 2.6684,
      "theoretical_loss": 3.5269115091247287,
      "tokens_seen": 1448738816
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002833600802407222,
      "loss": 2.8164,
      "theoretical_loss": 3.5268973484884354,
      "tokens_seen": 1448804352
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028335005015045137,
      "loss": 2.6736,
      "theoretical_loss": 3.5268831886720227,
      "tokens_seen": 1448869888
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028334002006018055,
      "loss": 2.7766,
      "theoretical_loss": 3.526869029675405,
      "tokens_seen": 1448935424
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028332998996990973,
      "loss": 2.8608,
      "theoretical_loss": 3.5268548714984984,
      "tokens_seen": 1449000960
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002833199598796389,
      "loss": 2.6551,
      "theoretical_loss": 3.5268407141412186,
      "tokens_seen": 1449066496
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002833099297893681,
      "loss": 2.9252,
      "theoretical_loss": 3.5268265576034805,
      "tokens_seen": 1449132032
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002832998996990973,
      "loss": 3.0203,
      "theoretical_loss": 3.5268124018852003,
      "tokens_seen": 1449197568
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028328986960882645,
      "loss": 3.0891,
      "theoretical_loss": 3.526798246986293,
      "tokens_seen": 1449263104
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002832798395185557,
      "loss": 2.7325,
      "theoretical_loss": 3.5267840929066736,
      "tokens_seen": 1449328640
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002832698094282848,
      "loss": 2.9162,
      "theoretical_loss": 3.5267699396462593,
      "tokens_seen": 1449394176
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028325977933801405,
      "loss": 2.6923,
      "theoretical_loss": 3.526755787204964,
      "tokens_seen": 1449459712
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028324974924774323,
      "loss": 2.7045,
      "theoretical_loss": 3.5267416355827033,
      "tokens_seen": 1449525248
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002832397191574724,
      "loss": 2.8655,
      "theoretical_loss": 3.5267274847793946,
      "tokens_seen": 1449590784
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002832296890672016,
      "loss": 2.8265,
      "theoretical_loss": 3.526713334794951,
      "tokens_seen": 1449656320
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028321965897693083,
      "loss": 2.9944,
      "theoretical_loss": 3.52669918562929,
      "tokens_seen": 1449721856
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028320962888665996,
      "loss": 2.8826,
      "theoretical_loss": 3.526685037282326,
      "tokens_seen": 1449787392
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002831995987963892,
      "loss": 2.8522,
      "theoretical_loss": 3.526670889753976,
      "tokens_seen": 1449852928
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002831895687061183,
      "loss": 2.9577,
      "theoretical_loss": 3.526656743044154,
      "tokens_seen": 1449918464
    },
    {
      "epoch": 0.44,
      "objective/train/advantage_avg": 0.4950864911079407,
      "objective/train/docs_used": 822517,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.5652053356170654,
      "objective/train/original_loss": 2.5652055740356445,
      "objective/train/theoretical_loss": 3.526642597152777,
      "objective/train/tokens_used": 1470444000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2451600879430771,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507550239562988,
      "objective/train/weighted_lm_loss": 2.695286989212036,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0442860126495361,
      "theoretical_loss": 3.526642597152777,
      "tokens_seen": 1449984000
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028317953861584755,
      "loss": 2.7314,
      "theoretical_loss": 3.526642597152777,
      "tokens_seen": 1449984000
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028316950852557674,
      "loss": 2.8264,
      "theoretical_loss": 3.52662845207976,
      "tokens_seen": 1450049536
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002831594784353059,
      "loss": 2.8737,
      "theoretical_loss": 3.5266143078250183,
      "tokens_seen": 1450115072
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002831494483450351,
      "loss": 2.8006,
      "theoretical_loss": 3.5266001643884684,
      "tokens_seen": 1450180608
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002831394182547643,
      "loss": 2.712,
      "theoretical_loss": 3.526586021770026,
      "tokens_seen": 1450246144
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028312938816449346,
      "loss": 2.8927,
      "theoretical_loss": 3.526571879969606,
      "tokens_seen": 1450311680
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002831193580742227,
      "loss": 2.8692,
      "theoretical_loss": 3.5265577389871248,
      "tokens_seen": 1450377216
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002831093279839518,
      "loss": 2.8967,
      "theoretical_loss": 3.5265435988224976,
      "tokens_seen": 1450442752
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028309929789368106,
      "loss": 2.7587,
      "theoretical_loss": 3.5265294594756407,
      "tokens_seen": 1450508288
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002830892678034102,
      "loss": 2.8851,
      "theoretical_loss": 3.52651532094647,
      "tokens_seen": 1450573824
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002830792377131394,
      "loss": 2.9289,
      "theoretical_loss": 3.5265011832349007,
      "tokens_seen": 1450639360
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002830692076228686,
      "loss": 2.8987,
      "theoretical_loss": 3.526487046340849,
      "tokens_seen": 1450704896
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002830591775325978,
      "loss": 2.9318,
      "theoretical_loss": 3.5264729102642294,
      "tokens_seen": 1450770432
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028304914744232696,
      "loss": 2.8512,
      "theoretical_loss": 3.5264587750049596,
      "tokens_seen": 1450835968
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002830391173520562,
      "loss": 2.769,
      "theoretical_loss": 3.526444640562955,
      "tokens_seen": 1450901504
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002830290872617854,
      "loss": 3.014,
      "theoretical_loss": 3.5264305069381305,
      "tokens_seen": 1450967040
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028301905717151456,
      "loss": 2.8247,
      "theoretical_loss": 3.526416374130403,
      "tokens_seen": 1451032576
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028300902708124374,
      "loss": 2.5275,
      "theoretical_loss": 3.5264022421396874,
      "tokens_seen": 1451098112
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002829989969909729,
      "loss": 2.9224,
      "theoretical_loss": 3.5263881109659003,
      "tokens_seen": 1451163648
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028298896690070216,
      "loss": 2.6712,
      "theoretical_loss": 3.526373980608957,
      "tokens_seen": 1451229184
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002829789368104313,
      "loss": 3.0252,
      "theoretical_loss": 3.526359851068774,
      "tokens_seen": 1451294720
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002829689067201605,
      "loss": 2.6949,
      "theoretical_loss": 3.526345722345267,
      "tokens_seen": 1451360256
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028295887662988965,
      "loss": 2.8236,
      "theoretical_loss": 3.5263315944383518,
      "tokens_seen": 1451425792
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002829488465396189,
      "loss": 2.9062,
      "theoretical_loss": 3.5263174673479445,
      "tokens_seen": 1451491328
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028293881644934806,
      "loss": 2.902,
      "theoretical_loss": 3.5263033410739606,
      "tokens_seen": 1451556864
    },
    {
      "epoch": 0.44,
      "objective/train/advantage_avg": 0.4913836121559143,
      "objective/train/docs_used": 823297,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.1471612453460693,
      "objective/train/original_loss": 3.147160530090332,
      "objective/train/theoretical_loss": 3.526289215616317,
      "objective/train/tokens_used": 1472082400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24163448810577393,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0503665208816528,
      "objective/train/weighted_lm_loss": 3.3056864738464355,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0304546356201172,
      "theoretical_loss": 3.526289215616317,
      "tokens_seen": 1451622400
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028292878635907725,
      "loss": 3.1203,
      "theoretical_loss": 3.526289215616317,
      "tokens_seen": 1451622400
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002829187562688064,
      "loss": 2.8576,
      "theoretical_loss": 3.5262750909749285,
      "tokens_seen": 1451687936
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028290872617853566,
      "loss": 2.7187,
      "theoretical_loss": 3.526260967149712,
      "tokens_seen": 1451753472
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002828986960882648,
      "loss": 2.8878,
      "theoretical_loss": 3.5262468441405828,
      "tokens_seen": 1451819008
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.000282888665997994,
      "loss": 2.7579,
      "theoretical_loss": 3.5262327219474576,
      "tokens_seen": 1451884544
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028287863590772315,
      "loss": 2.9111,
      "theoretical_loss": 3.5262186005702523,
      "tokens_seen": 1451950080
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002828686058174524,
      "loss": 2.8076,
      "theoretical_loss": 3.5262044800088823,
      "tokens_seen": 1452015616
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028285857572718157,
      "loss": 3.0668,
      "theoretical_loss": 3.5261903602632643,
      "tokens_seen": 1452081152
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028284854563691075,
      "loss": 2.9663,
      "theoretical_loss": 3.526176241333314,
      "tokens_seen": 1452146688
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028283851554663993,
      "loss": 2.7457,
      "theoretical_loss": 3.526162123218948,
      "tokens_seen": 1452212224
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002828284854563691,
      "loss": 2.9509,
      "theoretical_loss": 3.5261480059200814,
      "tokens_seen": 1452277760
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002828184553660983,
      "loss": 2.8563,
      "theoretical_loss": 3.5261338894366316,
      "tokens_seen": 1452343296
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002828084252758275,
      "loss": 2.7868,
      "theoretical_loss": 3.5261197737685137,
      "tokens_seen": 1452408832
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028279839518555665,
      "loss": 2.7904,
      "theoretical_loss": 3.526105658915644,
      "tokens_seen": 1452474368
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002827883650952859,
      "loss": 2.6475,
      "theoretical_loss": 3.5260915448779384,
      "tokens_seen": 1452539904
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.000282778335005015,
      "loss": 3.0529,
      "theoretical_loss": 3.526077431655314,
      "tokens_seen": 1452605440
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028276830491474425,
      "loss": 2.6582,
      "theoretical_loss": 3.5260633192476862,
      "tokens_seen": 1452670976
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028275827482447343,
      "loss": 2.9218,
      "theoretical_loss": 3.5260492076549714,
      "tokens_seen": 1452736512
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002827482447342026,
      "loss": 2.7087,
      "theoretical_loss": 3.5260350968770857,
      "tokens_seen": 1452802048
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002827382146439318,
      "loss": 2.742,
      "theoretical_loss": 3.526020986913945,
      "tokens_seen": 1452867584
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028272818455366103,
      "loss": 2.7647,
      "theoretical_loss": 3.5260068777654663,
      "tokens_seen": 1452933120
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028271815446339016,
      "loss": 2.8252,
      "theoretical_loss": 3.525992769431565,
      "tokens_seen": 1452998656
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002827081243731194,
      "loss": 2.8376,
      "theoretical_loss": 3.5259786619121583,
      "tokens_seen": 1453064192
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002826980942828485,
      "loss": 2.8689,
      "theoretical_loss": 3.5259645552071612,
      "tokens_seen": 1453129728
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028268806419257775,
      "loss": 2.8019,
      "theoretical_loss": 3.5259504493164906,
      "tokens_seen": 1453195264
    },
    {
      "epoch": 0.44,
      "objective/train/advantage_avg": 0.496718168258667,
      "objective/train/docs_used": 824661,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.759120225906372,
      "objective/train/original_loss": 2.759119749069214,
      "objective/train/theoretical_loss": 3.525936344240063,
      "objective/train/tokens_used": 1473720800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24677777290344238,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0509263277053833,
      "objective/train/weighted_lm_loss": 2.8995707035064697,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0417574644088745,
      "theoretical_loss": 3.525936344240063,
      "tokens_seen": 1453260800
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028267803410230694,
      "loss": 2.9728,
      "theoretical_loss": 3.525936344240063,
      "tokens_seen": 1453260800
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002826680040120361,
      "loss": 2.7242,
      "theoretical_loss": 3.5259222399777945,
      "tokens_seen": 1453326336
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002826579739217653,
      "loss": 2.88,
      "theoretical_loss": 3.5259081365296012,
      "tokens_seen": 1453391872
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002826479438314945,
      "loss": 2.8342,
      "theoretical_loss": 3.5258940338953995,
      "tokens_seen": 1453457408
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028263791374122366,
      "loss": 2.9003,
      "theoretical_loss": 3.5258799320751057,
      "tokens_seen": 1453522944
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002826278836509529,
      "loss": 2.791,
      "theoretical_loss": 3.5258658310686366,
      "tokens_seen": 1453588480
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.000282617853560682,
      "loss": 2.9074,
      "theoretical_loss": 3.5258517308759076,
      "tokens_seen": 1453654016
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028260782347041126,
      "loss": 2.6948,
      "theoretical_loss": 3.525837631496836,
      "tokens_seen": 1453719552
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002825977933801404,
      "loss": 2.8883,
      "theoretical_loss": 3.525823532931337,
      "tokens_seen": 1453785088
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002825877632898696,
      "loss": 2.7968,
      "theoretical_loss": 3.5258094351793288,
      "tokens_seen": 1453850624
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002825777331995988,
      "loss": 2.8818,
      "theoretical_loss": 3.5257953382407266,
      "tokens_seen": 1453916160
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.000282567703109328,
      "loss": 2.8926,
      "theoretical_loss": 3.5257812421154466,
      "tokens_seen": 1453981696
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028255767301905716,
      "loss": 2.8757,
      "theoretical_loss": 3.5257671468034055,
      "tokens_seen": 1454047232
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002825476429287864,
      "loss": 2.9417,
      "theoretical_loss": 3.5257530523045197,
      "tokens_seen": 1454112768
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002825376128385155,
      "loss": 2.7264,
      "theoretical_loss": 3.5257389586187067,
      "tokens_seen": 1454178304
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028252758274824476,
      "loss": 2.5822,
      "theoretical_loss": 3.525724865745881,
      "tokens_seen": 1454243840
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002825175526579739,
      "loss": 2.8554,
      "theoretical_loss": 3.5257107736859603,
      "tokens_seen": 1454309376
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002825075225677031,
      "loss": 2.9217,
      "theoretical_loss": 3.525696682438861,
      "tokens_seen": 1454374912
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002824974924774323,
      "loss": 2.9367,
      "theoretical_loss": 3.5256825920044994,
      "tokens_seen": 1454440448
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002824874623871615,
      "loss": 2.8853,
      "theoretical_loss": 3.525668502382792,
      "tokens_seen": 1454505984
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028247743229689067,
      "loss": 2.9131,
      "theoretical_loss": 3.5256544135736556,
      "tokens_seen": 1454571520
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028246740220661985,
      "loss": 2.8455,
      "theoretical_loss": 3.5256403255770064,
      "tokens_seen": 1454637056
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028245737211634903,
      "loss": 2.9529,
      "theoretical_loss": 3.5256262383927606,
      "tokens_seen": 1454702592
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028244734202607826,
      "loss": 2.7642,
      "theoretical_loss": 3.5256121520208357,
      "tokens_seen": 1454768128
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002824373119358074,
      "loss": 3.148,
      "theoretical_loss": 3.525598066461148,
      "tokens_seen": 1454833664
    },
    {
      "epoch": 0.44,
      "objective/train/advantage_avg": 0.4938875138759613,
      "objective/train/docs_used": 824875,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.662714958190918,
      "objective/train/original_loss": 2.662714719772339,
      "objective/train/theoretical_loss": 3.525583981713613,
      "objective/train/tokens_used": 1475359200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2439824640750885,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506290197372437,
      "objective/train/weighted_lm_loss": 2.7976882457733154,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0363596677780151,
      "theoretical_loss": 3.525583981713613,
      "tokens_seen": 1454899200
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002824272818455366,
      "loss": 2.8207,
      "theoretical_loss": 3.525583981713613,
      "tokens_seen": 1454899200
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002824172517552658,
      "loss": 2.84,
      "theoretical_loss": 3.5255698977781487,
      "tokens_seen": 1454964736
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.000282407221664995,
      "loss": 2.8037,
      "theoretical_loss": 3.5255558146546715,
      "tokens_seen": 1455030272
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028239719157472417,
      "loss": 2.8936,
      "theoretical_loss": 3.525541732343097,
      "tokens_seen": 1455095808
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028238716148445335,
      "loss": 2.8563,
      "theoretical_loss": 3.5255276508433426,
      "tokens_seen": 1455161344
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028237713139418253,
      "loss": 3.089,
      "theoretical_loss": 3.5255135701553253,
      "tokens_seen": 1455226880
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028236710130391177,
      "loss": 2.8439,
      "theoretical_loss": 3.525499490278961,
      "tokens_seen": 1455292416
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028236710130391177,
      "loss": 2.9978,
      "theoretical_loss": 3.5254854112141665,
      "tokens_seen": 1455357952
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002823570712136409,
      "loss": 2.8608,
      "theoretical_loss": 3.5254713329608585,
      "tokens_seen": 1455423488
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028234704112337013,
      "loss": 3.0372,
      "theoretical_loss": 3.5254572555189543,
      "tokens_seen": 1455489024
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028233701103309926,
      "loss": 3.0779,
      "theoretical_loss": 3.52544317888837,
      "tokens_seen": 1455554560
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002823269809428285,
      "loss": 2.754,
      "theoretical_loss": 3.525429103069022,
      "tokens_seen": 1455620096
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002823169508525577,
      "loss": 3.0248,
      "theoretical_loss": 3.525415028060828,
      "tokens_seen": 1455685632
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028230692076228685,
      "loss": 3.0621,
      "theoretical_loss": 3.525400953863704,
      "tokens_seen": 1455751168
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028229689067201604,
      "loss": 3.1086,
      "theoretical_loss": 3.5253868804775665,
      "tokens_seen": 1455816704
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002822868605817452,
      "loss": 2.7675,
      "theoretical_loss": 3.5253728079023334,
      "tokens_seen": 1455882240
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028227683049147445,
      "loss": 3.0289,
      "theoretical_loss": 3.5253587361379206,
      "tokens_seen": 1455947776
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028226680040120363,
      "loss": 2.9936,
      "theoretical_loss": 3.525344665184245,
      "tokens_seen": 1456013312
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002822567703109328,
      "loss": 3.0234,
      "theoretical_loss": 3.5253305950412237,
      "tokens_seen": 1456078848
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.000282246740220662,
      "loss": 2.9025,
      "theoretical_loss": 3.525316525708773,
      "tokens_seen": 1456144384
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028223671013039123,
      "loss": 2.745,
      "theoretical_loss": 3.52530245718681,
      "tokens_seen": 1456209920
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028222668004012036,
      "loss": 2.8491,
      "theoretical_loss": 3.5252883894752522,
      "tokens_seen": 1456275456
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002822166499498496,
      "loss": 2.8785,
      "theoretical_loss": 3.525274322574015,
      "tokens_seen": 1456340992
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002822066198595787,
      "loss": 2.9389,
      "theoretical_loss": 3.5252602564830164,
      "tokens_seen": 1456406528
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028219658976930795,
      "loss": 2.9241,
      "theoretical_loss": 3.5252461912021733,
      "tokens_seen": 1456472064
    },
    {
      "epoch": 0.44,
      "objective/train/advantage_avg": 0.49269911646842957,
      "objective/train/docs_used": 826174,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.4546267986297607,
      "objective/train/original_loss": 2.454627275466919,
      "objective/train/theoretical_loss": 3.525232126731402,
      "objective/train/tokens_used": 1476997600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24282300472259521,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050504207611084,
      "objective/train/weighted_lm_loss": 2.578758955001831,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0437887907028198,
      "theoretical_loss": 3.525232126731402,
      "tokens_seen": 1456537600
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028218655967903714,
      "loss": 2.9573,
      "theoretical_loss": 3.525232126731402,
      "tokens_seen": 1456537600
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002821765295887663,
      "loss": 3.1114,
      "theoretical_loss": 3.5252180630706196,
      "tokens_seen": 1456603136
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002821664994984955,
      "loss": 2.8835,
      "theoretical_loss": 3.525204000219743,
      "tokens_seen": 1456668672
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002821564694082247,
      "loss": 2.8606,
      "theoretical_loss": 3.5251899381786895,
      "tokens_seen": 1456734208
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028214643931795386,
      "loss": 3.0288,
      "theoretical_loss": 3.525175876947375,
      "tokens_seen": 1456799744
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002821364092276831,
      "loss": 3.0108,
      "theoretical_loss": 3.525161816525718,
      "tokens_seen": 1456865280
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002821263791374122,
      "loss": 2.8138,
      "theoretical_loss": 3.5251477569136336,
      "tokens_seen": 1456930816
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028211634904714146,
      "loss": 2.9774,
      "theoretical_loss": 3.525133698111041,
      "tokens_seen": 1456996352
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002821063189568706,
      "loss": 2.8146,
      "theoretical_loss": 3.525119640117855,
      "tokens_seen": 1457061888
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002820962888665998,
      "loss": 2.6954,
      "theoretical_loss": 3.525105582933994,
      "tokens_seen": 1457127424
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.000282086258776329,
      "loss": 2.7061,
      "theoretical_loss": 3.5250915265593745,
      "tokens_seen": 1457192960
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002820762286860582,
      "loss": 2.8904,
      "theoretical_loss": 3.5250774709939137,
      "tokens_seen": 1457258496
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028206619859578736,
      "loss": 2.8872,
      "theoretical_loss": 3.5250634162375283,
      "tokens_seen": 1457324032
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002820561685055166,
      "loss": 2.8639,
      "theoretical_loss": 3.5250493622901358,
      "tokens_seen": 1457389568
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002820461384152457,
      "loss": 2.9912,
      "theoretical_loss": 3.525035309151653,
      "tokens_seen": 1457455104
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028203610832497496,
      "loss": 2.9855,
      "theoretical_loss": 3.525021256821997,
      "tokens_seen": 1457520640
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002820260782347041,
      "loss": 2.9959,
      "theoretical_loss": 3.525007205301085,
      "tokens_seen": 1457586176
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002820160481444333,
      "loss": 2.9785,
      "theoretical_loss": 3.524993154588834,
      "tokens_seen": 1457651712
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002820060180541625,
      "loss": 2.9816,
      "theoretical_loss": 3.524979104685161,
      "tokens_seen": 1457717248
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002819959879638917,
      "loss": 2.8782,
      "theoretical_loss": 3.524965055589983,
      "tokens_seen": 1457782784
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028198595787362087,
      "loss": 2.9613,
      "theoretical_loss": 3.524951007303218,
      "tokens_seen": 1457848320
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028197592778335005,
      "loss": 3.0084,
      "theoretical_loss": 3.524936959824782,
      "tokens_seen": 1457913856
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028196589769307923,
      "loss": 2.8823,
      "theoretical_loss": 3.524922913154593,
      "tokens_seen": 1457979392
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028195586760280846,
      "loss": 3.154,
      "theoretical_loss": 3.5249088672925675,
      "tokens_seen": 1458044928
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002819458375125376,
      "loss": 3.0885,
      "theoretical_loss": 3.524894822238623,
      "tokens_seen": 1458110464
    },
    {
      "epoch": 0.44,
      "objective/train/advantage_avg": 0.4921771287918091,
      "objective/train/docs_used": 826810,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.837728977203369,
      "objective/train/original_loss": 2.837728977203369,
      "objective/train/theoretical_loss": 3.524880777992677,
      "objective/train/tokens_used": 1478636000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2423056662082672,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050449252128601,
      "objective/train/weighted_lm_loss": 2.9809322357177734,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0467102527618408,
      "theoretical_loss": 3.524880777992677,
      "tokens_seen": 1458176000
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002819358074222668,
      "loss": 2.9449,
      "theoretical_loss": 3.524880777992677,
      "tokens_seen": 1458176000
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.000281925777331996,
      "loss": 2.8612,
      "theoretical_loss": 3.524866734554646,
      "tokens_seen": 1458241536
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002819157472417252,
      "loss": 2.9304,
      "theoretical_loss": 3.524852691924448,
      "tokens_seen": 1458307072
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028190571715145437,
      "loss": 2.9643,
      "theoretical_loss": 3.5248386501019997,
      "tokens_seen": 1458372608
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028189568706118355,
      "loss": 2.8556,
      "theoretical_loss": 3.5248246090872186,
      "tokens_seen": 1458438144
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028188565697091273,
      "loss": 2.7018,
      "theoretical_loss": 3.5248105688800218,
      "tokens_seen": 1458503680
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028187562688064197,
      "loss": 2.6978,
      "theoretical_loss": 3.5247965294803265,
      "tokens_seen": 1458569216
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002818655967903711,
      "loss": 2.8132,
      "theoretical_loss": 3.52478249088805,
      "tokens_seen": 1458634752
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028185556670010033,
      "loss": 3.0911,
      "theoretical_loss": 3.5247684531031096,
      "tokens_seen": 1458700288
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028184553660982946,
      "loss": 2.9721,
      "theoretical_loss": 3.524754416125423,
      "tokens_seen": 1458765824
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002818355065195587,
      "loss": 2.9236,
      "theoretical_loss": 3.524740379954907,
      "tokens_seen": 1458831360
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002818254764292879,
      "loss": 2.7774,
      "theoretical_loss": 3.524726344591479,
      "tokens_seen": 1458896896
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028181544633901705,
      "loss": 2.7308,
      "theoretical_loss": 3.5247123100350572,
      "tokens_seen": 1458962432
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028180541624874624,
      "loss": 2.9554,
      "theoretical_loss": 3.524698276285558,
      "tokens_seen": 1459027968
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002817953861584754,
      "loss": 2.9036,
      "theoretical_loss": 3.5246842433428984,
      "tokens_seen": 1459093504
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002817853560682046,
      "loss": 2.8887,
      "theoretical_loss": 3.5246702112069963,
      "tokens_seen": 1459159040
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028177532597793383,
      "loss": 2.7622,
      "theoretical_loss": 3.52465617987777,
      "tokens_seen": 1459224576
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028176529588766296,
      "loss": 2.8374,
      "theoretical_loss": 3.5246421493551354,
      "tokens_seen": 1459290112
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002817552657973922,
      "loss": 2.682,
      "theoretical_loss": 3.5246281196390106,
      "tokens_seen": 1459355648
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002817452357071214,
      "loss": 2.6337,
      "theoretical_loss": 3.5246140907293135,
      "tokens_seen": 1459421184
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028173520561685056,
      "loss": 2.9703,
      "theoretical_loss": 3.5246000626259604,
      "tokens_seen": 1459486720
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028172517552657974,
      "loss": 2.7805,
      "theoretical_loss": 3.52458603532887,
      "tokens_seen": 1459552256
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002817151454363089,
      "loss": 2.765,
      "theoretical_loss": 3.524572008837958,
      "tokens_seen": 1459617792
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002817051153460381,
      "loss": 2.7576,
      "theoretical_loss": 3.524557983153144,
      "tokens_seen": 1459683328
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028169508525576734,
      "loss": 2.9032,
      "theoretical_loss": 3.5245439582743443,
      "tokens_seen": 1459748864
    },
    {
      "epoch": 0.44,
      "objective/train/advantage_avg": 0.4950461983680725,
      "objective/train/docs_used": 827996,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.7657337188720703,
      "objective/train/original_loss": 2.7657337188720703,
      "objective/train/theoretical_loss": 3.5245299342014764,
      "objective/train/tokens_used": 1480274400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2451310008764267,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050750732421875,
      "objective/train/weighted_lm_loss": 2.9060332775115967,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0425469875335693,
      "theoretical_loss": 3.5245299342014764,
      "tokens_seen": 1459814400
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028168505516549646,
      "loss": 2.6681,
      "theoretical_loss": 3.5245299342014764,
      "tokens_seen": 1459814400
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002816750250752257,
      "loss": 2.7547,
      "theoretical_loss": 3.524515910934458,
      "tokens_seen": 1459879936
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002816649949849548,
      "loss": 2.8533,
      "theoretical_loss": 3.524501888473207,
      "tokens_seen": 1459945472
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028165496489468406,
      "loss": 2.9688,
      "theoretical_loss": 3.52448786681764,
      "tokens_seen": 1460011008
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028164493480441324,
      "loss": 3.0157,
      "theoretical_loss": 3.5244738459676754,
      "tokens_seen": 1460076544
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002816349047141424,
      "loss": 2.7888,
      "theoretical_loss": 3.5244598259232305,
      "tokens_seen": 1460142080
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002816248746238716,
      "loss": 2.9081,
      "theoretical_loss": 3.5244458066842226,
      "tokens_seen": 1460207616
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002816148445336008,
      "loss": 2.6532,
      "theoretical_loss": 3.5244317882505696,
      "tokens_seen": 1460273152
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028160481444332997,
      "loss": 3.0732,
      "theoretical_loss": 3.524417770622189,
      "tokens_seen": 1460338688
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002815947843530592,
      "loss": 2.7795,
      "theoretical_loss": 3.5244037537989987,
      "tokens_seen": 1460404224
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028158475426278833,
      "loss": 2.8011,
      "theoretical_loss": 3.5243897377809157,
      "tokens_seen": 1460469760
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028157472417251756,
      "loss": 2.921,
      "theoretical_loss": 3.524375722567858,
      "tokens_seen": 1460535296
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028156469408224674,
      "loss": 2.6861,
      "theoretical_loss": 3.524361708159743,
      "tokens_seen": 1460600832
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002815546639919759,
      "loss": 2.8695,
      "theoretical_loss": 3.5243476945564893,
      "tokens_seen": 1460666368
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002815446339017051,
      "loss": 2.8285,
      "theoretical_loss": 3.524333681758013,
      "tokens_seen": 1460731904
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002815346038114343,
      "loss": 2.8409,
      "theoretical_loss": 3.524319669764233,
      "tokens_seen": 1460797440
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002815245737211635,
      "loss": 2.8096,
      "theoretical_loss": 3.524305658575067,
      "tokens_seen": 1460862976
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002815145436308927,
      "loss": 2.8174,
      "theoretical_loss": 3.524291648190432,
      "tokens_seen": 1460928512
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002815045135406219,
      "loss": 2.939,
      "theoretical_loss": 3.5242776386102452,
      "tokens_seen": 1460994048
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028149448345035107,
      "loss": 2.6776,
      "theoretical_loss": 3.524263629834426,
      "tokens_seen": 1461059584
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028148445336008025,
      "loss": 2.9685,
      "theoretical_loss": 3.524249621862891,
      "tokens_seen": 1461125120
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028147442326980943,
      "loss": 3.0455,
      "theoretical_loss": 3.5242356146955585,
      "tokens_seen": 1461190656
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028146439317953866,
      "loss": 2.7829,
      "theoretical_loss": 3.524221608332346,
      "tokens_seen": 1461256192
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002814543630892678,
      "loss": 3.0561,
      "theoretical_loss": 3.524207602773171,
      "tokens_seen": 1461321728
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.000281444332998997,
      "loss": 2.8412,
      "theoretical_loss": 3.524193598017952,
      "tokens_seen": 1461387264
    },
    {
      "epoch": 0.44,
      "objective/train/advantage_avg": 0.49334433674812317,
      "objective/train/docs_used": 828576,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.7861146926879883,
      "objective/train/original_loss": 2.7861149311065674,
      "objective/train/theoretical_loss": 3.524179594066606,
      "objective/train/tokens_used": 1481912800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2434796839952469,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505720376968384,
      "objective/train/weighted_lm_loss": 2.9269418716430664,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0425469875335693,
      "theoretical_loss": 3.524179594066606,
      "tokens_seen": 1461452800
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002814343029087262,
      "loss": 2.7593,
      "theoretical_loss": 3.524179594066606,
      "tokens_seen": 1461452800
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002814242728184554,
      "loss": 3.0261,
      "theoretical_loss": 3.524165590919051,
      "tokens_seen": 1461518336
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028141424272818457,
      "loss": 2.7784,
      "theoretical_loss": 3.5241515885752053,
      "tokens_seen": 1461583872
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028140421263791375,
      "loss": 2.903,
      "theoretical_loss": 3.5241375870349865,
      "tokens_seen": 1461649408
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028139418254764293,
      "loss": 2.9088,
      "theoretical_loss": 3.524123586298312,
      "tokens_seen": 1461714944
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028138415245737217,
      "loss": 3.0093,
      "theoretical_loss": 3.5241095863651006,
      "tokens_seen": 1461780480
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002813741223671013,
      "loss": 2.6677,
      "theoretical_loss": 3.5240955872352693,
      "tokens_seen": 1461846016
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028136409227683053,
      "loss": 3.0925,
      "theoretical_loss": 3.5240815889087362,
      "tokens_seen": 1461911552
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028135406218655966,
      "loss": 2.868,
      "theoretical_loss": 3.5240675913854194,
      "tokens_seen": 1461977088
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002813440320962889,
      "loss": 2.7651,
      "theoretical_loss": 3.524053594665237,
      "tokens_seen": 1462042624
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002813340020060181,
      "loss": 2.8511,
      "theoretical_loss": 3.5240395987481064,
      "tokens_seen": 1462108160
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028132397191574725,
      "loss": 2.8184,
      "theoretical_loss": 3.5240256036339455,
      "tokens_seen": 1462173696
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028131394182547644,
      "loss": 3.0112,
      "theoretical_loss": 3.524011609322673,
      "tokens_seen": 1462239232
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002813039117352056,
      "loss": 3.0919,
      "theoretical_loss": 3.5239976158142063,
      "tokens_seen": 1462304768
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002812938816449348,
      "loss": 3.1062,
      "theoretical_loss": 3.523983623108463,
      "tokens_seen": 1462370304
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028128385155466403,
      "loss": 2.8717,
      "theoretical_loss": 3.523969631205362,
      "tokens_seen": 1462435840
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028127382146439316,
      "loss": 3.0303,
      "theoretical_loss": 3.5239556401048207,
      "tokens_seen": 1462501376
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002812637913741224,
      "loss": 2.7774,
      "theoretical_loss": 3.523941649806757,
      "tokens_seen": 1462566912
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002812537612838516,
      "loss": 2.888,
      "theoretical_loss": 3.5239276603110894,
      "tokens_seen": 1462632448
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028124373119358076,
      "loss": 2.6995,
      "theoretical_loss": 3.5239136716177355,
      "tokens_seen": 1462697984
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028123370110330994,
      "loss": 2.7887,
      "theoretical_loss": 3.5238996837266137,
      "tokens_seen": 1462763520
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002812236710130391,
      "loss": 2.9736,
      "theoretical_loss": 3.5238856966376417,
      "tokens_seen": 1462829056
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002812136409227683,
      "loss": 2.7534,
      "theoretical_loss": 3.523871710350738,
      "tokens_seen": 1462894592
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028120361083249754,
      "loss": 3.0123,
      "theoretical_loss": 3.5238577248658203,
      "tokens_seen": 1462960128
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028119358074222666,
      "loss": 2.9188,
      "theoretical_loss": 3.523843740182807,
      "tokens_seen": 1463025664
    },
    {
      "epoch": 0.44,
      "objective/train/advantage_avg": 0.49464312195777893,
      "objective/train/docs_used": 830024,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.7842483520507812,
      "objective/train/original_loss": 2.784247875213623,
      "objective/train/theoretical_loss": 3.523829756301615,
      "objective/train/tokens_used": 1483551200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2447342425584793,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050708293914795,
      "objective/train/weighted_lm_loss": 2.9254021644592285,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0425469875335693,
      "theoretical_loss": 3.523829756301615,
      "tokens_seen": 1463091200
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002811835506519559,
      "loss": 2.9258,
      "theoretical_loss": 3.523829756301615,
      "tokens_seen": 1463091200
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.000281173520561685,
      "loss": 3.0745,
      "theoretical_loss": 3.5238157732221644,
      "tokens_seen": 1463156736
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028116349047141426,
      "loss": 2.9154,
      "theoretical_loss": 3.523801790944372,
      "tokens_seen": 1463222272
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028115346038114344,
      "loss": 2.5952,
      "theoretical_loss": 3.523787809468156,
      "tokens_seen": 1463287808
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002811434302908726,
      "loss": 2.8101,
      "theoretical_loss": 3.5237738287934355,
      "tokens_seen": 1463353344
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002811334002006018,
      "loss": 2.9453,
      "theoretical_loss": 3.523759848920128,
      "tokens_seen": 1463418880
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.000281123370110331,
      "loss": 2.9692,
      "theoretical_loss": 3.523745869848151,
      "tokens_seen": 1463484416
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028111334002006017,
      "loss": 2.9538,
      "theoretical_loss": 3.523731891577424,
      "tokens_seen": 1463549952
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002811033099297894,
      "loss": 2.9775,
      "theoretical_loss": 3.523717914107864,
      "tokens_seen": 1463615488
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028109327983951853,
      "loss": 2.7822,
      "theoretical_loss": 3.5237039374393904,
      "tokens_seen": 1463681024
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028108324974924776,
      "loss": 2.7686,
      "theoretical_loss": 3.52368996157192,
      "tokens_seen": 1463746560
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028107321965897695,
      "loss": 2.8267,
      "theoretical_loss": 3.5236759865053724,
      "tokens_seen": 1463812096
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002810631895687061,
      "loss": 2.6922,
      "theoretical_loss": 3.523662012239665,
      "tokens_seen": 1463877632
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002810531594784353,
      "loss": 2.8129,
      "theoretical_loss": 3.5236480387747164,
      "tokens_seen": 1463943168
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002810431293881645,
      "loss": 2.8273,
      "theoretical_loss": 3.5236340661104446,
      "tokens_seen": 1464008704
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028103309929789367,
      "loss": 2.7919,
      "theoretical_loss": 3.523620094246769,
      "tokens_seen": 1464074240
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002810230692076229,
      "loss": 2.8353,
      "theoretical_loss": 3.5236061231836056,
      "tokens_seen": 1464139776
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028101303911735203,
      "loss": 2.9283,
      "theoretical_loss": 3.523592152920875,
      "tokens_seen": 1464205312
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028100300902708127,
      "loss": 2.7395,
      "theoretical_loss": 3.5235781834584943,
      "tokens_seen": 1464270848
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028100300902708127,
      "loss": 2.8072,
      "theoretical_loss": 3.523564214796382,
      "tokens_seen": 1464336384
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002809929789368104,
      "loss": 3.0718,
      "theoretical_loss": 3.523550246934457,
      "tokens_seen": 1464401920
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028098294884653963,
      "loss": 2.8378,
      "theoretical_loss": 3.523536279872637,
      "tokens_seen": 1464467456
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002809729187562688,
      "loss": 2.6122,
      "theoretical_loss": 3.5235223136108402,
      "tokens_seen": 1464532992
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.000280962888665998,
      "loss": 2.6955,
      "theoretical_loss": 3.5235083481489857,
      "tokens_seen": 1464598528
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028095285857572717,
      "loss": 2.6339,
      "theoretical_loss": 3.523494383486992,
      "tokens_seen": 1464664064
    },
    {
      "epoch": 0.44,
      "objective/train/advantage_avg": 0.4931897222995758,
      "objective/train/docs_used": 830651,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.433992862701416,
      "objective/train/original_loss": 2.433992624282837,
      "objective/train/theoretical_loss": 3.5234804196247764,
      "objective/train/tokens_used": 1485189600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24331146478652954,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.05055570602417,
      "objective/train/weighted_lm_loss": 2.557037353515625,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 3.5234804196247764,
      "tokens_seen": 1464729600
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028094282848545635,
      "loss": 2.6584,
      "theoretical_loss": 3.5234804196247764,
      "tokens_seen": 1464729600
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028093279839518553,
      "loss": 3.1686,
      "theoretical_loss": 3.523466456562258,
      "tokens_seen": 1464795136
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028092276830491477,
      "loss": 2.9383,
      "theoretical_loss": 3.523452494299356,
      "tokens_seen": 1464860672
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002809127382146439,
      "loss": 2.8869,
      "theoretical_loss": 3.523438532835987,
      "tokens_seen": 1464926208
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028090270812437313,
      "loss": 2.6961,
      "theoretical_loss": 3.523424572172071,
      "tokens_seen": 1464991744
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002808926780341023,
      "loss": 3.1197,
      "theoretical_loss": 3.5234106123075257,
      "tokens_seen": 1465057280
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002808826479438315,
      "loss": 2.6472,
      "theoretical_loss": 3.52339665324227,
      "tokens_seen": 1465122816
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002808726178535607,
      "loss": 2.9734,
      "theoretical_loss": 3.523382694976222,
      "tokens_seen": 1465188352
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028086258776328986,
      "loss": 2.9959,
      "theoretical_loss": 3.5233687375093012,
      "tokens_seen": 1465253888
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028085255767301904,
      "loss": 2.8296,
      "theoretical_loss": 3.5233547808414247,
      "tokens_seen": 1465319424
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002808425275827483,
      "loss": 2.7322,
      "theoretical_loss": 3.5233408249725113,
      "tokens_seen": 1465384960
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002808324974924774,
      "loss": 2.7165,
      "theoretical_loss": 3.5233268699024807,
      "tokens_seen": 1465450496
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028082246740220664,
      "loss": 2.8759,
      "theoretical_loss": 3.5233129156312497,
      "tokens_seen": 1465516032
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028081243731193576,
      "loss": 2.6407,
      "theoretical_loss": 3.523298962158738,
      "tokens_seen": 1465581568
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.000280802407221665,
      "loss": 3.0034,
      "theoretical_loss": 3.523285009484865,
      "tokens_seen": 1465647104
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002807923771313942,
      "loss": 2.8795,
      "theoretical_loss": 3.523271057609547,
      "tokens_seen": 1465712640
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028078234704112336,
      "loss": 2.9191,
      "theoretical_loss": 3.5232571065327045,
      "tokens_seen": 1465778176
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002807723169508526,
      "loss": 2.869,
      "theoretical_loss": 3.5232431562542548,
      "tokens_seen": 1465843712
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002807622868605818,
      "loss": 2.8346,
      "theoretical_loss": 3.5232292067741176,
      "tokens_seen": 1465909248
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028075225677031096,
      "loss": 2.7451,
      "theoretical_loss": 3.523215258092211,
      "tokens_seen": 1465974784
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028074222668004014,
      "loss": 2.7081,
      "theoretical_loss": 3.523201310208454,
      "tokens_seen": 1466040320
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002807321965897693,
      "loss": 2.7074,
      "theoretical_loss": 3.5231873631227644,
      "tokens_seen": 1466105856
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002807221664994985,
      "loss": 2.8896,
      "theoretical_loss": 3.5231734168350615,
      "tokens_seen": 1466171392
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028071213640922774,
      "loss": 2.8476,
      "theoretical_loss": 3.5231594713452647,
      "tokens_seen": 1466236928
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028070210631895686,
      "loss": 2.9722,
      "theoretical_loss": 3.523145526653291,
      "tokens_seen": 1466302464
    },
    {
      "epoch": 0.44,
      "objective/train/advantage_avg": 0.4947197735309601,
      "objective/train/docs_used": 831840,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.9411041736602783,
      "objective/train/original_loss": 2.941103935241699,
      "objective/train/theoretical_loss": 3.5231315827590604,
      "objective/train/tokens_used": 1486828000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24482570588588715,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050716519355774,
      "objective/train/weighted_lm_loss": 3.0902981758117676,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0396549701690674,
      "theoretical_loss": 3.5231315827590604,
      "tokens_seen": 1466368000
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002806920762286861,
      "loss": 3.0052,
      "theoretical_loss": 3.5231315827590604,
      "tokens_seen": 1466368000
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002806820461384152,
      "loss": 2.8646,
      "theoretical_loss": 3.523117639662491,
      "tokens_seen": 1466433536
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028067201604814446,
      "loss": 2.8756,
      "theoretical_loss": 3.523103697363502,
      "tokens_seen": 1466499072
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028066198595787364,
      "loss": 2.8446,
      "theoretical_loss": 3.523089755862012,
      "tokens_seen": 1466564608
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002806519558676028,
      "loss": 2.7994,
      "theoretical_loss": 3.523075815157939,
      "tokens_seen": 1466630144
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.000280641925777332,
      "loss": 2.595,
      "theoretical_loss": 3.5230618752512024,
      "tokens_seen": 1466695680
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002806318956870612,
      "loss": 2.8648,
      "theoretical_loss": 3.5230479361417215,
      "tokens_seen": 1466761216
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028062186559679037,
      "loss": 2.8232,
      "theoretical_loss": 3.5230339978294145,
      "tokens_seen": 1466826752
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002806118355065196,
      "loss": 2.6544,
      "theoretical_loss": 3.5230200603142,
      "tokens_seen": 1466892288
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028060180541624873,
      "loss": 2.9725,
      "theoretical_loss": 3.523006123595997,
      "tokens_seen": 1466957824
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028059177532597796,
      "loss": 3.0577,
      "theoretical_loss": 3.522992187674725,
      "tokens_seen": 1467023360
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028058174523570715,
      "loss": 3.0457,
      "theoretical_loss": 3.5229782525503013,
      "tokens_seen": 1467088896
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002805717151454363,
      "loss": 3.0207,
      "theoretical_loss": 3.5229643182226464,
      "tokens_seen": 1467154432
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002805616850551655,
      "loss": 2.7663,
      "theoretical_loss": 3.5229503846916774,
      "tokens_seen": 1467219968
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002805516549648947,
      "loss": 2.7495,
      "theoretical_loss": 3.5229364519573148,
      "tokens_seen": 1467285504
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028054162487462387,
      "loss": 3.0024,
      "theoretical_loss": 3.522922520019477,
      "tokens_seen": 1467351040
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002805315947843531,
      "loss": 2.6673,
      "theoretical_loss": 3.5229085888780824,
      "tokens_seen": 1467416576
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028052156469408223,
      "loss": 3.1445,
      "theoretical_loss": 3.5228946585330507,
      "tokens_seen": 1467482112
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028051153460381147,
      "loss": 2.8888,
      "theoretical_loss": 3.5228807289843003,
      "tokens_seen": 1467547648
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002805015045135406,
      "loss": 2.9429,
      "theoretical_loss": 3.52286680023175,
      "tokens_seen": 1467613184
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028049147442326983,
      "loss": 2.7172,
      "theoretical_loss": 3.522852872275318,
      "tokens_seen": 1467678720
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.000280481444332999,
      "loss": 3.0274,
      "theoretical_loss": 3.5228389451149256,
      "tokens_seen": 1467744256
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002804714142427282,
      "loss": 2.8872,
      "theoretical_loss": 3.5228250187504897,
      "tokens_seen": 1467809792
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028046138415245737,
      "loss": 2.6277,
      "theoretical_loss": 3.5228110931819296,
      "tokens_seen": 1467875328
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028045135406218655,
      "loss": 2.6425,
      "theoretical_loss": 3.5227971684091655,
      "tokens_seen": 1467940864
    },
    {
      "epoch": 0.44,
      "objective/train/advantage_avg": 0.4939970374107361,
      "objective/train/docs_used": 832610,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.6945459842681885,
      "objective/train/original_loss": 2.6945455074310303,
      "objective/train/theoretical_loss": 3.522783244432115,
      "objective/train/tokens_used": 1488466400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2441299706697464,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506407022476196,
      "objective/train/weighted_lm_loss": 2.8310298919677734,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0338950157165527,
      "theoretical_loss": 3.522783244432115,
      "tokens_seen": 1468006400
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028044132397191574,
      "loss": 2.8156,
      "theoretical_loss": 3.522783244432115,
      "tokens_seen": 1468006400
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028043129388164497,
      "loss": 2.8246,
      "theoretical_loss": 3.5227693212506974,
      "tokens_seen": 1468071936
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002804212637913741,
      "loss": 3.1415,
      "theoretical_loss": 3.522755398864832,
      "tokens_seen": 1468137472
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028041123370110333,
      "loss": 2.9577,
      "theoretical_loss": 3.5227414772744376,
      "tokens_seen": 1468203008
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002804012036108325,
      "loss": 2.9456,
      "theoretical_loss": 3.522727556479434,
      "tokens_seen": 1468268544
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002803911735205617,
      "loss": 2.9321,
      "theoretical_loss": 3.5227136364797387,
      "tokens_seen": 1468334080
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.0002803811434302909,
      "loss": 2.879,
      "theoretical_loss": 3.5226997172752723,
      "tokens_seen": 1468399616
    },
    {
      "epoch": 0.44,
      "learning_rate": 0.00028037111334002006,
      "loss": 2.8808,
      "theoretical_loss": 3.5226857988659535,
      "tokens_seen": 1468465152
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00028036108324974924,
      "loss": 2.5715,
      "theoretical_loss": 3.522671881251701,
      "tokens_seen": 1468530688
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002803510531594785,
      "loss": 2.8601,
      "theoretical_loss": 3.522657964432434,
      "tokens_seen": 1468596224
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002803410230692076,
      "loss": 2.7396,
      "theoretical_loss": 3.522644048408072,
      "tokens_seen": 1468661760
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00028033099297893684,
      "loss": 2.754,
      "theoretical_loss": 3.5226301331785335,
      "tokens_seen": 1468727296
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00028032096288866596,
      "loss": 2.8429,
      "theoretical_loss": 3.522616218743738,
      "tokens_seen": 1468792832
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002803109327983952,
      "loss": 2.6877,
      "theoretical_loss": 3.522602305103605,
      "tokens_seen": 1468858368
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002803009027081244,
      "loss": 2.5542,
      "theoretical_loss": 3.5225883922580534,
      "tokens_seen": 1468923904
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00028029087261785356,
      "loss": 2.8853,
      "theoretical_loss": 3.522574480207002,
      "tokens_seen": 1468989440
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00028028084252758274,
      "loss": 2.9328,
      "theoretical_loss": 3.52256056895037,
      "tokens_seen": 1469054976
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.000280270812437312,
      "loss": 2.725,
      "theoretical_loss": 3.522546658488077,
      "tokens_seen": 1469120512
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002802607823470411,
      "loss": 3.0031,
      "theoretical_loss": 3.5225327488200424,
      "tokens_seen": 1469186048
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00028025075225677034,
      "loss": 2.871,
      "theoretical_loss": 3.522518839946185,
      "tokens_seen": 1469251584
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00028024072216649947,
      "loss": 2.7967,
      "theoretical_loss": 3.5225049318664237,
      "tokens_seen": 1469317120
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002802306920762287,
      "loss": 3.024,
      "theoretical_loss": 3.522491024580679,
      "tokens_seen": 1469382656
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002802206619859579,
      "loss": 2.8226,
      "theoretical_loss": 3.5224771180888688,
      "tokens_seen": 1469448192
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00028021063189568706,
      "loss": 2.8883,
      "theoretical_loss": 3.522463212390913,
      "tokens_seen": 1469513728
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00028020060180541624,
      "loss": 2.9657,
      "theoretical_loss": 3.5224493074867307,
      "tokens_seen": 1469579264
    },
    {
      "epoch": 0.45,
      "objective/train/advantage_avg": 0.495137482881546,
      "objective/train/docs_used": 834187,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.79181170463562,
      "objective/train/original_loss": 2.791811466217041,
      "objective/train/theoretical_loss": 3.5224354033762415,
      "objective/train/tokens_used": 1490104800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2452307790517807,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050760269165039,
      "objective/train/weighted_lm_loss": 2.9335546493530273,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 3.5224354033762415,
      "tokens_seen": 1469644800
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002801905717151454,
      "loss": 3.0568,
      "theoretical_loss": 3.5224354033762415,
      "tokens_seen": 1469644800
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002801805416248746,
      "loss": 2.7983,
      "theoretical_loss": 3.522421500059364,
      "tokens_seen": 1469710336
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00028017051153460384,
      "loss": 2.9883,
      "theoretical_loss": 3.5224075975360187,
      "tokens_seen": 1469775872
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00028016048144433297,
      "loss": 2.5815,
      "theoretical_loss": 3.522393695806124,
      "tokens_seen": 1469841408
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002801504513540622,
      "loss": 2.8982,
      "theoretical_loss": 3.5223797948695994,
      "tokens_seen": 1469906944
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00028014042126379133,
      "loss": 2.8092,
      "theoretical_loss": 3.5223658947263643,
      "tokens_seen": 1469972480
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00028013039117352057,
      "loss": 3.006,
      "theoretical_loss": 3.5223519953763383,
      "tokens_seen": 1470038016
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00028012036108324975,
      "loss": 2.986,
      "theoretical_loss": 3.5223380968194404,
      "tokens_seen": 1470103552
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00028011033099297893,
      "loss": 2.7075,
      "theoretical_loss": 3.52232419905559,
      "tokens_seen": 1470169088
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002801003009027081,
      "loss": 2.822,
      "theoretical_loss": 3.5223103020847075,
      "tokens_seen": 1470234624
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00028009027081243735,
      "loss": 2.6889,
      "theoretical_loss": 3.522296405906711,
      "tokens_seen": 1470300160
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00028008024072216647,
      "loss": 2.7821,
      "theoretical_loss": 3.52228251052152,
      "tokens_seen": 1470365696
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002800702106318957,
      "loss": 3.1881,
      "theoretical_loss": 3.5222686159290544,
      "tokens_seen": 1470431232
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00028006018054162483,
      "loss": 2.9627,
      "theoretical_loss": 3.522254722129234,
      "tokens_seen": 1470496768
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00028005015045135407,
      "loss": 2.8502,
      "theoretical_loss": 3.5222408291219773,
      "tokens_seen": 1470562304
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00028004012036108325,
      "loss": 2.6707,
      "theoretical_loss": 3.522226936907205,
      "tokens_seen": 1470627840
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00028003009027081243,
      "loss": 2.6274,
      "theoretical_loss": 3.5222130454848353,
      "tokens_seen": 1470693376
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00028002006018054167,
      "loss": 2.8018,
      "theoretical_loss": 3.5221991548547886,
      "tokens_seen": 1470758912
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002800100300902708,
      "loss": 2.6815,
      "theoretical_loss": 3.5221852650169843,
      "tokens_seen": 1470824448
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00028000000000000003,
      "loss": 2.9486,
      "theoretical_loss": 3.5221713759713413,
      "tokens_seen": 1470889984
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002799899699097292,
      "loss": 3.1022,
      "theoretical_loss": 3.52215748771778,
      "tokens_seen": 1470955520
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002799799398194584,
      "loss": 2.8672,
      "theoretical_loss": 3.5221436002562188,
      "tokens_seen": 1471021056
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002799699097291876,
      "loss": 2.6962,
      "theoretical_loss": 3.522129713586579,
      "tokens_seen": 1471086592
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027995987963891675,
      "loss": 2.9748,
      "theoretical_loss": 3.522115827708778,
      "tokens_seen": 1471152128
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027994984954864594,
      "loss": 2.7488,
      "theoretical_loss": 3.522101942622737,
      "tokens_seen": 1471217664
    },
    {
      "epoch": 0.45,
      "objective/train/advantage_avg": 0.49268078804016113,
      "objective/train/docs_used": 834912,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.705773115158081,
      "objective/train/original_loss": 2.705772876739502,
      "objective/train/theoretical_loss": 3.522088058328375,
      "objective/train/tokens_used": 1491743200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2428411990404129,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505025386810303,
      "objective/train/weighted_lm_loss": 2.8424150943756104,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0304546356201172,
      "theoretical_loss": 3.522088058328375,
      "tokens_seen": 1471283200
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027993981945837517,
      "loss": 2.689,
      "theoretical_loss": 3.522088058328375,
      "tokens_seen": 1471283200
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002799297893681043,
      "loss": 2.9797,
      "theoretical_loss": 3.5220741748256117,
      "tokens_seen": 1471348736
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027991975927783353,
      "loss": 2.7612,
      "theoretical_loss": 3.522060292114366,
      "tokens_seen": 1471414272
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002799097291875627,
      "loss": 2.6619,
      "theoretical_loss": 3.522046410194559,
      "tokens_seen": 1471479808
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002798996990972919,
      "loss": 2.6353,
      "theoretical_loss": 3.52203252906611,
      "tokens_seen": 1471545344
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002798896690070211,
      "loss": 2.6818,
      "theoretical_loss": 3.5220186487289373,
      "tokens_seen": 1471610880
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027987963891675026,
      "loss": 2.8625,
      "theoretical_loss": 3.5220047691829617,
      "tokens_seen": 1471676416
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027986960882647944,
      "loss": 2.7745,
      "theoretical_loss": 3.5219908904281025,
      "tokens_seen": 1471741952
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002798595787362087,
      "loss": 2.5563,
      "theoretical_loss": 3.52197701246428,
      "tokens_seen": 1471807488
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002798495486459378,
      "loss": 2.9823,
      "theoretical_loss": 3.521963135291413,
      "tokens_seen": 1471873024
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027983951855566704,
      "loss": 2.9567,
      "theoretical_loss": 3.5219492589094217,
      "tokens_seen": 1471938560
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027982948846539616,
      "loss": 2.9656,
      "theoretical_loss": 3.5219353833182256,
      "tokens_seen": 1472004096
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002798194583751254,
      "loss": 2.6689,
      "theoretical_loss": 3.5219215085177447,
      "tokens_seen": 1472069632
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002798094282848546,
      "loss": 2.6995,
      "theoretical_loss": 3.5219076345078983,
      "tokens_seen": 1472135168
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027979939819458376,
      "loss": 2.8404,
      "theoretical_loss": 3.5218937612886068,
      "tokens_seen": 1472200704
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027978936810431294,
      "loss": 3.0057,
      "theoretical_loss": 3.5218798888597895,
      "tokens_seen": 1472266240
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002797793380140422,
      "loss": 2.7993,
      "theoretical_loss": 3.5218660172213663,
      "tokens_seen": 1472331776
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002797693079237713,
      "loss": 2.6168,
      "theoretical_loss": 3.5218521463732566,
      "tokens_seen": 1472397312
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027975927783350054,
      "loss": 2.7381,
      "theoretical_loss": 3.521838276315381,
      "tokens_seen": 1472462848
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027974924774322967,
      "loss": 2.7741,
      "theoretical_loss": 3.521824407047659,
      "tokens_seen": 1472528384
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002797392176529589,
      "loss": 2.8261,
      "theoretical_loss": 3.5218105385700103,
      "tokens_seen": 1472593920
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002797291875626881,
      "loss": 2.9056,
      "theoretical_loss": 3.5217966708823543,
      "tokens_seen": 1472659456
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027971915747241726,
      "loss": 2.7242,
      "theoretical_loss": 3.5217828039846113,
      "tokens_seen": 1472724992
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027970912738214644,
      "loss": 2.711,
      "theoretical_loss": 3.5217689378767014,
      "tokens_seen": 1472790528
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002796990972918756,
      "loss": 2.8963,
      "theoretical_loss": 3.521755072558544,
      "tokens_seen": 1472856064
    },
    {
      "epoch": 0.45,
      "objective/train/advantage_avg": 0.4934503734111786,
      "objective/train/docs_used": 835528,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.6959166526794434,
      "objective/train/original_loss": 2.6959166526794434,
      "objective/train/theoretical_loss": 3.5217412080300594,
      "objective/train/tokens_used": 1493381600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2435677945613861,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505831241607666,
      "objective/train/weighted_lm_loss": 2.832193374633789,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0408108234405518,
      "theoretical_loss": 3.5217412080300594,
      "tokens_seen": 1472921600
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002796890672016048,
      "loss": 2.7985,
      "theoretical_loss": 3.5217412080300594,
      "tokens_seen": 1472921600
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027967903711133404,
      "loss": 2.7084,
      "theoretical_loss": 3.5217273442911674,
      "tokens_seen": 1472987136
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027966900702106317,
      "loss": 2.5421,
      "theoretical_loss": 3.5217134813417874,
      "tokens_seen": 1473052672
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002796589769307924,
      "loss": 2.912,
      "theoretical_loss": 3.52169961918184,
      "tokens_seen": 1473118208
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027964894684052153,
      "loss": 2.56,
      "theoretical_loss": 3.5216857578112446,
      "tokens_seen": 1473183744
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027963891675025077,
      "loss": 2.7491,
      "theoretical_loss": 3.521671897229922,
      "tokens_seen": 1473249280
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027962888665997995,
      "loss": 2.8231,
      "theoretical_loss": 3.521658037437791,
      "tokens_seen": 1473314816
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027961885656970913,
      "loss": 2.6355,
      "theoretical_loss": 3.521644178434772,
      "tokens_seen": 1473380352
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002796088264794383,
      "loss": 2.6818,
      "theoretical_loss": 3.521630320220785,
      "tokens_seen": 1473445888
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027959879638916755,
      "loss": 3.0377,
      "theoretical_loss": 3.5216164627957505,
      "tokens_seen": 1473511424
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027958876629889667,
      "loss": 2.5856,
      "theoretical_loss": 3.521602606159588,
      "tokens_seen": 1473576960
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002795787362086259,
      "loss": 2.7522,
      "theoretical_loss": 3.521588750312217,
      "tokens_seen": 1473642496
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027956870611835503,
      "loss": 2.8212,
      "theoretical_loss": 3.5215748952535586,
      "tokens_seen": 1473708032
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027955867602808427,
      "loss": 2.7001,
      "theoretical_loss": 3.5215610409835323,
      "tokens_seen": 1473773568
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027954864593781345,
      "loss": 2.7374,
      "theoretical_loss": 3.521547187502058,
      "tokens_seen": 1473839104
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027953861584754263,
      "loss": 2.6699,
      "theoretical_loss": 3.521533334809056,
      "tokens_seen": 1473904640
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002795285857572718,
      "loss": 2.6546,
      "theoretical_loss": 3.521519482904446,
      "tokens_seen": 1473970176
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.000279518555667001,
      "loss": 2.8058,
      "theoretical_loss": 3.5215056317881483,
      "tokens_seen": 1474035712
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002795085255767302,
      "loss": 2.7486,
      "theoretical_loss": 3.5214917814600835,
      "tokens_seen": 1474101248
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002794984954864594,
      "loss": 2.9619,
      "theoretical_loss": 3.5214779319201712,
      "tokens_seen": 1474166784
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027948846539618854,
      "loss": 2.667,
      "theoretical_loss": 3.5214640831683317,
      "tokens_seen": 1474232320
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002794784353059178,
      "loss": 2.668,
      "theoretical_loss": 3.521450235204485,
      "tokens_seen": 1474297856
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002794684052156469,
      "loss": 2.7509,
      "theoretical_loss": 3.5214363880285506,
      "tokens_seen": 1474363392
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027945837512537614,
      "loss": 2.8065,
      "theoretical_loss": 3.5214225416404497,
      "tokens_seen": 1474428928
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002794483450351053,
      "loss": 2.7163,
      "theoretical_loss": 3.521408696040102,
      "tokens_seen": 1474494464
    },
    {
      "debugging/Self-BLEU-5": 0.0051973805067677846,
      "debugging/distinct-1-grams": 0.8630434782608696,
      "debugging/distinct-2-grams": 1.0,
      "debugging/entropy-1-grams": 3.6737481794560187,
      "debugging/entropy-2-grams": 3.988984046564275,
      "debugging/length": 241.0,
      "debugging/num_segments": 2,
      "debugging/raw_token_scores_avg": 0.004314254969358444,
      "debugging/raw_token_scores_std": 0.0061650583520531654,
      "debugging/score": 0.008243313630907897,
      "debugging/score_std": 0.00011323233009488849,
      "epoch": 0.45,
      "objective/train/advantage_avg": 0.4956842362880707,
      "objective/train/docs_used": 836610,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.46694016456604,
      "objective/train/original_loss": 2.46694016456604,
      "objective/train/theoretical_loss": 3.521394851227428,
      "objective/train/tokens_used": 1495020000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2457408756017685,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050817608833313,
      "objective/train/weighted_lm_loss": 2.5923399925231934,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0396549701690674,
      "theoretical_loss": 3.521394851227428,
      "tokens_seen": 1474560000
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002794383149448345,
      "loss": 2.5624,
      "theoretical_loss": 3.521394851227428,
      "tokens_seen": 1474560000
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002794282848545637,
      "loss": 2.6715,
      "theoretical_loss": 3.521381007202347,
      "tokens_seen": 1474625536
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002794182547642929,
      "loss": 2.8409,
      "theoretical_loss": 3.5213671639647806,
      "tokens_seen": 1474691072
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027940822467402204,
      "loss": 2.846,
      "theoretical_loss": 3.5213533215146477,
      "tokens_seen": 1474756608
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002793981945837513,
      "loss": 3.0966,
      "theoretical_loss": 3.521339479851869,
      "tokens_seen": 1474822144
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002793881644934804,
      "loss": 2.7569,
      "theoretical_loss": 3.521325638976365,
      "tokens_seen": 1474887680
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027937813440320964,
      "loss": 2.9351,
      "theoretical_loss": 3.5213117988880556,
      "tokens_seen": 1474953216
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002793681043129388,
      "loss": 2.9418,
      "theoretical_loss": 3.521297959586861,
      "tokens_seen": 1475018752
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.000279358074222668,
      "loss": 2.6302,
      "theoretical_loss": 3.5212841210727017,
      "tokens_seen": 1475084288
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002793480441323972,
      "loss": 2.8876,
      "theoretical_loss": 3.5212702833454985,
      "tokens_seen": 1475149824
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027933801404212636,
      "loss": 3.0284,
      "theoretical_loss": 3.5212564464051703,
      "tokens_seen": 1475215360
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027932798395185554,
      "loss": 2.8301,
      "theoretical_loss": 3.5212426102516385,
      "tokens_seen": 1475280896
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002793179538615848,
      "loss": 2.761,
      "theoretical_loss": 3.521228774884823,
      "tokens_seen": 1475346432
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002793079237713139,
      "loss": 2.8876,
      "theoretical_loss": 3.5212149403046444,
      "tokens_seen": 1475411968
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027929789368104314,
      "loss": 2.9721,
      "theoretical_loss": 3.5212011065110227,
      "tokens_seen": 1475477504
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027928786359077227,
      "loss": 2.7493,
      "theoretical_loss": 3.5211872735038785,
      "tokens_seen": 1475543040
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002792778335005015,
      "loss": 2.9009,
      "theoretical_loss": 3.521173441283132,
      "tokens_seen": 1475608576
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027926780341023074,
      "loss": 2.9137,
      "theoretical_loss": 3.521159609848704,
      "tokens_seen": 1475674112
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027925777331995987,
      "loss": 2.6737,
      "theoretical_loss": 3.521145779200514,
      "tokens_seen": 1475739648
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002792477432296891,
      "loss": 2.6415,
      "theoretical_loss": 3.5211319493384834,
      "tokens_seen": 1475805184
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002792377131394183,
      "loss": 2.7374,
      "theoretical_loss": 3.521118120262532,
      "tokens_seen": 1475870720
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027922768304914746,
      "loss": 2.7524,
      "theoretical_loss": 3.52110429197258,
      "tokens_seen": 1475936256
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027921765295887664,
      "loss": 3.0209,
      "theoretical_loss": 3.521090464468548,
      "tokens_seen": 1476001792
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002792076228686058,
      "loss": 2.7024,
      "theoretical_loss": 3.5210766377503573,
      "tokens_seen": 1476067328
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.000279197592778335,
      "loss": 2.6889,
      "theoretical_loss": 3.5210628118179272,
      "tokens_seen": 1476132864
    },
    {
      "epoch": 0.45,
      "objective/train/advantage_avg": 0.4932645857334137,
      "objective/train/docs_used": 837052,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.4770302772521973,
      "objective/train/original_loss": 2.4770305156707764,
      "objective/train/theoretical_loss": 3.521048986671179,
      "objective/train/tokens_used": 1496658400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24345482885837555,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505640506744385,
      "objective/train/weighted_lm_loss": 2.6022560596466064,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0408108234405518,
      "theoretical_loss": 3.521048986671179,
      "tokens_seen": 1476198400
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027918756268806424,
      "loss": 2.7499,
      "theoretical_loss": 3.521048986671179,
      "tokens_seen": 1476198400
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027917753259779337,
      "loss": 2.86,
      "theoretical_loss": 3.5210351623100324,
      "tokens_seen": 1476263936
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002791675025075226,
      "loss": 2.7712,
      "theoretical_loss": 3.5210213387344087,
      "tokens_seen": 1476329472
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027915747241725173,
      "loss": 2.9104,
      "theoretical_loss": 3.521007515944228,
      "tokens_seen": 1476395008
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027914744232698097,
      "loss": 2.6546,
      "theoretical_loss": 3.5209936939394106,
      "tokens_seen": 1476460544
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027913741223671015,
      "loss": 2.9648,
      "theoretical_loss": 3.520979872719877,
      "tokens_seen": 1476526080
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027912738214643933,
      "loss": 2.7336,
      "theoretical_loss": 3.520966052285548,
      "tokens_seen": 1476591616
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002791173520561685,
      "loss": 3.0027,
      "theoretical_loss": 3.5209522326363443,
      "tokens_seen": 1476657152
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027910732196589775,
      "loss": 2.7156,
      "theoretical_loss": 3.5209384137721864,
      "tokens_seen": 1476722688
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027909729187562687,
      "loss": 2.7319,
      "theoretical_loss": 3.5209245956929944,
      "tokens_seen": 1476788224
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002790872617853561,
      "loss": 2.7716,
      "theoretical_loss": 3.5209107783986893,
      "tokens_seen": 1476853760
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027907723169508523,
      "loss": 2.862,
      "theoretical_loss": 3.5208969618891914,
      "tokens_seen": 1476919296
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027906720160481447,
      "loss": 2.8131,
      "theoretical_loss": 3.5208831461644214,
      "tokens_seen": 1476984832
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027905717151454365,
      "loss": 2.922,
      "theoretical_loss": 3.5208693312243007,
      "tokens_seen": 1477050368
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027904714142427283,
      "loss": 2.7789,
      "theoretical_loss": 3.5208555170687488,
      "tokens_seen": 1477115904
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.000279037111334002,
      "loss": 3.0042,
      "theoretical_loss": 3.5208417036976867,
      "tokens_seen": 1477181440
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002790270812437312,
      "loss": 2.8684,
      "theoretical_loss": 3.520827891111035,
      "tokens_seen": 1477246976
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002790170511534604,
      "loss": 2.7635,
      "theoretical_loss": 3.520814079308715,
      "tokens_seen": 1477312512
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002790070210631896,
      "loss": 3.0028,
      "theoretical_loss": 3.520800268290646,
      "tokens_seen": 1477378048
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027899699097291874,
      "loss": 2.9439,
      "theoretical_loss": 3.52078645805675,
      "tokens_seen": 1477443584
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.000278986960882648,
      "loss": 2.7591,
      "theoretical_loss": 3.5207726486069473,
      "tokens_seen": 1477509120
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002789769307923771,
      "loss": 2.8868,
      "theoretical_loss": 3.5207588399411582,
      "tokens_seen": 1477574656
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027896690070210634,
      "loss": 2.862,
      "theoretical_loss": 3.5207450320593043,
      "tokens_seen": 1477640192
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002789568706118355,
      "loss": 2.907,
      "theoretical_loss": 3.520731224961305,
      "tokens_seen": 1477705728
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002789468405215647,
      "loss": 2.9361,
      "theoretical_loss": 3.5207174186470827,
      "tokens_seen": 1477771264
    },
    {
      "epoch": 0.45,
      "objective/train/advantage_avg": 0.4943833649158478,
      "objective/train/docs_used": 837925,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.6808488368988037,
      "objective/train/original_loss": 2.680849313735962,
      "objective/train/theoretical_loss": 3.5207036131165568,
      "objective/train/tokens_used": 1498296800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2444789558649063,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506811141967773,
      "objective/train/weighted_lm_loss": 2.8165557384490967,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0396549701690674,
      "theoretical_loss": 3.5207036131165568,
      "tokens_seen": 1477836800
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002789368104312939,
      "loss": 2.8083,
      "theoretical_loss": 3.5207036131165568,
      "tokens_seen": 1477836800
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002789267803410231,
      "loss": 2.9078,
      "theoretical_loss": 3.5206898083696485,
      "tokens_seen": 1477902336
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027891675025075224,
      "loss": 2.8855,
      "theoretical_loss": 3.520676004406279,
      "tokens_seen": 1477967872
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002789067201604815,
      "loss": 2.745,
      "theoretical_loss": 3.520662201226368,
      "tokens_seen": 1478033408
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002788966900702106,
      "loss": 2.5943,
      "theoretical_loss": 3.5206483988298376,
      "tokens_seen": 1478098944
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027888665997993984,
      "loss": 2.7107,
      "theoretical_loss": 3.520634597216608,
      "tokens_seen": 1478164480
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.000278876629889669,
      "loss": 2.6102,
      "theoretical_loss": 3.5206207963865994,
      "tokens_seen": 1478230016
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002788665997993982,
      "loss": 2.9576,
      "theoretical_loss": 3.5206069963397333,
      "tokens_seen": 1478295552
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002788565697091274,
      "loss": 2.6969,
      "theoretical_loss": 3.5205931970759314,
      "tokens_seen": 1478361088
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027884653961885656,
      "loss": 2.7443,
      "theoretical_loss": 3.520579398595113,
      "tokens_seen": 1478426624
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027883650952858574,
      "loss": 2.7516,
      "theoretical_loss": 3.5205656008972,
      "tokens_seen": 1478492160
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.000278826479438315,
      "loss": 2.7216,
      "theoretical_loss": 3.520551803982112,
      "tokens_seen": 1478557696
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002788164493480441,
      "loss": 2.9747,
      "theoretical_loss": 3.5205380078497717,
      "tokens_seen": 1478623232
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027880641925777334,
      "loss": 3.0198,
      "theoretical_loss": 3.5205242125000984,
      "tokens_seen": 1478688768
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027879638916750247,
      "loss": 2.9188,
      "theoretical_loss": 3.5205104179330142,
      "tokens_seen": 1478754304
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002787863590772317,
      "loss": 2.7595,
      "theoretical_loss": 3.5204966241484397,
      "tokens_seen": 1478819840
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002787763289869609,
      "loss": 2.7555,
      "theoretical_loss": 3.5204828311462952,
      "tokens_seen": 1478885376
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027876629889669007,
      "loss": 2.4395,
      "theoretical_loss": 3.5204690389265023,
      "tokens_seen": 1478950912
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027875626880641925,
      "loss": 2.9997,
      "theoretical_loss": 3.520455247488982,
      "tokens_seen": 1479016448
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002787462387161485,
      "loss": 2.7879,
      "theoretical_loss": 3.520441456833655,
      "tokens_seen": 1479081984
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002787362086258776,
      "loss": 2.8623,
      "theoretical_loss": 3.520427666960442,
      "tokens_seen": 1479147520
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027872617853560685,
      "loss": 2.8489,
      "theoretical_loss": 3.5204138778692644,
      "tokens_seen": 1479213056
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027871614844533597,
      "loss": 3.0166,
      "theoretical_loss": 3.5204000895600434,
      "tokens_seen": 1479278592
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002787061183550652,
      "loss": 2.8597,
      "theoretical_loss": 3.5203863020326995,
      "tokens_seen": 1479344128
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002786960882647944,
      "loss": 2.8493,
      "theoretical_loss": 3.520372515287154,
      "tokens_seen": 1479409664
    },
    {
      "epoch": 0.45,
      "objective/train/advantage_avg": 0.4960938096046448,
      "objective/train/docs_used": 838738,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.921964645385742,
      "objective/train/original_loss": 2.9219648838043213,
      "objective/train/theoretical_loss": 3.520358729323328,
      "objective/train/tokens_used": 1499935200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24615660309791565,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0508607625961304,
      "objective/train/weighted_lm_loss": 3.0704691410064697,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0408108234405518,
      "theoretical_loss": 3.520358729323328,
      "tokens_seen": 1479475200
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027868605817452357,
      "loss": 2.9212,
      "theoretical_loss": 3.520358729323328,
      "tokens_seen": 1479475200
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027867602808425275,
      "loss": 2.9279,
      "theoretical_loss": 3.5203449441411423,
      "tokens_seen": 1479540736
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027866599799398193,
      "loss": 3.1469,
      "theoretical_loss": 3.5203311597405182,
      "tokens_seen": 1479606272
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002786559679037111,
      "loss": 2.6942,
      "theoretical_loss": 3.520317376121377,
      "tokens_seen": 1479671808
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027864593781344035,
      "loss": 2.7208,
      "theoretical_loss": 3.5203035932836393,
      "tokens_seen": 1479737344
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002786359077231695,
      "loss": 2.6946,
      "theoretical_loss": 3.520289811227226,
      "tokens_seen": 1479802880
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002786258776328987,
      "loss": 2.9046,
      "theoretical_loss": 3.520276029952059,
      "tokens_seen": 1479868416
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002786158475426279,
      "loss": 2.7582,
      "theoretical_loss": 3.520262249458059,
      "tokens_seen": 1479933952
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027860581745235707,
      "loss": 2.8369,
      "theoretical_loss": 3.5202484697451473,
      "tokens_seen": 1479999488
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027859578736208625,
      "loss": 2.9474,
      "theoretical_loss": 3.5202346908132442,
      "tokens_seen": 1480065024
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027858575727181544,
      "loss": 2.576,
      "theoretical_loss": 3.520220912662272,
      "tokens_seen": 1480130560
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002785757271815446,
      "loss": 2.5593,
      "theoretical_loss": 3.520207135292152,
      "tokens_seen": 1480196096
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027856569709127385,
      "loss": 2.8154,
      "theoretical_loss": 3.5201933587028034,
      "tokens_seen": 1480261632
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.000278555667001003,
      "loss": 2.6382,
      "theoretical_loss": 3.5201795828941496,
      "tokens_seen": 1480327168
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002785456369107322,
      "loss": 2.8783,
      "theoretical_loss": 3.5201658078661113,
      "tokens_seen": 1480392704
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027853560682046134,
      "loss": 3.0474,
      "theoretical_loss": 3.5201520336186087,
      "tokens_seen": 1480458240
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002785255767301906,
      "loss": 2.986,
      "theoretical_loss": 3.5201382601515636,
      "tokens_seen": 1480523776
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002785155466399198,
      "loss": 3.1115,
      "theoretical_loss": 3.5201244874648983,
      "tokens_seen": 1480589312
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027850551654964894,
      "loss": 3.1533,
      "theoretical_loss": 3.5201107155585323,
      "tokens_seen": 1480654848
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002784954864593782,
      "loss": 2.9877,
      "theoretical_loss": 3.520096944432388,
      "tokens_seen": 1480720384
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002784854563691073,
      "loss": 2.8325,
      "theoretical_loss": 3.5200831740863863,
      "tokens_seen": 1480785920
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027847542627883654,
      "loss": 2.9951,
      "theoretical_loss": 3.520069404520448,
      "tokens_seen": 1480851456
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002784653961885657,
      "loss": 2.6579,
      "theoretical_loss": 3.5200556357344954,
      "tokens_seen": 1480916992
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002784553660982949,
      "loss": 2.7737,
      "theoretical_loss": 3.520041867728449,
      "tokens_seen": 1480982528
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002784453360080241,
      "loss": 2.7079,
      "theoretical_loss": 3.5200281005022305,
      "tokens_seen": 1481048064
    },
    {
      "epoch": 0.45,
      "objective/train/advantage_avg": 0.494108647108078,
      "objective/train/docs_used": 839976,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.4886057376861572,
      "objective/train/original_loss": 2.4886062145233154,
      "objective/train/theoretical_loss": 3.5200143340557615,
      "objective/train/tokens_used": 1501573600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24419398605823517,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506521463394165,
      "objective/train/weighted_lm_loss": 2.614680767059326,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 3.5200143340557615,
      "tokens_seen": 1481113600
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002784353059177533,
      "loss": 2.6338,
      "theoretical_loss": 3.5200143340557615,
      "tokens_seen": 1481113600
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027842527582748244,
      "loss": 2.8889,
      "theoretical_loss": 3.520000568388962,
      "tokens_seen": 1481179136
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002784152457372117,
      "loss": 3.0379,
      "theoretical_loss": 3.519986803501755,
      "tokens_seen": 1481244672
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002784052156469408,
      "loss": 2.7265,
      "theoretical_loss": 3.5199730393940616,
      "tokens_seen": 1481310208
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027839518555667004,
      "loss": 2.8572,
      "theoretical_loss": 3.5199592760658023,
      "tokens_seen": 1481375744
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002783851554663992,
      "loss": 2.8268,
      "theoretical_loss": 3.5199455135168987,
      "tokens_seen": 1481441280
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002783751253761284,
      "loss": 2.9711,
      "theoretical_loss": 3.5199317517472726,
      "tokens_seen": 1481506816
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002783650952858576,
      "loss": 2.948,
      "theoretical_loss": 3.519917990756845,
      "tokens_seen": 1481572352
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027835506519558676,
      "loss": 2.8699,
      "theoretical_loss": 3.519904230545538,
      "tokens_seen": 1481637888
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027834503510531594,
      "loss": 2.9241,
      "theoretical_loss": 3.519890471113272,
      "tokens_seen": 1481703424
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002783350050150452,
      "loss": 2.9185,
      "theoretical_loss": 3.51987671245997,
      "tokens_seen": 1481768960
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002783249749247743,
      "loss": 2.9981,
      "theoretical_loss": 3.5198629545855518,
      "tokens_seen": 1481834496
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027831494483450354,
      "loss": 2.9286,
      "theoretical_loss": 3.5198491974899397,
      "tokens_seen": 1481900032
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027830491474423267,
      "loss": 2.6904,
      "theoretical_loss": 3.5198354411730546,
      "tokens_seen": 1481965568
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002782948846539619,
      "loss": 2.8436,
      "theoretical_loss": 3.5198216856348186,
      "tokens_seen": 1482031104
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002782848545636911,
      "loss": 2.6998,
      "theoretical_loss": 3.519807930875153,
      "tokens_seen": 1482096640
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027827482447342027,
      "loss": 2.895,
      "theoretical_loss": 3.5197941768939796,
      "tokens_seen": 1482162176
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027826479438314945,
      "loss": 2.8944,
      "theoretical_loss": 3.519780423691219,
      "tokens_seen": 1482227712
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002782547642928787,
      "loss": 2.8429,
      "theoretical_loss": 3.519766671266794,
      "tokens_seen": 1482293248
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002782447342026078,
      "loss": 2.9139,
      "theoretical_loss": 3.519752919620625,
      "tokens_seen": 1482358784
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027823470411233705,
      "loss": 3.0048,
      "theoretical_loss": 3.519739168752634,
      "tokens_seen": 1482424320
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027822467402206617,
      "loss": 2.9343,
      "theoretical_loss": 3.519725418662743,
      "tokens_seen": 1482489856
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002782146439317954,
      "loss": 2.7476,
      "theoretical_loss": 3.519711669350873,
      "tokens_seen": 1482555392
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002782046138415246,
      "loss": 2.7951,
      "theoretical_loss": 3.5196979208169457,
      "tokens_seen": 1482620928
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027819458375125377,
      "loss": 2.817,
      "theoretical_loss": 3.5196841730608828,
      "tokens_seen": 1482686464
    },
    {
      "epoch": 0.45,
      "objective/train/advantage_avg": 0.4953335225582123,
      "objective/train/docs_used": 840589,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.7920713424682617,
      "objective/train/original_loss": 2.7920713424682617,
      "objective/train/theoretical_loss": 3.519670426082606,
      "objective/train/tokens_used": 1503212000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2454260289669037,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050781011581421,
      "objective/train/weighted_lm_loss": 2.933943033218384,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0417574644088745,
      "theoretical_loss": 3.519670426082606,
      "tokens_seen": 1482752000
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027818455366098295,
      "loss": 2.7139,
      "theoretical_loss": 3.519670426082606,
      "tokens_seen": 1482752000
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027817452357071213,
      "loss": 2.7864,
      "theoretical_loss": 3.519656679882037,
      "tokens_seen": 1482817536
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002781644934804413,
      "loss": 2.6777,
      "theoretical_loss": 3.519642934459097,
      "tokens_seen": 1482883072
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027815446339017055,
      "loss": 2.7268,
      "theoretical_loss": 3.5196291898137075,
      "tokens_seen": 1482948608
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002781444332998997,
      "loss": 2.9345,
      "theoretical_loss": 3.5196154459457905,
      "tokens_seen": 1483014144
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002781344032096289,
      "loss": 2.8699,
      "theoretical_loss": 3.519601702855268,
      "tokens_seen": 1483079680
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002781243731193581,
      "loss": 2.7931,
      "theoretical_loss": 3.519587960542062,
      "tokens_seen": 1483145216
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002781143430290873,
      "loss": 2.8394,
      "theoretical_loss": 3.519574219006093,
      "tokens_seen": 1483210752
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027810431293881645,
      "loss": 2.7497,
      "theoretical_loss": 3.519560478247283,
      "tokens_seen": 1483276288
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027809428284854564,
      "loss": 2.9468,
      "theoretical_loss": 3.5195467382655545,
      "tokens_seen": 1483341824
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002780842527582748,
      "loss": 3.0853,
      "theoretical_loss": 3.5195329990608286,
      "tokens_seen": 1483407360
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027807422266800405,
      "loss": 2.7623,
      "theoretical_loss": 3.5195192606330266,
      "tokens_seen": 1483472896
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002780641925777332,
      "loss": 2.568,
      "theoretical_loss": 3.5195055229820715,
      "tokens_seen": 1483538432
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002780541624874624,
      "loss": 2.8941,
      "theoretical_loss": 3.519491786107884,
      "tokens_seen": 1483603968
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027804413239719154,
      "loss": 2.7514,
      "theoretical_loss": 3.5194780500103864,
      "tokens_seen": 1483669504
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002780341023069208,
      "loss": 2.905,
      "theoretical_loss": 3.5194643146895,
      "tokens_seen": 1483735040
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027802407221664996,
      "loss": 2.9541,
      "theoretical_loss": 3.519450580145147,
      "tokens_seen": 1483800576
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027801404212637914,
      "loss": 2.9329,
      "theoretical_loss": 3.519436846377249,
      "tokens_seen": 1483866112
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002780040120361083,
      "loss": 2.5151,
      "theoretical_loss": 3.519423113385728,
      "tokens_seen": 1483931648
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002779939819458375,
      "loss": 3.0664,
      "theoretical_loss": 3.5194093811705063,
      "tokens_seen": 1483997184
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002779839518555667,
      "loss": 2.67,
      "theoretical_loss": 3.5193956497315044,
      "tokens_seen": 1484062720
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002779739217652959,
      "loss": 2.7605,
      "theoretical_loss": 3.5193819190686453,
      "tokens_seen": 1484128256
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027796389167502504,
      "loss": 2.817,
      "theoretical_loss": 3.5193681891818507,
      "tokens_seen": 1484193792
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002779538615847543,
      "loss": 2.7495,
      "theoretical_loss": 3.5193544600710416,
      "tokens_seen": 1484259328
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027794383149448346,
      "loss": 2.7009,
      "theoretical_loss": 3.5193407317361407,
      "tokens_seen": 1484324864
    },
    {
      "epoch": 0.45,
      "objective/train/advantage_avg": 0.4938831627368927,
      "objective/train/docs_used": 841910,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.92267107963562,
      "objective/train/original_loss": 2.92267107963562,
      "objective/train/theoretical_loss": 3.5193270041770703,
      "objective/train/tokens_used": 1504850400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24399685859680176,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506285429000854,
      "objective/train/weighted_lm_loss": 3.07061505317688,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0408108234405518,
      "theoretical_loss": 3.5193270041770703,
      "tokens_seen": 1484390400
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027793380140421264,
      "loss": 2.8534,
      "theoretical_loss": 3.5193270041770703,
      "tokens_seen": 1484390400
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002779237713139418,
      "loss": 2.7009,
      "theoretical_loss": 3.5193132773937514,
      "tokens_seen": 1484455936
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.000277913741223671,
      "loss": 2.77,
      "theoretical_loss": 3.519299551386106,
      "tokens_seen": 1484521472
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002779037111334002,
      "loss": 2.6309,
      "theoretical_loss": 3.5192858261540567,
      "tokens_seen": 1484587008
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002778936810431294,
      "loss": 2.6975,
      "theoretical_loss": 3.519272101697525,
      "tokens_seen": 1484652544
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027788365095285855,
      "loss": 2.6455,
      "theoretical_loss": 3.519258378016432,
      "tokens_seen": 1484718080
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002778736208625878,
      "loss": 2.7907,
      "theoretical_loss": 3.5192446551107017,
      "tokens_seen": 1484783616
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002778635907723169,
      "loss": 2.8904,
      "theoretical_loss": 3.5192309329802542,
      "tokens_seen": 1484849152
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027785356068204614,
      "loss": 3.1855,
      "theoretical_loss": 3.519217211625013,
      "tokens_seen": 1484914688
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002778435305917753,
      "loss": 2.6429,
      "theoretical_loss": 3.5192034910448986,
      "tokens_seen": 1484980224
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002778335005015045,
      "loss": 2.8997,
      "theoretical_loss": 3.5191897712398337,
      "tokens_seen": 1485045760
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002778234704112337,
      "loss": 2.7888,
      "theoretical_loss": 3.5191760522097404,
      "tokens_seen": 1485111296
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027781344032096287,
      "loss": 2.5895,
      "theoretical_loss": 3.519162333954541,
      "tokens_seen": 1485176832
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027780341023069205,
      "loss": 2.7919,
      "theoretical_loss": 3.5191486164741566,
      "tokens_seen": 1485242368
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002777933801404213,
      "loss": 2.8209,
      "theoretical_loss": 3.51913489976851,
      "tokens_seen": 1485307904
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002777833500501504,
      "loss": 2.8902,
      "theoretical_loss": 3.519121183837523,
      "tokens_seen": 1485373440
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027777331995987965,
      "loss": 3.0317,
      "theoretical_loss": 3.5191074686811183,
      "tokens_seen": 1485438976
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002777632898696089,
      "loss": 2.8933,
      "theoretical_loss": 3.519093754299217,
      "tokens_seen": 1485504512
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.000277753259779338,
      "loss": 2.7821,
      "theoretical_loss": 3.519080040691742,
      "tokens_seen": 1485570048
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027774322968906725,
      "loss": 2.9059,
      "theoretical_loss": 3.5190663278586145,
      "tokens_seen": 1485635584
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027773319959879637,
      "loss": 2.8295,
      "theoretical_loss": 3.519052615799758,
      "tokens_seen": 1485701120
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002777231695085256,
      "loss": 2.8475,
      "theoretical_loss": 3.519038904515093,
      "tokens_seen": 1485766656
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002777131394182548,
      "loss": 2.7105,
      "theoretical_loss": 3.519025194004543,
      "tokens_seen": 1485832192
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027770310932798397,
      "loss": 2.56,
      "theoretical_loss": 3.5190114842680296,
      "tokens_seen": 1485897728
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027769307923771315,
      "loss": 2.9216,
      "theoretical_loss": 3.5189977753054746,
      "tokens_seen": 1485963264
    },
    {
      "epoch": 0.45,
      "objective/train/advantage_avg": 0.49575427174568176,
      "objective/train/docs_used": 842517,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.4610183238983154,
      "objective/train/original_loss": 2.4610180854797363,
      "objective/train/theoretical_loss": 3.5189840671168007,
      "objective/train/tokens_used": 1506488800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2458370327949524,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0508251190185547,
      "objective/train/weighted_lm_loss": 2.5861361026763916,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0396549701690674,
      "theoretical_loss": 3.5189840671168007,
      "tokens_seen": 1486028800
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027768304914744233,
      "loss": 2.706,
      "theoretical_loss": 3.5189840671168007,
      "tokens_seen": 1486028800
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002776730190571715,
      "loss": 2.8482,
      "theoretical_loss": 3.51897035970193,
      "tokens_seen": 1486094336
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027766298896690075,
      "loss": 2.753,
      "theoretical_loss": 3.518956653060785,
      "tokens_seen": 1486159872
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002776529588766299,
      "loss": 2.562,
      "theoretical_loss": 3.5189429471932874,
      "tokens_seen": 1486225408
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002776429287863591,
      "loss": 2.7452,
      "theoretical_loss": 3.518929242099359,
      "tokens_seen": 1486290944
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002776328986960883,
      "loss": 2.9037,
      "theoretical_loss": 3.5189155377789234,
      "tokens_seen": 1486356480
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002776228686058175,
      "loss": 2.8544,
      "theoretical_loss": 3.5189018342319014,
      "tokens_seen": 1486422016
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027761283851554665,
      "loss": 2.8381,
      "theoretical_loss": 3.5188881314582163,
      "tokens_seen": 1486487552
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027760280842527584,
      "loss": 2.7373,
      "theoretical_loss": 3.5188744294577896,
      "tokens_seen": 1486553088
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.000277592778335005,
      "loss": 2.9782,
      "theoretical_loss": 3.5188607282305444,
      "tokens_seen": 1486618624
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027758274824473425,
      "loss": 2.8194,
      "theoretical_loss": 3.518847027776402,
      "tokens_seen": 1486684160
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002775727181544634,
      "loss": 2.7718,
      "theoretical_loss": 3.518833328095286,
      "tokens_seen": 1486749696
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002775626880641926,
      "loss": 2.748,
      "theoretical_loss": 3.518819629187117,
      "tokens_seen": 1486815232
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027755265797392174,
      "loss": 2.6336,
      "theoretical_loss": 3.518805931051819,
      "tokens_seen": 1486880768
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.000277542627883651,
      "loss": 2.6226,
      "theoretical_loss": 3.5187922336893136,
      "tokens_seen": 1486946304
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027753259779338016,
      "loss": 2.8499,
      "theoretical_loss": 3.518778537099523,
      "tokens_seen": 1487011840
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027752256770310934,
      "loss": 2.7255,
      "theoretical_loss": 3.518764841282369,
      "tokens_seen": 1487077376
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002775125376128385,
      "loss": 2.8691,
      "theoretical_loss": 3.5187511462377756,
      "tokens_seen": 1487142912
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002775025075225677,
      "loss": 2.6785,
      "theoretical_loss": 3.518737451965664,
      "tokens_seen": 1487208448
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002774924774322969,
      "loss": 2.7698,
      "theoretical_loss": 3.5187237584659568,
      "tokens_seen": 1487273984
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002774824473420261,
      "loss": 2.7329,
      "theoretical_loss": 3.5187100657385764,
      "tokens_seen": 1487339520
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027747241725175524,
      "loss": 2.8394,
      "theoretical_loss": 3.5186963737834454,
      "tokens_seen": 1487405056
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002774623871614845,
      "loss": 2.7817,
      "theoretical_loss": 3.5186826826004856,
      "tokens_seen": 1487470592
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027745235707121366,
      "loss": 2.9814,
      "theoretical_loss": 3.5186689921896206,
      "tokens_seen": 1487536128
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027744232698094284,
      "loss": 2.7844,
      "theoretical_loss": 3.518655302550772,
      "tokens_seen": 1487601664
    },
    {
      "epoch": 0.45,
      "objective/train/advantage_avg": 0.49224743247032166,
      "objective/train/docs_used": 843832,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.540544033050537,
      "objective/train/original_loss": 2.540543556213379,
      "objective/train/theoretical_loss": 3.518641613683862,
      "objective/train/tokens_used": 1508127200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24242711067199707,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0504568815231323,
      "objective/train/weighted_lm_loss": 2.669233560562134,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0417574644088745,
      "theoretical_loss": 3.518641613683862,
      "tokens_seen": 1487667200
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.000277432296890672,
      "loss": 2.8336,
      "theoretical_loss": 3.518641613683862,
      "tokens_seen": 1487667200
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002774222668004012,
      "loss": 2.5034,
      "theoretical_loss": 3.518627925588814,
      "tokens_seen": 1487732736
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002774122367101304,
      "loss": 2.7193,
      "theoretical_loss": 3.5186142382655494,
      "tokens_seen": 1487798272
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002774022066198596,
      "loss": 2.7746,
      "theoretical_loss": 3.518600551713991,
      "tokens_seen": 1487863808
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027739217652958875,
      "loss": 2.7905,
      "theoretical_loss": 3.5185868659340627,
      "tokens_seen": 1487929344
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.000277382146439318,
      "loss": 2.6884,
      "theoretical_loss": 3.518573180925685,
      "tokens_seen": 1487994880
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002773721163490471,
      "loss": 2.9737,
      "theoretical_loss": 3.5185594966887814,
      "tokens_seen": 1488060416
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027736208625877634,
      "loss": 2.7981,
      "theoretical_loss": 3.518545813223275,
      "tokens_seen": 1488125952
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002773520561685055,
      "loss": 2.8604,
      "theoretical_loss": 3.5185321305290866,
      "tokens_seen": 1488191488
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002773420260782347,
      "loss": 2.8206,
      "theoretical_loss": 3.518518448606141,
      "tokens_seen": 1488257024
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002773319959879639,
      "loss": 2.8347,
      "theoretical_loss": 3.518504767454359,
      "tokens_seen": 1488322560
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027732196589769307,
      "loss": 2.7628,
      "theoretical_loss": 3.5184910870736634,
      "tokens_seen": 1488388096
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027731193580742225,
      "loss": 2.8339,
      "theoretical_loss": 3.5184774074639775,
      "tokens_seen": 1488453632
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002773019057171515,
      "loss": 2.9965,
      "theoretical_loss": 3.518463728625224,
      "tokens_seen": 1488519168
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002772918756268806,
      "loss": 2.6769,
      "theoretical_loss": 3.518450050557324,
      "tokens_seen": 1488584704
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027728184553660985,
      "loss": 2.9909,
      "theoretical_loss": 3.5184363732602026,
      "tokens_seen": 1488650240
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027727181544633903,
      "loss": 2.6858,
      "theoretical_loss": 3.51842269673378,
      "tokens_seen": 1488715776
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002772617853560682,
      "loss": 2.8485,
      "theoretical_loss": 3.5184090209779804,
      "tokens_seen": 1488781312
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002772517552657974,
      "loss": 2.8311,
      "theoretical_loss": 3.518395345992726,
      "tokens_seen": 1488846848
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027724172517552657,
      "loss": 2.795,
      "theoretical_loss": 3.518381671777939,
      "tokens_seen": 1488912384
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027723169508525575,
      "loss": 2.7227,
      "theoretical_loss": 3.518367998333543,
      "tokens_seen": 1488977920
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.000277221664994985,
      "loss": 3.037,
      "theoretical_loss": 3.51835432565946,
      "tokens_seen": 1489043456
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002772116349047141,
      "loss": 2.7338,
      "theoretical_loss": 3.5183406537556126,
      "tokens_seen": 1489108992
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027720160481444335,
      "loss": 2.9682,
      "theoretical_loss": 3.5183269826219243,
      "tokens_seen": 1489174528
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002771915747241725,
      "loss": 2.8422,
      "theoretical_loss": 3.518313312258317,
      "tokens_seen": 1489240064
    },
    {
      "epoch": 0.45,
      "objective/train/advantage_avg": 0.49331310391426086,
      "objective/train/docs_used": 844519,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.608154535293579,
      "objective/train/original_loss": 2.608154773712158,
      "objective/train/theoretical_loss": 3.5182996426647137,
      "objective/train/tokens_used": 1509765600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2434207946062088,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505685806274414,
      "objective/train/weighted_lm_loss": 2.740140914916992,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 3.5182996426647137,
      "tokens_seen": 1489305600
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002771815446339017,
      "loss": 2.7332,
      "theoretical_loss": 3.5182996426647137,
      "tokens_seen": 1489305600
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002771715145436309,
      "loss": 2.7736,
      "theoretical_loss": 3.5182859738410373,
      "tokens_seen": 1489371136
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002771614844533601,
      "loss": 2.8592,
      "theoretical_loss": 3.5182723057872107,
      "tokens_seen": 1489436672
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027715145436308926,
      "loss": 2.8561,
      "theoretical_loss": 3.5182586385031565,
      "tokens_seen": 1489502208
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002771414242728185,
      "loss": 2.8569,
      "theoretical_loss": 3.518244971988797,
      "tokens_seen": 1489567744
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002771313941825476,
      "loss": 2.8811,
      "theoretical_loss": 3.518231306244056,
      "tokens_seen": 1489633280
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027712136409227685,
      "loss": 3.0836,
      "theoretical_loss": 3.5182176412688553,
      "tokens_seen": 1489698816
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.000277111334002006,
      "loss": 2.6954,
      "theoretical_loss": 3.518203977063118,
      "tokens_seen": 1489764352
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002771013039117352,
      "loss": 2.8123,
      "theoretical_loss": 3.5181903136267674,
      "tokens_seen": 1489829888
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002770912738214644,
      "loss": 3.0322,
      "theoretical_loss": 3.5181766509597256,
      "tokens_seen": 1489895424
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002770812437311936,
      "loss": 2.8169,
      "theoretical_loss": 3.5181629890619166,
      "tokens_seen": 1489960960
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027707121364092276,
      "loss": 3.0085,
      "theoretical_loss": 3.518149327933262,
      "tokens_seen": 1490026496
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027706118355065194,
      "loss": 2.9629,
      "theoretical_loss": 3.5181356675736852,
      "tokens_seen": 1490092032
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002770511534603811,
      "loss": 2.6355,
      "theoretical_loss": 3.5181220079831093,
      "tokens_seen": 1490157568
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027704112337011036,
      "loss": 2.9427,
      "theoretical_loss": 3.518108349161457,
      "tokens_seen": 1490223104
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002770310932798395,
      "loss": 2.75,
      "theoretical_loss": 3.518094691108651,
      "tokens_seen": 1490288640
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002770210631895687,
      "loss": 2.8822,
      "theoretical_loss": 3.5180810338246142,
      "tokens_seen": 1490354176
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002770110330992979,
      "loss": 2.7816,
      "theoretical_loss": 3.5180673773092694,
      "tokens_seen": 1490419712
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002770010030090271,
      "loss": 2.9486,
      "theoretical_loss": 3.5180537215625405,
      "tokens_seen": 1490485248
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002769909729187563,
      "loss": 2.7459,
      "theoretical_loss": 3.5180400665843496,
      "tokens_seen": 1490550784
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027698094282848544,
      "loss": 2.6925,
      "theoretical_loss": 3.5180264123746197,
      "tokens_seen": 1490616320
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002769709127382147,
      "loss": 2.4546,
      "theoretical_loss": 3.518012758933274,
      "tokens_seen": 1490681856
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027696088264794386,
      "loss": 2.8098,
      "theoretical_loss": 3.517999106260236,
      "tokens_seen": 1490747392
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027695085255767304,
      "loss": 2.8075,
      "theoretical_loss": 3.5179854543554274,
      "tokens_seen": 1490812928
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002769408224674022,
      "loss": 2.7732,
      "theoretical_loss": 3.5179718032187717,
      "tokens_seen": 1490878464
    },
    {
      "epoch": 0.45,
      "objective/train/advantage_avg": 0.4924141466617584,
      "objective/train/docs_used": 845874,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.6872446537017822,
      "objective/train/original_loss": 2.687244415283203,
      "objective/train/theoretical_loss": 3.517958152850192,
      "objective/train/tokens_used": 1511404000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2425483763217926,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0504742860794067,
      "objective/train/weighted_lm_loss": 2.8228018283843994,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0451053380966187,
      "theoretical_loss": 3.517958152850192,
      "tokens_seen": 1490944000
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002769307923771314,
      "loss": 2.7473,
      "theoretical_loss": 3.517958152850192,
      "tokens_seen": 1490944000
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002769207622868606,
      "loss": 2.7541,
      "theoretical_loss": 3.5179445032496126,
      "tokens_seen": 1491009536
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002769107321965898,
      "loss": 2.7221,
      "theoretical_loss": 3.5179308544169543,
      "tokens_seen": 1491075072
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027690070210631895,
      "loss": 2.7543,
      "theoretical_loss": 3.5179172063521413,
      "tokens_seen": 1491140608
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002768906720160482,
      "loss": 2.9228,
      "theoretical_loss": 3.517903559055097,
      "tokens_seen": 1491206144
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002768806419257773,
      "loss": 2.8788,
      "theoretical_loss": 3.517889912525744,
      "tokens_seen": 1491271680
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027687061183550655,
      "loss": 2.8463,
      "theoretical_loss": 3.5178762667640053,
      "tokens_seen": 1491337216
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002768605817452357,
      "loss": 2.9267,
      "theoretical_loss": 3.5178626217698037,
      "tokens_seen": 1491402752
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002768505516549649,
      "loss": 2.9862,
      "theoretical_loss": 3.5178489775430632,
      "tokens_seen": 1491468288
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002768405215646941,
      "loss": 2.9401,
      "theoretical_loss": 3.5178353340837063,
      "tokens_seen": 1491533824
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027683049147442327,
      "loss": 2.8835,
      "theoretical_loss": 3.5178216913916565,
      "tokens_seen": 1491599360
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027682046138415245,
      "loss": 2.6984,
      "theoretical_loss": 3.5178080494668365,
      "tokens_seen": 1491664896
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002768104312938817,
      "loss": 2.951,
      "theoretical_loss": 3.51779440830917,
      "tokens_seen": 1491730432
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002768004012036108,
      "loss": 2.7927,
      "theoretical_loss": 3.5177807679185795,
      "tokens_seen": 1491795968
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027679037111334005,
      "loss": 3.0288,
      "theoretical_loss": 3.517767128294988,
      "tokens_seen": 1491861504
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027678034102306923,
      "loss": 2.7122,
      "theoretical_loss": 3.51775348943832,
      "tokens_seen": 1491927040
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002767703109327984,
      "loss": 2.6161,
      "theoretical_loss": 3.517739851348497,
      "tokens_seen": 1491992576
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002767602808425276,
      "loss": 2.8129,
      "theoretical_loss": 3.517726214025444,
      "tokens_seen": 1492058112
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027675025075225677,
      "loss": 2.8993,
      "theoretical_loss": 3.5177125774690827,
      "tokens_seen": 1492123648
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027674022066198595,
      "loss": 2.4078,
      "theoretical_loss": 3.517698941679337,
      "tokens_seen": 1492189184
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002767301905717152,
      "loss": 2.7668,
      "theoretical_loss": 3.51768530665613,
      "tokens_seen": 1492254720
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002767201604814443,
      "loss": 2.9139,
      "theoretical_loss": 3.517671672399385,
      "tokens_seen": 1492320256
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027671013039117355,
      "loss": 2.4712,
      "theoretical_loss": 3.5176580389090253,
      "tokens_seen": 1492385792
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002767001003009027,
      "loss": 2.9157,
      "theoretical_loss": 3.517644406184974,
      "tokens_seen": 1492451328
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002766900702106319,
      "loss": 2.7116,
      "theoretical_loss": 3.5176307742271544,
      "tokens_seen": 1492516864
    },
    {
      "epoch": 0.45,
      "objective/train/advantage_avg": 0.4946475923061371,
      "objective/train/docs_used": 846561,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.4950332641601562,
      "objective/train/original_loss": 2.495033025741577,
      "objective/train/theoretical_loss": 3.5176171430354897,
      "objective/train/tokens_used": 1513042400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24474520981311798,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507088899612427,
      "objective/train/weighted_lm_loss": 2.621471881866455,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0425469875335693,
      "theoretical_loss": 3.5176171430354897,
      "tokens_seen": 1492582400
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002766800401203611,
      "loss": 2.8168,
      "theoretical_loss": 3.5176171430354897,
      "tokens_seen": 1492582400
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002766700100300903,
      "loss": 2.9724,
      "theoretical_loss": 3.517603512609904,
      "tokens_seen": 1492647936
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027665997993981946,
      "loss": 2.5971,
      "theoretical_loss": 3.517589882950319,
      "tokens_seen": 1492713472
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002766499498495487,
      "loss": 2.8229,
      "theoretical_loss": 3.51757625405666,
      "tokens_seen": 1492779008
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002766399197592778,
      "loss": 2.9271,
      "theoretical_loss": 3.517562625928849,
      "tokens_seen": 1492844544
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027662988966900705,
      "loss": 3.0458,
      "theoretical_loss": 3.51754899856681,
      "tokens_seen": 1492910080
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002766198595787362,
      "loss": 2.8731,
      "theoretical_loss": 3.5175353719704656,
      "tokens_seen": 1492975616
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002766098294884654,
      "loss": 2.7911,
      "theoretical_loss": 3.5175217461397397,
      "tokens_seen": 1493041152
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002765997993981946,
      "loss": 2.8763,
      "theoretical_loss": 3.5175081210745556,
      "tokens_seen": 1493106688
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002765897693079238,
      "loss": 2.7484,
      "theoretical_loss": 3.517494496774837,
      "tokens_seen": 1493172224
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027657973921765296,
      "loss": 2.8225,
      "theoretical_loss": 3.5174808732405065,
      "tokens_seen": 1493237760
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027656970912738214,
      "loss": 2.8557,
      "theoretical_loss": 3.5174672504714883,
      "tokens_seen": 1493303296
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002765596790371113,
      "loss": 2.8647,
      "theoretical_loss": 3.5174536284677056,
      "tokens_seen": 1493368832
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027654964894684056,
      "loss": 2.6997,
      "theoretical_loss": 3.517440007229082,
      "tokens_seen": 1493434368
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002765396188565697,
      "loss": 2.5396,
      "theoretical_loss": 3.51742638675554,
      "tokens_seen": 1493499904
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002765295887662989,
      "loss": 2.6824,
      "theoretical_loss": 3.5174127670470043,
      "tokens_seen": 1493565440
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027651955867602805,
      "loss": 2.6656,
      "theoretical_loss": 3.5173991481033977,
      "tokens_seen": 1493630976
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002765095285857573,
      "loss": 2.7568,
      "theoretical_loss": 3.5173855299246437,
      "tokens_seen": 1493696512
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027649949849548646,
      "loss": 2.5186,
      "theoretical_loss": 3.517371912510666,
      "tokens_seen": 1493762048
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027648946840521564,
      "loss": 2.8142,
      "theoretical_loss": 3.5173582958613885,
      "tokens_seen": 1493827584
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002764794383149448,
      "loss": 3.1345,
      "theoretical_loss": 3.5173446799767336,
      "tokens_seen": 1493893120
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027646940822467406,
      "loss": 2.6334,
      "theoretical_loss": 3.5173310648566254,
      "tokens_seen": 1493958656
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002764593781344032,
      "loss": 2.7047,
      "theoretical_loss": 3.517317450500988,
      "tokens_seen": 1494024192
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002764493480441324,
      "loss": 2.7827,
      "theoretical_loss": 3.5173038369097442,
      "tokens_seen": 1494089728
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027643931795386155,
      "loss": 2.621,
      "theoretical_loss": 3.517290224082817,
      "tokens_seen": 1494155264
    },
    {
      "epoch": 0.45,
      "objective/train/advantage_avg": 0.4946543276309967,
      "objective/train/docs_used": 848095,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.9171464443206787,
      "objective/train/original_loss": 2.917146921157837,
      "objective/train/theoretical_loss": 3.517276612020132,
      "objective/train/tokens_used": 1514680800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24474447965621948,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.05070960521698,
      "objective/train/weighted_lm_loss": 3.0651144981384277,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0408108234405518,
      "theoretical_loss": 3.517276612020132,
      "tokens_seen": 1494220800
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002764292878635908,
      "loss": 2.8567,
      "theoretical_loss": 3.517276612020132,
      "tokens_seen": 1494220800
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027641925777331997,
      "loss": 2.8627,
      "theoretical_loss": 3.517263000721611,
      "tokens_seen": 1494286336
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027640922768304915,
      "loss": 2.8665,
      "theoretical_loss": 3.517249390187178,
      "tokens_seen": 1494351872
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027639919759277833,
      "loss": 2.7956,
      "theoretical_loss": 3.517235780416757,
      "tokens_seen": 1494417408
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002763891675025075,
      "loss": 2.9042,
      "theoretical_loss": 3.517222171410271,
      "tokens_seen": 1494482944
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002763791374122367,
      "loss": 2.8808,
      "theoretical_loss": 3.517208563167644,
      "tokens_seen": 1494548480
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002763691073219659,
      "loss": 2.8106,
      "theoretical_loss": 3.5171949556888,
      "tokens_seen": 1494614016
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027635907723169505,
      "loss": 2.6269,
      "theoretical_loss": 3.5171813489736614,
      "tokens_seen": 1494679552
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002763490471414243,
      "loss": 2.8179,
      "theoretical_loss": 3.5171677430221533,
      "tokens_seen": 1494745088
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002763390170511534,
      "loss": 2.8745,
      "theoretical_loss": 3.517154137834199,
      "tokens_seen": 1494810624
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027632898696088265,
      "loss": 2.7384,
      "theoretical_loss": 3.517140533409721,
      "tokens_seen": 1494876160
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027631895687061183,
      "loss": 2.8779,
      "theoretical_loss": 3.5171269297486445,
      "tokens_seen": 1494941696
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.000276308926780341,
      "loss": 2.5625,
      "theoretical_loss": 3.5171133268508927,
      "tokens_seen": 1495007232
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002762988966900702,
      "loss": 2.8681,
      "theoretical_loss": 3.517099724716389,
      "tokens_seen": 1495072768
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027628886659979943,
      "loss": 2.7809,
      "theoretical_loss": 3.5170861233450577,
      "tokens_seen": 1495138304
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027627883650952856,
      "loss": 2.7808,
      "theoretical_loss": 3.5170725227368216,
      "tokens_seen": 1495203840
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002762688064192578,
      "loss": 2.7128,
      "theoretical_loss": 3.5170589228916054,
      "tokens_seen": 1495269376
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027625877632898697,
      "loss": 2.7504,
      "theoretical_loss": 3.5170453238093327,
      "tokens_seen": 1495334912
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027624874623871615,
      "loss": 2.6079,
      "theoretical_loss": 3.517031725489926,
      "tokens_seen": 1495400448
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002762387161484454,
      "loss": 2.7319,
      "theoretical_loss": 3.5170181279333113,
      "tokens_seen": 1495465984
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002762286860581745,
      "loss": 2.6952,
      "theoretical_loss": 3.5170045311394107,
      "tokens_seen": 1495531520
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027621865596790375,
      "loss": 2.6577,
      "theoretical_loss": 3.516990935108148,
      "tokens_seen": 1495597056
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002762086258776329,
      "loss": 2.6588,
      "theoretical_loss": 3.5169773398394484,
      "tokens_seen": 1495662592
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002761985957873621,
      "loss": 2.9486,
      "theoretical_loss": 3.5169637453332343,
      "tokens_seen": 1495728128
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002761885656970913,
      "loss": 2.8016,
      "theoretical_loss": 3.5169501515894304,
      "tokens_seen": 1495793664
    },
    {
      "epoch": 0.45,
      "objective/train/advantage_avg": 0.49593737721443176,
      "objective/train/docs_used": 848766,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.595860242843628,
      "objective/train/original_loss": 2.5958597660064697,
      "objective/train/theoretical_loss": 3.5169365586079597,
      "objective/train/tokens_used": 1516319200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24601034820079803,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050844430923462,
      "objective/train/weighted_lm_loss": 2.727834939956665,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0363596677780151,
      "theoretical_loss": 3.5169365586079597,
      "tokens_seen": 1495859200
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002761785356068205,
      "loss": 2.9321,
      "theoretical_loss": 3.5169365586079597,
      "tokens_seen": 1495859200
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027616850551654966,
      "loss": 2.8372,
      "theoretical_loss": 3.516922966388747,
      "tokens_seen": 1495924736
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002761584754262789,
      "loss": 2.804,
      "theoretical_loss": 3.516909374931715,
      "tokens_seen": 1495990272
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.000276148445336008,
      "loss": 2.7017,
      "theoretical_loss": 3.516895784236789,
      "tokens_seen": 1496055808
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027613841524573725,
      "loss": 2.7712,
      "theoretical_loss": 3.516882194303892,
      "tokens_seen": 1496121344
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002761283851554664,
      "loss": 2.8563,
      "theoretical_loss": 3.516868605132948,
      "tokens_seen": 1496186880
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002761183550651956,
      "loss": 2.7598,
      "theoretical_loss": 3.516855016723881,
      "tokens_seen": 1496252416
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002761083249749248,
      "loss": 2.679,
      "theoretical_loss": 3.516841429076615,
      "tokens_seen": 1496317952
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.000276098294884654,
      "loss": 2.9091,
      "theoretical_loss": 3.5168278421910735,
      "tokens_seen": 1496383488
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027608826479438316,
      "loss": 2.7514,
      "theoretical_loss": 3.516814256067181,
      "tokens_seen": 1496449024
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027607823470411234,
      "loss": 2.837,
      "theoretical_loss": 3.5168006707048614,
      "tokens_seen": 1496514560
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002760682046138415,
      "loss": 2.9183,
      "theoretical_loss": 3.516787086104038,
      "tokens_seen": 1496580096
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027605817452357076,
      "loss": 3.055,
      "theoretical_loss": 3.5167735022646354,
      "tokens_seen": 1496645632
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002760481444332999,
      "loss": 2.7467,
      "theoretical_loss": 3.5167599191865775,
      "tokens_seen": 1496711168
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002760381143430291,
      "loss": 2.7718,
      "theoretical_loss": 3.5167463368697884,
      "tokens_seen": 1496776704
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027602808425275825,
      "loss": 2.8339,
      "theoretical_loss": 3.516732755314192,
      "tokens_seen": 1496842240
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002760180541624875,
      "loss": 2.8252,
      "theoretical_loss": 3.5167191745197117,
      "tokens_seen": 1496907776
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027600802407221666,
      "loss": 2.7266,
      "theoretical_loss": 3.516705594486272,
      "tokens_seen": 1496973312
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027599799398194584,
      "loss": 2.8111,
      "theoretical_loss": 3.5166920152137973,
      "tokens_seen": 1497038848
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.000275987963891675,
      "loss": 2.6951,
      "theoretical_loss": 3.516678436702212,
      "tokens_seen": 1497104384
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027597793380140426,
      "loss": 2.9459,
      "theoretical_loss": 3.5166648589514384,
      "tokens_seen": 1497169920
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002759679037111334,
      "loss": 2.857,
      "theoretical_loss": 3.516651281961402,
      "tokens_seen": 1497235456
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002759578736208626,
      "loss": 2.7191,
      "theoretical_loss": 3.516637705732027,
      "tokens_seen": 1497300992
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027594784353059175,
      "loss": 2.8244,
      "theoretical_loss": 3.516624130263237,
      "tokens_seen": 1497366528
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.000275937813440321,
      "loss": 2.7456,
      "theoretical_loss": 3.516610555554956,
      "tokens_seen": 1497432064
    },
    {
      "epoch": 0.45,
      "objective/train/advantage_avg": 0.494194358587265,
      "objective/train/docs_used": 849888,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.7602741718292236,
      "objective/train/original_loss": 2.7602739334106445,
      "objective/train/theoretical_loss": 3.5165969816071083,
      "objective/train/tokens_used": 1517957600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24431748688220978,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050661325454712,
      "objective/train/weighted_lm_loss": 2.8998870849609375,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0432155132293701,
      "theoretical_loss": 3.5165969816071083,
      "tokens_seen": 1497497600
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027592778335005017,
      "loss": 2.7914,
      "theoretical_loss": 3.5165969816071083,
      "tokens_seen": 1497497600
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027591775325977935,
      "loss": 2.6284,
      "theoretical_loss": 3.516583408419618,
      "tokens_seen": 1497563136
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027590772316950853,
      "loss": 2.817,
      "theoretical_loss": 3.516569835992409,
      "tokens_seen": 1497628672
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002758976930792377,
      "loss": 2.4954,
      "theoretical_loss": 3.5165562643254056,
      "tokens_seen": 1497694208
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002758876629889669,
      "loss": 2.7096,
      "theoretical_loss": 3.5165426934185327,
      "tokens_seen": 1497759744
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002758776328986961,
      "loss": 2.8658,
      "theoretical_loss": 3.516529123271713,
      "tokens_seen": 1497825280
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027586760280842525,
      "loss": 2.7871,
      "theoretical_loss": 3.5165155538848722,
      "tokens_seen": 1497890816
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002758575727181545,
      "loss": 2.8375,
      "theoretical_loss": 3.5165019852579333,
      "tokens_seen": 1497956352
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002758475426278836,
      "loss": 2.8608,
      "theoretical_loss": 3.5164884173908213,
      "tokens_seen": 1498021888
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027583751253761285,
      "loss": 2.8123,
      "theoretical_loss": 3.51647485028346,
      "tokens_seen": 1498087424
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027582748244734203,
      "loss": 2.8773,
      "theoretical_loss": 3.516461283935774,
      "tokens_seen": 1498152960
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002758174523570712,
      "loss": 2.7276,
      "theoretical_loss": 3.5164477183476865,
      "tokens_seen": 1498218496
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002758074222668004,
      "loss": 2.7996,
      "theoretical_loss": 3.516434153519123,
      "tokens_seen": 1498284032
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027579739217652963,
      "loss": 2.8473,
      "theoretical_loss": 3.516420589450007,
      "tokens_seen": 1498349568
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027578736208625876,
      "loss": 2.7191,
      "theoretical_loss": 3.5164070261402633,
      "tokens_seen": 1498415104
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.000275777331995988,
      "loss": 2.8368,
      "theoretical_loss": 3.516393463589816,
      "tokens_seen": 1498480640
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002757673019057171,
      "loss": 2.7668,
      "theoretical_loss": 3.5163799017985893,
      "tokens_seen": 1498546176
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027575727181544635,
      "loss": 2.5992,
      "theoretical_loss": 3.516366340766507,
      "tokens_seen": 1498611712
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027574724172517554,
      "loss": 2.6117,
      "theoretical_loss": 3.516352780493494,
      "tokens_seen": 1498677248
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002757372116349047,
      "loss": 2.7909,
      "theoretical_loss": 3.516339220979475,
      "tokens_seen": 1498742784
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002757271815446339,
      "loss": 2.9897,
      "theoretical_loss": 3.5163256622243733,
      "tokens_seen": 1498808320
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002757171514543631,
      "loss": 2.8735,
      "theoretical_loss": 3.5163121042281142,
      "tokens_seen": 1498873856
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027570712136409226,
      "loss": 2.5287,
      "theoretical_loss": 3.5162985469906216,
      "tokens_seen": 1498939392
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002756970912738215,
      "loss": 2.8052,
      "theoretical_loss": 3.51628499051182,
      "tokens_seen": 1499004928
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002756870611835506,
      "loss": 3.0101,
      "theoretical_loss": 3.5162714347916335,
      "tokens_seen": 1499070464
    },
    {
      "epoch": 0.45,
      "objective/train/advantage_avg": 0.49608859419822693,
      "objective/train/docs_used": 850610,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.8257594108581543,
      "objective/train/original_loss": 2.8257596492767334,
      "objective/train/theoretical_loss": 3.516257879829986,
      "objective/train/tokens_used": 1519596000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24615038931369781,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0508602857589722,
      "objective/train/weighted_lm_loss": 2.9694602489471436,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 3.516257879829986,
      "tokens_seen": 1499136000
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027567703109327986,
      "loss": 2.975,
      "theoretical_loss": 3.516257879829986,
      "tokens_seen": 1499136000
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.000275667001003009,
      "loss": 2.8247,
      "theoretical_loss": 3.5162443256268032,
      "tokens_seen": 1499201536
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002756569709127382,
      "loss": 2.8748,
      "theoretical_loss": 3.516230772182009,
      "tokens_seen": 1499267072
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002756469408224674,
      "loss": 2.9948,
      "theoretical_loss": 3.5162172194955277,
      "tokens_seen": 1499332608
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002756369107321966,
      "loss": 2.9495,
      "theoretical_loss": 3.516203667567283,
      "tokens_seen": 1499398144
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027562688064192576,
      "loss": 2.6417,
      "theoretical_loss": 3.516190116397201,
      "tokens_seen": 1499463680
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.000275616850551655,
      "loss": 2.7067,
      "theoretical_loss": 3.516176565985205,
      "tokens_seen": 1499529216
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002756068204613841,
      "loss": 2.7804,
      "theoretical_loss": 3.5161630163312196,
      "tokens_seen": 1499594752
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027559679037111336,
      "loss": 2.6813,
      "theoretical_loss": 3.516149467435169,
      "tokens_seen": 1499660288
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002755867602808425,
      "loss": 2.5743,
      "theoretical_loss": 3.5161359192969783,
      "tokens_seen": 1499725824
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002755767301905717,
      "loss": 2.8111,
      "theoretical_loss": 3.516122371916572,
      "tokens_seen": 1499791360
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002755667001003009,
      "loss": 2.8351,
      "theoretical_loss": 3.516108825293874,
      "tokens_seen": 1499856896
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002755566700100301,
      "loss": 2.8449,
      "theoretical_loss": 3.516095279428809,
      "tokens_seen": 1499922432
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027554663991975927,
      "loss": 2.8085,
      "theoretical_loss": 3.516081734321302,
      "tokens_seen": 1499987968
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027553660982948845,
      "loss": 2.898,
      "theoretical_loss": 3.516068189971277,
      "tokens_seen": 1500053504
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027552657973921763,
      "loss": 2.6141,
      "theoretical_loss": 3.516054646378659,
      "tokens_seen": 1500119040
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027551654964894686,
      "loss": 2.8261,
      "theoretical_loss": 3.5160411035433725,
      "tokens_seen": 1500184576
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027550651955867604,
      "loss": 2.9402,
      "theoretical_loss": 3.516027561465341,
      "tokens_seen": 1500250112
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002754964894684052,
      "loss": 2.8036,
      "theoretical_loss": 3.5160140201444907,
      "tokens_seen": 1500315648
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027548645937813446,
      "loss": 2.7347,
      "theoretical_loss": 3.5160004795807454,
      "tokens_seen": 1500381184
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002754764292878636,
      "loss": 2.8021,
      "theoretical_loss": 3.5159869397740295,
      "tokens_seen": 1500446720
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002754663991975928,
      "loss": 2.8849,
      "theoretical_loss": 3.5159734007242682,
      "tokens_seen": 1500512256
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027545636910732195,
      "loss": 2.9028,
      "theoretical_loss": 3.5159598624313855,
      "tokens_seen": 1500577792
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002754463390170512,
      "loss": 2.5679,
      "theoretical_loss": 3.5159463248953067,
      "tokens_seen": 1500643328
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027543630892678037,
      "loss": 2.9251,
      "theoretical_loss": 3.515932788115956,
      "tokens_seen": 1500708864
    },
    {
      "epoch": 0.45,
      "objective/train/advantage_avg": 0.4952465891838074,
      "objective/train/docs_used": 851359,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.009622573852539,
      "objective/train/original_loss": 3.00962233543396,
      "objective/train/theoretical_loss": 3.5159192520932576,
      "objective/train/tokens_used": 1521234400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24531859159469604,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507718324661255,
      "objective/train/weighted_lm_loss": 3.1624574661254883,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0425469875335693,
      "theoretical_loss": 3.5159192520932576,
      "tokens_seen": 1500774400
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027542627883650955,
      "loss": 2.8632,
      "theoretical_loss": 3.5159192520932576,
      "tokens_seen": 1500774400
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027541624874623873,
      "loss": 2.9026,
      "theoretical_loss": 3.5159057168271373,
      "tokens_seen": 1500839936
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002754062186559679,
      "loss": 2.9022,
      "theoretical_loss": 3.5158921823175184,
      "tokens_seen": 1500905472
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002753961885656971,
      "loss": 2.6389,
      "theoretical_loss": 3.5158786485643274,
      "tokens_seen": 1500971008
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002753861584754263,
      "loss": 2.6029,
      "theoretical_loss": 3.5158651155674874,
      "tokens_seen": 1501036544
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027537612838515545,
      "loss": 2.6814,
      "theoretical_loss": 3.5158515833269237,
      "tokens_seen": 1501102080
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002753660982948847,
      "loss": 2.7738,
      "theoretical_loss": 3.515838051842561,
      "tokens_seen": 1501167616
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002753560682046138,
      "loss": 2.6688,
      "theoretical_loss": 3.5158245211143244,
      "tokens_seen": 1501233152
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027534603811434305,
      "loss": 2.6953,
      "theoretical_loss": 3.515810991142138,
      "tokens_seen": 1501298688
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.00027533600802407223,
      "loss": 2.585,
      "theoretical_loss": 3.515797461925927,
      "tokens_seen": 1501364224
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002753259779338014,
      "loss": 2.5951,
      "theoretical_loss": 3.5157839334656154,
      "tokens_seen": 1501429760
    },
    {
      "epoch": 0.45,
      "learning_rate": 0.0002753159478435306,
      "loss": 2.9264,
      "theoretical_loss": 3.515770405761129,
      "tokens_seen": 1501495296
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027530591775325983,
      "loss": 2.9689,
      "theoretical_loss": 3.5157568788123923,
      "tokens_seen": 1501560832
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027529588766298896,
      "loss": 2.8783,
      "theoretical_loss": 3.51574335261933,
      "tokens_seen": 1501626368
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002752858575727182,
      "loss": 2.8693,
      "theoretical_loss": 3.5157298271818664,
      "tokens_seen": 1501691904
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002752758274824473,
      "loss": 2.9794,
      "theoretical_loss": 3.5157163024999267,
      "tokens_seen": 1501757440
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027526579739217655,
      "loss": 2.756,
      "theoretical_loss": 3.5157027785734365,
      "tokens_seen": 1501822976
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027525576730190574,
      "loss": 2.7247,
      "theoretical_loss": 3.5156892554023194,
      "tokens_seen": 1501888512
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002752457372116349,
      "loss": 2.9641,
      "theoretical_loss": 3.5156757329865007,
      "tokens_seen": 1501954048
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002752357071213641,
      "loss": 2.8435,
      "theoretical_loss": 3.515662211325906,
      "tokens_seen": 1502019584
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002752256770310933,
      "loss": 2.8622,
      "theoretical_loss": 3.515648690420459,
      "tokens_seen": 1502085120
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027521564694082246,
      "loss": 2.8803,
      "theoretical_loss": 3.5156351702700857,
      "tokens_seen": 1502150656
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002752056168505517,
      "loss": 2.7767,
      "theoretical_loss": 3.5156216508747096,
      "tokens_seen": 1502216192
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002751955867602808,
      "loss": 2.9612,
      "theoretical_loss": 3.5156081322342567,
      "tokens_seen": 1502281728
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027518555667001006,
      "loss": 2.7912,
      "theoretical_loss": 3.515594614348652,
      "tokens_seen": 1502347264
    },
    {
      "epoch": 0.46,
      "objective/train/advantage_avg": 0.4945014715194702,
      "objective/train/docs_used": 852603,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.707899570465088,
      "objective/train/original_loss": 2.707899570465088,
      "objective/train/theoretical_loss": 3.5155810972178196,
      "objective/train/tokens_used": 1522872800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.244576558470726,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050693392753601,
      "objective/train/weighted_lm_loss": 2.845108985900879,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0432155132293701,
      "theoretical_loss": 3.5155810972178196,
      "tokens_seen": 1502412800
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002751755265797392,
      "loss": 2.837,
      "theoretical_loss": 3.5155810972178196,
      "tokens_seen": 1502412800
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002751654964894684,
      "loss": 2.753,
      "theoretical_loss": 3.515567580841685,
      "tokens_seen": 1502478336
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002751554663991976,
      "loss": 2.8523,
      "theoretical_loss": 3.515554065220173,
      "tokens_seen": 1502543872
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002751454363089268,
      "loss": 2.721,
      "theoretical_loss": 3.515540550353209,
      "tokens_seen": 1502609408
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027513540621865596,
      "loss": 2.832,
      "theoretical_loss": 3.515527036240717,
      "tokens_seen": 1502674944
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002751253761283852,
      "loss": 2.9375,
      "theoretical_loss": 3.515513522882623,
      "tokens_seen": 1502740480
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002751153460381143,
      "loss": 2.5912,
      "theoretical_loss": 3.515500010278851,
      "tokens_seen": 1502806016
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027510531594784356,
      "loss": 2.7732,
      "theoretical_loss": 3.5154864984293273,
      "tokens_seen": 1502871552
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002750952858575727,
      "loss": 2.9864,
      "theoretical_loss": 3.5154729873339754,
      "tokens_seen": 1502937088
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002750852557673019,
      "loss": 2.8663,
      "theoretical_loss": 3.5154594769927217,
      "tokens_seen": 1503002624
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002750752256770311,
      "loss": 2.7281,
      "theoretical_loss": 3.5154459674054896,
      "tokens_seen": 1503068160
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002750651955867603,
      "loss": 3.0194,
      "theoretical_loss": 3.515432458572206,
      "tokens_seen": 1503133696
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027505516549648947,
      "loss": 2.8942,
      "theoretical_loss": 3.515418950492795,
      "tokens_seen": 1503199232
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027504513540621865,
      "loss": 3.1161,
      "theoretical_loss": 3.5154054431671815,
      "tokens_seen": 1503264768
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027503510531594783,
      "loss": 2.9798,
      "theoretical_loss": 3.5153919365952913,
      "tokens_seen": 1503330304
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027502507522567706,
      "loss": 3.0863,
      "theoretical_loss": 3.515378430777049,
      "tokens_seen": 1503395840
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002750150451354062,
      "loss": 2.6986,
      "theoretical_loss": 3.5153649257123787,
      "tokens_seen": 1503461376
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002750050150451354,
      "loss": 2.6645,
      "theoretical_loss": 3.515351421401207,
      "tokens_seen": 1503526912
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027499498495486455,
      "loss": 2.7736,
      "theoretical_loss": 3.515337917843459,
      "tokens_seen": 1503592448
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002749849548645938,
      "loss": 2.7777,
      "theoretical_loss": 3.5153244150390597,
      "tokens_seen": 1503657984
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027497492477432297,
      "loss": 2.9496,
      "theoretical_loss": 3.515310912987933,
      "tokens_seen": 1503723520
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027496489468405215,
      "loss": 2.6052,
      "theoretical_loss": 3.5152974116900055,
      "tokens_seen": 1503789056
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027495486459378133,
      "loss": 2.6648,
      "theoretical_loss": 3.5152839111452017,
      "tokens_seen": 1503854592
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027494483450351057,
      "loss": 3.0176,
      "theoretical_loss": 3.5152704113534465,
      "tokens_seen": 1503920128
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002749348044132397,
      "loss": 2.6487,
      "theoretical_loss": 3.5152569123146664,
      "tokens_seen": 1503985664
    },
    {
      "epoch": 0.46,
      "objective/train/advantage_avg": 0.4926580786705017,
      "objective/train/docs_used": 853123,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.281559944152832,
      "objective/train/original_loss": 2.281559705734253,
      "objective/train/theoretical_loss": 3.515243414028785,
      "objective/train/tokens_used": 1524511200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24279901385307312,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505000352859497,
      "objective/train/weighted_lm_loss": 2.3969197273254395,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0253151655197144,
      "theoretical_loss": 3.515243414028785,
      "tokens_seen": 1504051200
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027492477432296893,
      "loss": 2.7122,
      "theoretical_loss": 3.515243414028785,
      "tokens_seen": 1504051200
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027491474423269806,
      "loss": 2.8287,
      "theoretical_loss": 3.5152299164957284,
      "tokens_seen": 1504116736
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002749047141424273,
      "loss": 2.7346,
      "theoretical_loss": 3.515216419715421,
      "tokens_seen": 1504182272
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027489468405215647,
      "loss": 2.893,
      "theoretical_loss": 3.5152029236877897,
      "tokens_seen": 1504247808
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027488465396188565,
      "loss": 2.6734,
      "theoretical_loss": 3.515189428412758,
      "tokens_seen": 1504313344
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027487462387161483,
      "loss": 2.7883,
      "theoretical_loss": 3.5151759338902515,
      "tokens_seen": 1504378880
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.000274864593781344,
      "loss": 2.9232,
      "theoretical_loss": 3.5151624401201964,
      "tokens_seen": 1504444416
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002748545636910732,
      "loss": 2.7499,
      "theoretical_loss": 3.5151489471025172,
      "tokens_seen": 1504509952
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027484453360080243,
      "loss": 2.5643,
      "theoretical_loss": 3.5151354548371394,
      "tokens_seen": 1504575488
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027483450351053156,
      "loss": 2.7326,
      "theoretical_loss": 3.5151219633239874,
      "tokens_seen": 1504641024
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002748244734202608,
      "loss": 2.8799,
      "theoretical_loss": 3.5151084725629884,
      "tokens_seen": 1504706560
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027481444332999,
      "loss": 2.6207,
      "theoretical_loss": 3.515094982554066,
      "tokens_seen": 1504772096
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027480441323971916,
      "loss": 2.5431,
      "theoretical_loss": 3.515081493297146,
      "tokens_seen": 1504837632
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027479438314944834,
      "loss": 2.7523,
      "theoretical_loss": 3.5150680047921545,
      "tokens_seen": 1504903168
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002747843530591775,
      "loss": 2.7626,
      "theoretical_loss": 3.5150545170390153,
      "tokens_seen": 1504968704
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027477432296890675,
      "loss": 2.6875,
      "theoretical_loss": 3.515041030037655,
      "tokens_seen": 1505034240
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027476429287863594,
      "loss": 3.0736,
      "theoretical_loss": 3.5150275437879994,
      "tokens_seen": 1505099776
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002747542627883651,
      "loss": 2.8098,
      "theoretical_loss": 3.5150140582899723,
      "tokens_seen": 1505165312
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002747442326980943,
      "loss": 2.5728,
      "theoretical_loss": 3.5150005735435004,
      "tokens_seen": 1505230848
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002747342026078235,
      "loss": 2.7425,
      "theoretical_loss": 3.514987089548508,
      "tokens_seen": 1505296384
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027472417251755266,
      "loss": 2.7892,
      "theoretical_loss": 3.5149736063049213,
      "tokens_seen": 1505361920
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002747141424272819,
      "loss": 2.9847,
      "theoretical_loss": 3.5149601238126653,
      "tokens_seen": 1505427456
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.000274704112337011,
      "loss": 3.0031,
      "theoretical_loss": 3.514946642071666,
      "tokens_seen": 1505492992
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027469408224674026,
      "loss": 2.7516,
      "theoretical_loss": 3.5149331610818484,
      "tokens_seen": 1505558528
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002746840521564694,
      "loss": 2.7777,
      "theoretical_loss": 3.514919680843138,
      "tokens_seen": 1505624064
    },
    {
      "epoch": 0.46,
      "objective/train/advantage_avg": 0.4934435784816742,
      "objective/train/docs_used": 854391,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.7000315189361572,
      "objective/train/original_loss": 2.7000315189361572,
      "objective/train/theoretical_loss": 3.51490620135546,
      "objective/train/tokens_used": 1526149600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2435896098613739,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505825281143188,
      "objective/train/weighted_lm_loss": 2.836474895477295,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0373471975326538,
      "theoretical_loss": 3.51490620135546,
      "tokens_seen": 1505689600
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002746740220661986,
      "loss": 2.9234,
      "theoretical_loss": 3.51490620135546,
      "tokens_seen": 1505689600
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002746639919759278,
      "loss": 3.0927,
      "theoretical_loss": 3.5148927226187405,
      "tokens_seen": 1505755136
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.000274653961885657,
      "loss": 2.7449,
      "theoretical_loss": 3.5148792446329047,
      "tokens_seen": 1505820672
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027464393179538616,
      "loss": 2.7966,
      "theoretical_loss": 3.514865767397878,
      "tokens_seen": 1505886208
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002746339017051154,
      "loss": 2.8557,
      "theoretical_loss": 3.514852290913586,
      "tokens_seen": 1505951744
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002746238716148445,
      "loss": 2.8313,
      "theoretical_loss": 3.514838815179954,
      "tokens_seen": 1506017280
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027461384152457376,
      "loss": 2.8973,
      "theoretical_loss": 3.5148253401969076,
      "tokens_seen": 1506082816
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002746038114343029,
      "loss": 2.6715,
      "theoretical_loss": 3.514811865964372,
      "tokens_seen": 1506148352
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002745937813440321,
      "loss": 2.9212,
      "theoretical_loss": 3.514798392482274,
      "tokens_seen": 1506213888
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002745837512537613,
      "loss": 2.6802,
      "theoretical_loss": 3.514784919750538,
      "tokens_seen": 1506279424
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002745737211634905,
      "loss": 2.6037,
      "theoretical_loss": 3.51477144776909,
      "tokens_seen": 1506344960
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027456369107321967,
      "loss": 2.7125,
      "theoretical_loss": 3.514757976537855,
      "tokens_seen": 1506410496
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027455366098294885,
      "loss": 2.532,
      "theoretical_loss": 3.5147445060567595,
      "tokens_seen": 1506476032
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027454363089267803,
      "loss": 2.9146,
      "theoretical_loss": 3.514731036325728,
      "tokens_seen": 1506541568
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027453360080240726,
      "loss": 2.8342,
      "theoretical_loss": 3.5147175673446873,
      "tokens_seen": 1506607104
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002745235707121364,
      "loss": 2.7869,
      "theoretical_loss": 3.5147040991135623,
      "tokens_seen": 1506672640
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002745135406218656,
      "loss": 2.6817,
      "theoretical_loss": 3.514690631632279,
      "tokens_seen": 1506738176
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027450351053159475,
      "loss": 3.0764,
      "theoretical_loss": 3.514677164900762,
      "tokens_seen": 1506803712
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.000274493480441324,
      "loss": 2.7269,
      "theoretical_loss": 3.514663698918939,
      "tokens_seen": 1506869248
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027448345035105317,
      "loss": 2.8041,
      "theoretical_loss": 3.5146502336867336,
      "tokens_seen": 1506934784
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027447342026078235,
      "loss": 2.6115,
      "theoretical_loss": 3.5146367692040728,
      "tokens_seen": 1507000320
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027446339017051153,
      "loss": 2.8647,
      "theoretical_loss": 3.5146233054708818,
      "tokens_seen": 1507065856
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027445336008024077,
      "loss": 2.8068,
      "theoretical_loss": 3.5146098424870855,
      "tokens_seen": 1507131392
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002744433299899699,
      "loss": 2.9361,
      "theoretical_loss": 3.5145963802526112,
      "tokens_seen": 1507196928
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027443329989969913,
      "loss": 2.8192,
      "theoretical_loss": 3.514582918767384,
      "tokens_seen": 1507262464
    },
    {
      "epoch": 0.46,
      "objective/train/advantage_avg": 0.49413156509399414,
      "objective/train/docs_used": 855086,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.3066415786743164,
      "objective/train/original_loss": 3.3066418170928955,
      "objective/train/theoretical_loss": 3.5145694580313287,
      "objective/train/tokens_used": 1527788000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2442956268787384,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506550073623657,
      "objective/train/weighted_lm_loss": 3.4742584228515625,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0253151655197144,
      "theoretical_loss": 3.5145694580313287,
      "tokens_seen": 1507328000
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027442326980942826,
      "loss": 2.7925,
      "theoretical_loss": 3.5145694580313287,
      "tokens_seen": 1507328000
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002744132397191575,
      "loss": 2.7561,
      "theoretical_loss": 3.514555998044372,
      "tokens_seen": 1507393536
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027440320962888667,
      "loss": 2.9925,
      "theoretical_loss": 3.51454253880644,
      "tokens_seen": 1507459072
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027439317953861585,
      "loss": 3.1129,
      "theoretical_loss": 3.514529080317457,
      "tokens_seen": 1507524608
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027438314944834503,
      "loss": 2.5691,
      "theoretical_loss": 3.5145156225773504,
      "tokens_seen": 1507590144
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002743731193580742,
      "loss": 2.6848,
      "theoretical_loss": 3.5145021655860447,
      "tokens_seen": 1507655680
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002743630892678034,
      "loss": 2.8429,
      "theoretical_loss": 3.5144887093434667,
      "tokens_seen": 1507721216
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027435305917753263,
      "loss": 2.7683,
      "theoretical_loss": 3.5144752538495414,
      "tokens_seen": 1507786752
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027434302908726176,
      "loss": 3.0058,
      "theoretical_loss": 3.514461799104195,
      "tokens_seen": 1507852288
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.000274332998996991,
      "loss": 2.7876,
      "theoretical_loss": 3.514448345107353,
      "tokens_seen": 1507917824
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002743229689067202,
      "loss": 3.0298,
      "theoretical_loss": 3.514434891858942,
      "tokens_seen": 1507983360
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027431293881644936,
      "loss": 2.7948,
      "theoretical_loss": 3.514421439358887,
      "tokens_seen": 1508048896
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027430290872617854,
      "loss": 2.9055,
      "theoretical_loss": 3.514407987607114,
      "tokens_seen": 1508114432
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002742928786359077,
      "loss": 2.9515,
      "theoretical_loss": 3.5143945366035494,
      "tokens_seen": 1508179968
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002742828485456369,
      "loss": 2.7834,
      "theoretical_loss": 3.5143810863481186,
      "tokens_seen": 1508245504
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027427281845536614,
      "loss": 2.736,
      "theoretical_loss": 3.5143676368407477,
      "tokens_seen": 1508311040
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027426278836509526,
      "loss": 2.9774,
      "theoretical_loss": 3.5143541880813625,
      "tokens_seen": 1508376576
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002742527582748245,
      "loss": 2.8378,
      "theoretical_loss": 3.5143407400698887,
      "tokens_seen": 1508442112
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002742427281845536,
      "loss": 2.745,
      "theoretical_loss": 3.514327292806253,
      "tokens_seen": 1508507648
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027423269809428286,
      "loss": 2.7193,
      "theoretical_loss": 3.5143138462903796,
      "tokens_seen": 1508573184
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027422266800401204,
      "loss": 2.8149,
      "theoretical_loss": 3.5143004005221963,
      "tokens_seen": 1508638720
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002742126379137412,
      "loss": 3.0483,
      "theoretical_loss": 3.5142869555016283,
      "tokens_seen": 1508704256
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002742026078234704,
      "loss": 2.8644,
      "theoretical_loss": 3.5142735112286014,
      "tokens_seen": 1508769792
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002741925777331996,
      "loss": 2.9912,
      "theoretical_loss": 3.514260067703042,
      "tokens_seen": 1508835328
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027418254764292877,
      "loss": 2.6687,
      "theoretical_loss": 3.5142466249248754,
      "tokens_seen": 1508900864
    },
    {
      "epoch": 0.46,
      "objective/train/advantage_avg": 0.49428659677505493,
      "objective/train/docs_used": 856541,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.811544418334961,
      "objective/train/original_loss": 2.81154465675354,
      "objective/train/theoretical_loss": 3.5142331828940288,
      "objective/train/tokens_used": 1529426400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24439191818237305,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506709814071655,
      "objective/train/weighted_lm_loss": 2.9541738033294678,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0338950157165527,
      "theoretical_loss": 3.5142331828940288,
      "tokens_seen": 1508966400
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.000274172517552658,
      "loss": 2.9541,
      "theoretical_loss": 3.5142331828940288,
      "tokens_seen": 1508966400
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027416248746238713,
      "loss": 2.6079,
      "theoretical_loss": 3.514219741610427,
      "tokens_seen": 1509031936
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027415245737211636,
      "loss": 2.6199,
      "theoretical_loss": 3.514206301073996,
      "tokens_seen": 1509097472
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027414242728184554,
      "loss": 2.7672,
      "theoretical_loss": 3.5141928612846627,
      "tokens_seen": 1509163008
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002741323971915747,
      "loss": 2.6749,
      "theoretical_loss": 3.5141794222423526,
      "tokens_seen": 1509228544
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002741223671013039,
      "loss": 2.7496,
      "theoretical_loss": 3.514165983946992,
      "tokens_seen": 1509294080
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002741123370110331,
      "loss": 2.9546,
      "theoretical_loss": 3.5141525463985066,
      "tokens_seen": 1509359616
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027410230692076227,
      "loss": 2.8619,
      "theoretical_loss": 3.5141391095968224,
      "tokens_seen": 1509425152
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002740922768304915,
      "loss": 2.9852,
      "theoretical_loss": 3.5141256735418662,
      "tokens_seen": 1509490688
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027408224674022063,
      "loss": 2.933,
      "theoretical_loss": 3.5141122382335634,
      "tokens_seen": 1509556224
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027407221664994987,
      "loss": 2.8441,
      "theoretical_loss": 3.514098803671841,
      "tokens_seen": 1509621760
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.000274062186559679,
      "loss": 2.9633,
      "theoretical_loss": 3.5140853698566232,
      "tokens_seen": 1509687296
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027405215646940823,
      "loss": 2.8424,
      "theoretical_loss": 3.514071936787838,
      "tokens_seen": 1509752832
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002740421263791374,
      "loss": 2.9423,
      "theoretical_loss": 3.5140585044654102,
      "tokens_seen": 1509818368
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002740320962888666,
      "loss": 2.7273,
      "theoretical_loss": 3.5140450728892674,
      "tokens_seen": 1509883904
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002740220661985958,
      "loss": 2.5395,
      "theoretical_loss": 3.5140316420593347,
      "tokens_seen": 1509949440
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027401203610832495,
      "loss": 2.9189,
      "theoretical_loss": 3.5140182119755385,
      "tokens_seen": 1510014976
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002740020060180542,
      "loss": 2.7717,
      "theoretical_loss": 3.5140047826378047,
      "tokens_seen": 1510080512
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027399197592778337,
      "loss": 2.7479,
      "theoretical_loss": 3.51399135404606,
      "tokens_seen": 1510146048
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027398194583751255,
      "loss": 2.8436,
      "theoretical_loss": 3.51397792620023,
      "tokens_seen": 1510211584
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027397191574724173,
      "loss": 2.7664,
      "theoretical_loss": 3.5139644991002417,
      "tokens_seen": 1510277120
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027396188565697097,
      "loss": 2.9322,
      "theoretical_loss": 3.5139510727460204,
      "tokens_seen": 1510342656
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002739518555667001,
      "loss": 2.9347,
      "theoretical_loss": 3.5139376471374932,
      "tokens_seen": 1510408192
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027394182547642933,
      "loss": 2.5785,
      "theoretical_loss": 3.5139242222745857,
      "tokens_seen": 1510473728
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027393179538615846,
      "loss": 2.7392,
      "theoretical_loss": 3.513910798157224,
      "tokens_seen": 1510539264
    },
    {
      "epoch": 0.46,
      "objective/train/advantage_avg": 0.495487779378891,
      "objective/train/docs_used": 857299,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.5525007247924805,
      "objective/train/original_loss": 2.5525007247924805,
      "objective/train/theoretical_loss": 3.5138973747853353,
      "objective/train/tokens_used": 1531064800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24555324018001556,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507969856262207,
      "objective/train/weighted_lm_loss": 2.682046413421631,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0338950157165527,
      "theoretical_loss": 3.5138973747853353,
      "tokens_seen": 1510604800
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002739217652958877,
      "loss": 2.8765,
      "theoretical_loss": 3.5138973747853353,
      "tokens_seen": 1510604800
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027391173520561687,
      "loss": 2.8312,
      "theoretical_loss": 3.513883952158845,
      "tokens_seen": 1510670336
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027390170511534605,
      "loss": 2.6429,
      "theoretical_loss": 3.5138705302776794,
      "tokens_seen": 1510735872
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027389167502507524,
      "loss": 2.7143,
      "theoretical_loss": 3.513857109141765,
      "tokens_seen": 1510801408
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002738816449348044,
      "loss": 2.7853,
      "theoretical_loss": 3.5138436887510283,
      "tokens_seen": 1510866944
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002738716148445336,
      "loss": 2.6132,
      "theoretical_loss": 3.5138302691053953,
      "tokens_seen": 1510932480
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027386158475426283,
      "loss": 2.8966,
      "theoretical_loss": 3.513816850204793,
      "tokens_seen": 1510998016
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027385155466399196,
      "loss": 2.7664,
      "theoretical_loss": 3.5138034320491465,
      "tokens_seen": 1511063552
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002738415245737212,
      "loss": 2.6507,
      "theoretical_loss": 3.513790014638383,
      "tokens_seen": 1511129088
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002738314944834504,
      "loss": 2.6375,
      "theoretical_loss": 3.513776597972429,
      "tokens_seen": 1511194624
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027382146439317956,
      "loss": 2.712,
      "theoretical_loss": 3.5137631820512096,
      "tokens_seen": 1511260160
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027381143430290874,
      "loss": 2.9332,
      "theoretical_loss": 3.5137497668746525,
      "tokens_seen": 1511325696
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002738014042126379,
      "loss": 2.9447,
      "theoretical_loss": 3.5137363524426837,
      "tokens_seen": 1511391232
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002737913741223671,
      "loss": 2.6627,
      "theoretical_loss": 3.5137229387552296,
      "tokens_seen": 1511456768
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027378134403209634,
      "loss": 2.9021,
      "theoretical_loss": 3.5137095258122164,
      "tokens_seen": 1511522304
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027377131394182546,
      "loss": 2.7135,
      "theoretical_loss": 3.513696113613571,
      "tokens_seen": 1511587840
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002737612838515547,
      "loss": 2.6081,
      "theoretical_loss": 3.513682702159219,
      "tokens_seen": 1511653376
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002737512537612838,
      "loss": 3.0311,
      "theoretical_loss": 3.5136692914490877,
      "tokens_seen": 1511718912
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027374122367101306,
      "loss": 2.619,
      "theoretical_loss": 3.513655881483103,
      "tokens_seen": 1511784448
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027373119358074224,
      "loss": 2.8704,
      "theoretical_loss": 3.513642472261191,
      "tokens_seen": 1511849984
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002737211634904714,
      "loss": 2.6394,
      "theoretical_loss": 3.513629063783279,
      "tokens_seen": 1511915520
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002737111334002006,
      "loss": 2.6964,
      "theoretical_loss": 3.5136156560492937,
      "tokens_seen": 1511981056
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002737011033099298,
      "loss": 2.9624,
      "theoretical_loss": 3.5136022490591605,
      "tokens_seen": 1512046592
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027369107321965897,
      "loss": 2.8374,
      "theoretical_loss": 3.513588842812806,
      "tokens_seen": 1512112128
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002736810431293882,
      "loss": 2.8421,
      "theoretical_loss": 3.5135754373101578,
      "tokens_seen": 1512177664
    },
    {
      "epoch": 0.46,
      "objective/train/advantage_avg": 0.49355363845825195,
      "objective/train/docs_used": 858636,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.6845436096191406,
      "objective/train/original_loss": 2.6845431327819824,
      "objective/train/theoretical_loss": 3.513562032551141,
      "objective/train/tokens_used": 1532703200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24368596076965332,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505940914154053,
      "objective/train/weighted_lm_loss": 2.8203988075256348,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0304546356201172,
      "theoretical_loss": 3.513562032551141,
      "tokens_seen": 1512243200
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027367101303911733,
      "loss": 2.7462,
      "theoretical_loss": 3.513562032551141,
      "tokens_seen": 1512243200
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027366098294884656,
      "loss": 2.6987,
      "theoretical_loss": 3.513548628535683,
      "tokens_seen": 1512308736
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027365095285857574,
      "loss": 2.6738,
      "theoretical_loss": 3.513535225263711,
      "tokens_seen": 1512374272
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002736409227683049,
      "loss": 2.9885,
      "theoretical_loss": 3.51352182273515,
      "tokens_seen": 1512439808
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002736308926780341,
      "loss": 2.8323,
      "theoretical_loss": 3.513508420949927,
      "tokens_seen": 1512505344
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002736208625877633,
      "loss": 2.95,
      "theoretical_loss": 3.5134950199079698,
      "tokens_seen": 1512570880
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027361083249749247,
      "loss": 2.7979,
      "theoretical_loss": 3.5134816196092036,
      "tokens_seen": 1512636416
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002736008024072217,
      "loss": 2.94,
      "theoretical_loss": 3.513468220053555,
      "tokens_seen": 1512701952
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027359077231695083,
      "loss": 2.8502,
      "theoretical_loss": 3.5134548212409515,
      "tokens_seen": 1512767488
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027358074222668007,
      "loss": 2.7775,
      "theoretical_loss": 3.5134414231713187,
      "tokens_seen": 1512833024
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002735707121364092,
      "loss": 2.9046,
      "theoretical_loss": 3.5134280258445845,
      "tokens_seen": 1512898560
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027356068204613843,
      "loss": 2.7247,
      "theoretical_loss": 3.5134146292606743,
      "tokens_seen": 1512964096
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002735506519558676,
      "loss": 2.9088,
      "theoretical_loss": 3.513401233419515,
      "tokens_seen": 1513029632
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002735406218655968,
      "loss": 2.7273,
      "theoretical_loss": 3.5133878383210337,
      "tokens_seen": 1513095168
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027353059177532597,
      "loss": 2.7074,
      "theoretical_loss": 3.513374443965157,
      "tokens_seen": 1513160704
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027352056168505515,
      "loss": 2.6688,
      "theoretical_loss": 3.5133610503518113,
      "tokens_seen": 1513226240
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027351053159478433,
      "loss": 2.8196,
      "theoretical_loss": 3.513347657480923,
      "tokens_seen": 1513291776
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027350050150451357,
      "loss": 2.7726,
      "theoretical_loss": 3.5133342653524195,
      "tokens_seen": 1513357312
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002734904714142427,
      "loss": 2.5718,
      "theoretical_loss": 3.5133208739662267,
      "tokens_seen": 1513422848
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027348044132397193,
      "loss": 2.9327,
      "theoretical_loss": 3.5133074833222726,
      "tokens_seen": 1513488384
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002734704112337011,
      "loss": 2.5183,
      "theoretical_loss": 3.5132940934204826,
      "tokens_seen": 1513553920
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002734603811434303,
      "loss": 2.8464,
      "theoretical_loss": 3.513280704260784,
      "tokens_seen": 1513619456
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002734503510531595,
      "loss": 2.7211,
      "theoretical_loss": 3.5132673158431036,
      "tokens_seen": 1513684992
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027344032096288866,
      "loss": 2.8138,
      "theoretical_loss": 3.5132539281673676,
      "tokens_seen": 1513750528
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027343029087261784,
      "loss": 2.6792,
      "theoretical_loss": 3.513240541233504,
      "tokens_seen": 1513816064
    },
    {
      "epoch": 0.46,
      "objective/train/advantage_avg": 0.4947545826435089,
      "objective/train/docs_used": 859135,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.2203543186187744,
      "objective/train/original_loss": 2.2203540802001953,
      "objective/train/theoretical_loss": 3.513227155041438,
      "objective/train/tokens_used": 1534341600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24484670162200928,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507200956344604,
      "objective/train/weighted_lm_loss": 2.3328187465667725,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0338950157165527,
      "theoretical_loss": 3.513227155041438,
      "tokens_seen": 1513881600
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002734202607823471,
      "loss": 2.707,
      "theoretical_loss": 3.513227155041438,
      "tokens_seen": 1513881600
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002734102306920762,
      "loss": 2.8319,
      "theoretical_loss": 3.5132137695910974,
      "tokens_seen": 1513947136
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027340020060180544,
      "loss": 2.6916,
      "theoretical_loss": 3.513200384882409,
      "tokens_seen": 1514012672
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027339017051153456,
      "loss": 3.0297,
      "theoretical_loss": 3.5131870009152992,
      "tokens_seen": 1514078208
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002733801404212638,
      "loss": 2.614,
      "theoretical_loss": 3.513173617689695,
      "tokens_seen": 1514143744
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.000273370110330993,
      "loss": 2.7377,
      "theoretical_loss": 3.513160235205523,
      "tokens_seen": 1514209280
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027336008024072216,
      "loss": 2.7715,
      "theoretical_loss": 3.51314685346271,
      "tokens_seen": 1514274816
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027335005015045134,
      "loss": 2.728,
      "theoretical_loss": 3.513133472461184,
      "tokens_seen": 1514340352
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002733400200601806,
      "loss": 2.8193,
      "theoretical_loss": 3.5131200922008703,
      "tokens_seen": 1514405888
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002733299899699097,
      "loss": 2.9303,
      "theoretical_loss": 3.5131067126816964,
      "tokens_seen": 1514471424
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027331995987963894,
      "loss": 2.8559,
      "theoretical_loss": 3.5130933339035897,
      "tokens_seen": 1514536960
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027330992978936807,
      "loss": 2.7207,
      "theoretical_loss": 3.513079955866476,
      "tokens_seen": 1514602496
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002732998996990973,
      "loss": 2.8537,
      "theoretical_loss": 3.513066578570283,
      "tokens_seen": 1514668032
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002732898696088265,
      "loss": 2.6904,
      "theoretical_loss": 3.5130532020149374,
      "tokens_seen": 1514733568
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027327983951855566,
      "loss": 2.7809,
      "theoretical_loss": 3.513039826200366,
      "tokens_seen": 1514799104
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002732698094282849,
      "loss": 2.5751,
      "theoretical_loss": 3.513026451126496,
      "tokens_seen": 1514864640
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.000273259779338014,
      "loss": 2.792,
      "theoretical_loss": 3.5130130767932544,
      "tokens_seen": 1514930176
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027324974924774326,
      "loss": 2.8652,
      "theoretical_loss": 3.5129997032005673,
      "tokens_seen": 1514995712
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027323971915747244,
      "loss": 3.0983,
      "theoretical_loss": 3.5129863303483626,
      "tokens_seen": 1515061248
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002732296890672016,
      "loss": 2.9547,
      "theoretical_loss": 3.512972958236567,
      "tokens_seen": 1515126784
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002732196589769308,
      "loss": 2.7708,
      "theoretical_loss": 3.512959586865108,
      "tokens_seen": 1515192320
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027320962888666,
      "loss": 2.9035,
      "theoretical_loss": 3.512946216233911,
      "tokens_seen": 1515257856
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027319959879638917,
      "loss": 2.7684,
      "theoretical_loss": 3.5129328463429053,
      "tokens_seen": 1515323392
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002731895687061184,
      "loss": 2.7971,
      "theoretical_loss": 3.512919477192016,
      "tokens_seen": 1515388928
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027317953861584753,
      "loss": 2.923,
      "theoretical_loss": 3.5129061087811704,
      "tokens_seen": 1515454464
    },
    {
      "epoch": 0.46,
      "objective/train/advantage_avg": 0.49216827750205994,
      "objective/train/docs_used": 860397,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.5662729740142822,
      "objective/train/original_loss": 2.5662729740142822,
      "objective/train/theoretical_loss": 3.5128927411102966,
      "objective/train/tokens_used": 1535980000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24233262240886688,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0504485368728638,
      "objective/train/weighted_lm_loss": 2.6960980892181396,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0437887907028198,
      "theoretical_loss": 3.5128927411102966,
      "tokens_seen": 1515520000
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027316950852557676,
      "loss": 2.5307,
      "theoretical_loss": 3.5128927411102966,
      "tokens_seen": 1515520000
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027315947843530594,
      "loss": 2.7153,
      "theoretical_loss": 3.5128793741793203,
      "tokens_seen": 1515585536
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002731494483450351,
      "loss": 2.7349,
      "theoretical_loss": 3.51286600798817,
      "tokens_seen": 1515651072
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002731394182547643,
      "loss": 2.9038,
      "theoretical_loss": 3.5128526425367714,
      "tokens_seen": 1515716608
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002731293881644935,
      "loss": 2.8242,
      "theoretical_loss": 3.5128392778250523,
      "tokens_seen": 1515782144
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027311935807422267,
      "loss": 2.6946,
      "theoretical_loss": 3.5128259138529394,
      "tokens_seen": 1515847680
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002731093279839519,
      "loss": 2.9225,
      "theoretical_loss": 3.5128125506203607,
      "tokens_seen": 1515913216
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027309929789368103,
      "loss": 2.7608,
      "theoretical_loss": 3.512799188127242,
      "tokens_seen": 1515978752
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027308926780341027,
      "loss": 2.7672,
      "theoretical_loss": 3.512785826373512,
      "tokens_seen": 1516044288
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002730792377131394,
      "loss": 2.7827,
      "theoretical_loss": 3.512772465359096,
      "tokens_seen": 1516109824
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027306920762286863,
      "loss": 2.7316,
      "theoretical_loss": 3.512759105083922,
      "tokens_seen": 1516175360
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002730591775325978,
      "loss": 2.5118,
      "theoretical_loss": 3.512745745547918,
      "tokens_seen": 1516240896
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.000273049147442327,
      "loss": 2.9504,
      "theoretical_loss": 3.5127323867510096,
      "tokens_seen": 1516306432
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027303911735205617,
      "loss": 2.6835,
      "theoretical_loss": 3.512719028693125,
      "tokens_seen": 1516371968
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027302908726178535,
      "loss": 2.7875,
      "theoretical_loss": 3.5127056713741913,
      "tokens_seen": 1516437504
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027301905717151453,
      "loss": 2.9412,
      "theoretical_loss": 3.512692314794135,
      "tokens_seen": 1516503040
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027300902708124377,
      "loss": 2.8869,
      "theoretical_loss": 3.512678958952884,
      "tokens_seen": 1516568576
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002729989969909729,
      "loss": 2.6672,
      "theoretical_loss": 3.5126656038503654,
      "tokens_seen": 1516634112
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027298896690070213,
      "loss": 2.7416,
      "theoretical_loss": 3.5126522494865062,
      "tokens_seen": 1516699648
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002729789368104313,
      "loss": 2.9382,
      "theoretical_loss": 3.512638895861234,
      "tokens_seen": 1516765184
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002729689067201605,
      "loss": 2.7688,
      "theoretical_loss": 3.5126255429744755,
      "tokens_seen": 1516830720
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002729588766298897,
      "loss": 2.6804,
      "theoretical_loss": 3.5126121908261583,
      "tokens_seen": 1516896256
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027294884653961886,
      "loss": 2.7695,
      "theoretical_loss": 3.5125988394162095,
      "tokens_seen": 1516961792
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027293881644934804,
      "loss": 2.7049,
      "theoretical_loss": 3.5125854887445565,
      "tokens_seen": 1517027328
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002729287863590773,
      "loss": 3.0523,
      "theoretical_loss": 3.5125721388111266,
      "tokens_seen": 1517092864
    },
    {
      "epoch": 0.46,
      "objective/train/advantage_avg": 0.4953611493110657,
      "objective/train/docs_used": 860980,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.357553482055664,
      "objective/train/original_loss": 2.357553005218506,
      "objective/train/theoretical_loss": 3.5125587896158477,
      "objective/train/tokens_used": 1537618400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24541768431663513,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050783634185791,
      "objective/train/weighted_lm_loss": 2.477138042449951,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0451053380966187,
      "theoretical_loss": 3.5125587896158477,
      "tokens_seen": 1517158400
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002729187562688064,
      "loss": 2.8396,
      "theoretical_loss": 3.5125587896158477,
      "tokens_seen": 1517158400
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027290872617853564,
      "loss": 2.7636,
      "theoretical_loss": 3.5125454411586454,
      "tokens_seen": 1517223936
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027289869608826476,
      "loss": 3.0836,
      "theoretical_loss": 3.5125320934394484,
      "tokens_seen": 1517289472
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.000272888665997994,
      "loss": 2.9894,
      "theoretical_loss": 3.512518746458184,
      "tokens_seen": 1517355008
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002728786359077232,
      "loss": 2.5982,
      "theoretical_loss": 3.512505400214779,
      "tokens_seen": 1517420544
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027286860581745236,
      "loss": 2.8856,
      "theoretical_loss": 3.5124920547091607,
      "tokens_seen": 1517486080
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027285857572718154,
      "loss": 2.7399,
      "theoretical_loss": 3.5124787099412575,
      "tokens_seen": 1517551616
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002728485456369108,
      "loss": 2.5426,
      "theoretical_loss": 3.5124653659109955,
      "tokens_seen": 1517617152
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002728385155466399,
      "loss": 2.6815,
      "theoretical_loss": 3.5124520226183025,
      "tokens_seen": 1517682688
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027282848545636914,
      "loss": 2.5954,
      "theoretical_loss": 3.512438680063106,
      "tokens_seen": 1517748224
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027281845536609827,
      "loss": 2.8288,
      "theoretical_loss": 3.512425338245333,
      "tokens_seen": 1517813760
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002728084252758275,
      "loss": 2.622,
      "theoretical_loss": 3.5124119971649117,
      "tokens_seen": 1517879296
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002727983951855567,
      "loss": 2.733,
      "theoretical_loss": 3.5123986568217687,
      "tokens_seen": 1517944832
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027278836509528586,
      "loss": 2.8489,
      "theoretical_loss": 3.5123853172158324,
      "tokens_seen": 1518010368
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027277833500501504,
      "loss": 2.6917,
      "theoretical_loss": 3.512371978347029,
      "tokens_seen": 1518075904
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002727683049147442,
      "loss": 2.7082,
      "theoretical_loss": 3.512358640215287,
      "tokens_seen": 1518141440
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002727582748244734,
      "loss": 2.6602,
      "theoretical_loss": 3.512345302820533,
      "tokens_seen": 1518206976
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027274824473420264,
      "loss": 2.8922,
      "theoretical_loss": 3.5123319661626953,
      "tokens_seen": 1518272512
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027273821464393177,
      "loss": 2.6864,
      "theoretical_loss": 3.5123186302417007,
      "tokens_seen": 1518338048
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.000272728184553661,
      "loss": 2.9057,
      "theoretical_loss": 3.512305295057477,
      "tokens_seen": 1518403584
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027271815446339013,
      "loss": 2.7618,
      "theoretical_loss": 3.512291960609952,
      "tokens_seen": 1518469120
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027270812437311937,
      "loss": 2.8232,
      "theoretical_loss": 3.5122786268990525,
      "tokens_seen": 1518534656
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027269809428284855,
      "loss": 2.917,
      "theoretical_loss": 3.5122652939247065,
      "tokens_seen": 1518600192
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027268806419257773,
      "loss": 2.8844,
      "theoretical_loss": 3.5122519616868413,
      "tokens_seen": 1518665728
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002726780341023069,
      "loss": 2.9713,
      "theoretical_loss": 3.5122386301853847,
      "tokens_seen": 1518731264
    },
    {
      "epoch": 0.46,
      "objective/train/advantage_avg": 0.4938102066516876,
      "objective/train/docs_used": 862373,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.5721991062164307,
      "objective/train/original_loss": 2.572199821472168,
      "objective/train/theoretical_loss": 3.512225299420264,
      "objective/train/tokens_used": 1539256800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24390998482704163,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506209135055542,
      "objective/train/weighted_lm_loss": 2.70257568359375,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0432155132293701,
      "theoretical_loss": 3.512225299420264,
      "tokens_seen": 1518796800
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027266800401203614,
      "loss": 2.7863,
      "theoretical_loss": 3.512225299420264,
      "tokens_seen": 1518796800
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027265797392176527,
      "loss": 2.7729,
      "theoretical_loss": 3.512211969391407,
      "tokens_seen": 1518862336
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002726479438314945,
      "loss": 2.7081,
      "theoretical_loss": 3.5121986400987413,
      "tokens_seen": 1518927872
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027263791374122363,
      "loss": 2.8024,
      "theoretical_loss": 3.512185311542194,
      "tokens_seen": 1518993408
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027262788365095287,
      "loss": 2.7855,
      "theoretical_loss": 3.512171983721693,
      "tokens_seen": 1519058944
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027261785356068205,
      "loss": 2.8178,
      "theoretical_loss": 3.5121586566371663,
      "tokens_seen": 1519124480
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027260782347041123,
      "loss": 2.9066,
      "theoretical_loss": 3.512145330288541,
      "tokens_seen": 1519190016
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002725977933801404,
      "loss": 2.7072,
      "theoretical_loss": 3.5121320046757445,
      "tokens_seen": 1519255552
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002725877632898696,
      "loss": 2.7708,
      "theoretical_loss": 3.512118679798705,
      "tokens_seen": 1519321088
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002725777331995988,
      "loss": 2.8928,
      "theoretical_loss": 3.51210535565735,
      "tokens_seen": 1519386624
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.000272567703109328,
      "loss": 2.7945,
      "theoretical_loss": 3.5120920322516067,
      "tokens_seen": 1519452160
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027255767301905714,
      "loss": 2.7778,
      "theoretical_loss": 3.5120787095814032,
      "tokens_seen": 1519517696
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027254764292878637,
      "loss": 2.7167,
      "theoretical_loss": 3.5120653876466674,
      "tokens_seen": 1519583232
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002725376128385155,
      "loss": 2.9522,
      "theoretical_loss": 3.5120520664473265,
      "tokens_seen": 1519648768
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027252758274824473,
      "loss": 2.9348,
      "theoretical_loss": 3.5120387459833085,
      "tokens_seen": 1519714304
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027251755265797397,
      "loss": 2.8278,
      "theoretical_loss": 3.5120254262545405,
      "tokens_seen": 1519779840
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002725075225677031,
      "loss": 2.5705,
      "theoretical_loss": 3.5120121072609516,
      "tokens_seen": 1519845376
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027249749247743233,
      "loss": 2.8715,
      "theoretical_loss": 3.511998789002468,
      "tokens_seen": 1519910912
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002724874623871615,
      "loss": 2.9615,
      "theoretical_loss": 3.5119854714790177,
      "tokens_seen": 1519976448
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002724774322968907,
      "loss": 2.8493,
      "theoretical_loss": 3.51197215469053,
      "tokens_seen": 1520041984
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002724674022066199,
      "loss": 2.8124,
      "theoretical_loss": 3.5119588386369305,
      "tokens_seen": 1520107520
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027245737211634906,
      "loss": 2.9058,
      "theoretical_loss": 3.511945523318148,
      "tokens_seen": 1520173056
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027244734202607824,
      "loss": 2.8042,
      "theoretical_loss": 3.51193220873411,
      "tokens_seen": 1520238592
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002724373119358075,
      "loss": 2.8423,
      "theoretical_loss": 3.511918894884745,
      "tokens_seen": 1520304128
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002724272818455366,
      "loss": 2.5642,
      "theoretical_loss": 3.5119055817699802,
      "tokens_seen": 1520369664
    },
    {
      "epoch": 0.46,
      "objective/train/advantage_avg": 0.49497976899147034,
      "objective/train/docs_used": 862945,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.781200885772705,
      "objective/train/original_loss": 2.781200647354126,
      "objective/train/theoretical_loss": 3.511892269389743,
      "objective/train/tokens_used": 1540895200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24505653977394104,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507436990737915,
      "objective/train/weighted_lm_loss": 2.9222989082336426,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 3.511892269389743,
      "tokens_seen": 1520435200
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027241725175526584,
      "loss": 2.7408,
      "theoretical_loss": 3.511892269389743,
      "tokens_seen": 1520435200
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027240722166499496,
      "loss": 2.7049,
      "theoretical_loss": 3.511878957743962,
      "tokens_seen": 1520500736
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002723971915747242,
      "loss": 2.8617,
      "theoretical_loss": 3.511865646832565,
      "tokens_seen": 1520566272
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002723871614844534,
      "loss": 2.7578,
      "theoretical_loss": 3.5118523366554792,
      "tokens_seen": 1520631808
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027237713139418256,
      "loss": 2.7623,
      "theoretical_loss": 3.511839027212633,
      "tokens_seen": 1520697344
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027236710130391174,
      "loss": 2.6335,
      "theoretical_loss": 3.511825718503954,
      "tokens_seen": 1520762880
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.000272357071213641,
      "loss": 2.8656,
      "theoretical_loss": 3.51181241052937,
      "tokens_seen": 1520828416
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002723470411233701,
      "loss": 2.9976,
      "theoretical_loss": 3.511799103288809,
      "tokens_seen": 1520893952
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027233701103309934,
      "loss": 2.6909,
      "theoretical_loss": 3.511785796782199,
      "tokens_seen": 1520959488
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027232698094282847,
      "loss": 2.7315,
      "theoretical_loss": 3.5117724910094674,
      "tokens_seen": 1521025024
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002723169508525577,
      "loss": 2.6346,
      "theoretical_loss": 3.511759185970543,
      "tokens_seen": 1521090560
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002723069207622869,
      "loss": 2.8585,
      "theoretical_loss": 3.511745881665353,
      "tokens_seen": 1521156096
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027229689067201606,
      "loss": 2.739,
      "theoretical_loss": 3.5117325780938256,
      "tokens_seen": 1521221632
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027228686058174524,
      "loss": 2.7882,
      "theoretical_loss": 3.5117192752558886,
      "tokens_seen": 1521287168
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002722768304914744,
      "loss": 2.7699,
      "theoretical_loss": 3.51170597315147,
      "tokens_seen": 1521352704
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002722668004012036,
      "loss": 2.8334,
      "theoretical_loss": 3.511692671780498,
      "tokens_seen": 1521418240
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027225677031093284,
      "loss": 2.8707,
      "theoretical_loss": 3.5116793711429004,
      "tokens_seen": 1521483776
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027224674022066197,
      "loss": 2.6679,
      "theoretical_loss": 3.5116660712386047,
      "tokens_seen": 1521549312
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002722367101303912,
      "loss": 3.0373,
      "theoretical_loss": 3.5116527720675395,
      "tokens_seen": 1521614848
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027222668004012033,
      "loss": 2.8643,
      "theoretical_loss": 3.5116394736296326,
      "tokens_seen": 1521680384
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027221664994984957,
      "loss": 2.7753,
      "theoretical_loss": 3.511626175924812,
      "tokens_seen": 1521745920
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027220661985957875,
      "loss": 3.0084,
      "theoretical_loss": 3.511612878953006,
      "tokens_seen": 1521811456
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027219658976930793,
      "loss": 2.7093,
      "theoretical_loss": 3.511599582714142,
      "tokens_seen": 1521876992
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002721865596790371,
      "loss": 2.8532,
      "theoretical_loss": 3.5115862872081487,
      "tokens_seen": 1521942528
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027217652958876635,
      "loss": 2.8312,
      "theoretical_loss": 3.511572992434954,
      "tokens_seen": 1522008064
    },
    {
      "epoch": 0.46,
      "objective/train/advantage_avg": 0.49333882331848145,
      "objective/train/docs_used": 863367,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.563232183456421,
      "objective/train/original_loss": 2.563232421875,
      "objective/train/theoretical_loss": 3.5115596983944855,
      "objective/train/tokens_used": 1542533600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24351578950881958,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505716800689697,
      "objective/train/weighted_lm_loss": 2.6927871704101562,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 3.5115596983944855,
      "tokens_seen": 1522073600
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027216649949849547,
      "loss": 2.8984,
      "theoretical_loss": 3.5115596983944855,
      "tokens_seen": 1522073600
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002721564694082247,
      "loss": 2.7373,
      "theoretical_loss": 3.5115464050866714,
      "tokens_seen": 1522139136
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027214643931795383,
      "loss": 2.6547,
      "theoretical_loss": 3.51153311251144,
      "tokens_seen": 1522204672
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027213640922768307,
      "loss": 2.741,
      "theoretical_loss": 3.51151982066872,
      "tokens_seen": 1522270208
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027212637913741225,
      "loss": 2.7951,
      "theoretical_loss": 3.5115065295584382,
      "tokens_seen": 1522335744
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027211634904714143,
      "loss": 2.7449,
      "theoretical_loss": 3.511493239180524,
      "tokens_seen": 1522401280
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002721063189568706,
      "loss": 3.015,
      "theoretical_loss": 3.5114799495349045,
      "tokens_seen": 1522466816
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002720962888665998,
      "loss": 2.9143,
      "theoretical_loss": 3.511466660621508,
      "tokens_seen": 1522532352
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.000272086258776329,
      "loss": 2.7141,
      "theoretical_loss": 3.511453372440263,
      "tokens_seen": 1522597888
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002720762286860582,
      "loss": 2.9346,
      "theoretical_loss": 3.511440084991098,
      "tokens_seen": 1522663424
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027206619859578734,
      "loss": 2.7144,
      "theoretical_loss": 3.5114267982739404,
      "tokens_seen": 1522728960
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027205616850551657,
      "loss": 2.8158,
      "theoretical_loss": 3.5114135122887182,
      "tokens_seen": 1522794496
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002720461384152457,
      "loss": 2.6147,
      "theoretical_loss": 3.5114002270353604,
      "tokens_seen": 1522860032
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027203610832497493,
      "loss": 2.9684,
      "theoretical_loss": 3.511386942513795,
      "tokens_seen": 1522925568
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002720260782347041,
      "loss": 2.8092,
      "theoretical_loss": 3.51137365872395,
      "tokens_seen": 1522991104
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002720160481444333,
      "loss": 2.6231,
      "theoretical_loss": 3.5113603756657534,
      "tokens_seen": 1523056640
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002720060180541625,
      "loss": 2.9027,
      "theoretical_loss": 3.5113470933391335,
      "tokens_seen": 1523122176
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002719959879638917,
      "loss": 2.7358,
      "theoretical_loss": 3.511333811744019,
      "tokens_seen": 1523187712
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027198595787362084,
      "loss": 2.6514,
      "theoretical_loss": 3.5113205308803384,
      "tokens_seen": 1523253248
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002719759277833501,
      "loss": 2.6844,
      "theoretical_loss": 3.5113072507480183,
      "tokens_seen": 1523318784
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002719658976930792,
      "loss": 2.867,
      "theoretical_loss": 3.511293971346989,
      "tokens_seen": 1523384320
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027195586760280844,
      "loss": 2.7892,
      "theoretical_loss": 3.511280692677177,
      "tokens_seen": 1523449856
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002719458375125376,
      "loss": 2.8521,
      "theoretical_loss": 3.511267414738512,
      "tokens_seen": 1523515392
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002719358074222668,
      "loss": 2.4683,
      "theoretical_loss": 3.5112541375309214,
      "tokens_seen": 1523580928
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.000271925777331996,
      "loss": 2.9374,
      "theoretical_loss": 3.511240861054334,
      "tokens_seen": 1523646464
    },
    {
      "epoch": 0.46,
      "objective/train/advantage_avg": 0.4951799511909485,
      "objective/train/docs_used": 864510,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.271352767944336,
      "objective/train/original_loss": 2.271352529525757,
      "objective/train/theoretical_loss": 3.511227585308678,
      "objective/train/tokens_used": 1544172000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24526016414165497,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050764799118042,
      "objective/train/weighted_lm_loss": 2.3865723609924316,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0396549701690674,
      "theoretical_loss": 3.511227585308678,
      "tokens_seen": 1523712000
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027191574724172516,
      "loss": 2.6772,
      "theoretical_loss": 3.511227585308678,
      "tokens_seen": 1523712000
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027190571715145434,
      "loss": 2.8306,
      "theoretical_loss": 3.511214310293881,
      "tokens_seen": 1523777536
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002718956870611836,
      "loss": 2.9841,
      "theoretical_loss": 3.511201036009872,
      "tokens_seen": 1523843072
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002718856569709127,
      "loss": 2.9283,
      "theoretical_loss": 3.51118776245658,
      "tokens_seen": 1523908608
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027187562688064194,
      "loss": 2.581,
      "theoretical_loss": 3.511174489633933,
      "tokens_seen": 1523974144
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027186559679037107,
      "loss": 2.6344,
      "theoretical_loss": 3.5111612175418583,
      "tokens_seen": 1524039680
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002718555667001003,
      "loss": 2.7775,
      "theoretical_loss": 3.511147946180285,
      "tokens_seen": 1524105216
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002718455366098295,
      "loss": 2.9333,
      "theoretical_loss": 3.5111346755491417,
      "tokens_seen": 1524170752
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027183550651955867,
      "loss": 2.7987,
      "theoretical_loss": 3.511121405648357,
      "tokens_seen": 1524236288
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027182547642928785,
      "loss": 2.8417,
      "theoretical_loss": 3.511108136477858,
      "tokens_seen": 1524301824
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002718154463390171,
      "loss": 2.7309,
      "theoretical_loss": 3.511094868037574,
      "tokens_seen": 1524367360
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002718054162487462,
      "loss": 2.824,
      "theoretical_loss": 3.5110816003274348,
      "tokens_seen": 1524432896
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027179538615847544,
      "loss": 2.8697,
      "theoretical_loss": 3.5110683333473665,
      "tokens_seen": 1524498432
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027178535606820457,
      "loss": 2.9585,
      "theoretical_loss": 3.511055067097298,
      "tokens_seen": 1524563968
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002717753259779338,
      "loss": 2.6074,
      "theoretical_loss": 3.511041801577159,
      "tokens_seen": 1524629504
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027176529588766304,
      "loss": 2.6512,
      "theoretical_loss": 3.511028536786877,
      "tokens_seen": 1524695040
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027175526579739217,
      "loss": 2.7474,
      "theoretical_loss": 3.511015272726381,
      "tokens_seen": 1524760576
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002717452357071214,
      "loss": 2.7081,
      "theoretical_loss": 3.511002009395599,
      "tokens_seen": 1524826112
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027173520561685053,
      "loss": 2.6997,
      "theoretical_loss": 3.5109887467944594,
      "tokens_seen": 1524891648
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027172517552657977,
      "loss": 2.6955,
      "theoretical_loss": 3.5109754849228914,
      "tokens_seen": 1524957184
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027171514543630895,
      "loss": 2.7506,
      "theoretical_loss": 3.5109622237808233,
      "tokens_seen": 1525022720
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027170511534603813,
      "loss": 2.9648,
      "theoretical_loss": 3.510948963368183,
      "tokens_seen": 1525088256
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002716950852557673,
      "loss": 2.72,
      "theoretical_loss": 3.5109357036848996,
      "tokens_seen": 1525153792
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027168505516549655,
      "loss": 2.6035,
      "theoretical_loss": 3.5109224447309013,
      "tokens_seen": 1525219328
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027167502507522567,
      "loss": 2.8642,
      "theoretical_loss": 3.510909186506117,
      "tokens_seen": 1525284864
    },
    {
      "epoch": 0.46,
      "objective/train/advantage_avg": 0.4940264821052551,
      "objective/train/docs_used": 865194,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.9297471046447754,
      "objective/train/original_loss": 2.9297471046447754,
      "objective/train/theoretical_loss": 3.5108959290104753,
      "objective/train/tokens_used": 1545810400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24411864578723907,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506435632705688,
      "objective/train/weighted_lm_loss": 3.0782411098480225,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0363596677780151,
      "theoretical_loss": 3.5108959290104753,
      "tokens_seen": 1525350400
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002716649949849549,
      "loss": 2.9208,
      "theoretical_loss": 3.5108959290104753,
      "tokens_seen": 1525350400
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027165496489468403,
      "loss": 2.5972,
      "theoretical_loss": 3.5108826722439046,
      "tokens_seen": 1525415936
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027164493480441327,
      "loss": 2.7852,
      "theoretical_loss": 3.5108694162063334,
      "tokens_seen": 1525481472
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027163490471414245,
      "loss": 3.0708,
      "theoretical_loss": 3.5108561608976903,
      "tokens_seen": 1525547008
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027162487462387163,
      "loss": 2.6011,
      "theoretical_loss": 3.5108429063179045,
      "tokens_seen": 1525612544
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002716148445336008,
      "loss": 2.8133,
      "theoretical_loss": 3.510829652466904,
      "tokens_seen": 1525678080
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027160481444333,
      "loss": 2.7093,
      "theoretical_loss": 3.510816399344617,
      "tokens_seen": 1525743616
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002715947843530592,
      "loss": 2.7462,
      "theoretical_loss": 3.510803146950973,
      "tokens_seen": 1525809152
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002715847542627884,
      "loss": 2.6931,
      "theoretical_loss": 3.5107898952859,
      "tokens_seen": 1525874688
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027157472417251754,
      "loss": 2.6811,
      "theoretical_loss": 3.510776644349327,
      "tokens_seen": 1525940224
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002715646940822468,
      "loss": 2.6552,
      "theoretical_loss": 3.510763394141183,
      "tokens_seen": 1526005760
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002715546639919759,
      "loss": 2.7525,
      "theoretical_loss": 3.5107501446613965,
      "tokens_seen": 1526071296
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027154463390170514,
      "loss": 2.8423,
      "theoretical_loss": 3.5107368959098957,
      "tokens_seen": 1526136832
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002715346038114343,
      "loss": 2.6826,
      "theoretical_loss": 3.5107236478866097,
      "tokens_seen": 1526202368
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002715245737211635,
      "loss": 2.7838,
      "theoretical_loss": 3.510710400591467,
      "tokens_seen": 1526267904
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002715145436308927,
      "loss": 2.6255,
      "theoretical_loss": 3.5106971540243963,
      "tokens_seen": 1526333440
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002715045135406219,
      "loss": 2.6424,
      "theoretical_loss": 3.5106839081853267,
      "tokens_seen": 1526398976
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027149448345035104,
      "loss": 2.7433,
      "theoretical_loss": 3.510670663074187,
      "tokens_seen": 1526464512
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002714844533600803,
      "loss": 2.6663,
      "theoretical_loss": 3.510657418690905,
      "tokens_seen": 1526530048
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002714744232698094,
      "loss": 2.8141,
      "theoretical_loss": 3.5106441750354103,
      "tokens_seen": 1526595584
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027146439317953864,
      "loss": 2.9364,
      "theoretical_loss": 3.5106309321076314,
      "tokens_seen": 1526661120
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002714543630892678,
      "loss": 2.7797,
      "theoretical_loss": 3.5106176899074972,
      "tokens_seen": 1526726656
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.000271444332998997,
      "loss": 2.7317,
      "theoretical_loss": 3.5106044484349366,
      "tokens_seen": 1526792192
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002714343029087262,
      "loss": 2.9524,
      "theoretical_loss": 3.5105912076898784,
      "tokens_seen": 1526857728
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027142427281845536,
      "loss": 2.8613,
      "theoretical_loss": 3.510577967672251,
      "tokens_seen": 1526923264
    },
    {
      "epoch": 0.46,
      "objective/train/advantage_avg": 0.4948343336582184,
      "objective/train/docs_used": 866480,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.6757280826568604,
      "objective/train/original_loss": 2.6757280826568604,
      "objective/train/theoretical_loss": 3.510564728381983,
      "objective/train/tokens_used": 1547448800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2449209839105606,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507285594940186,
      "objective/train/weighted_lm_loss": 2.8114466667175293,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0396549701690674,
      "theoretical_loss": 3.510564728381983,
      "tokens_seen": 1526988800
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027141424272818454,
      "loss": 2.7418,
      "theoretical_loss": 3.510564728381983,
      "tokens_seen": 1526988800
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002714042126379138,
      "loss": 2.709,
      "theoretical_loss": 3.5105514898190044,
      "tokens_seen": 1527054336
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002713941825476429,
      "loss": 2.6641,
      "theoretical_loss": 3.5105382519832427,
      "tokens_seen": 1527119872
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027138415245737214,
      "loss": 2.8067,
      "theoretical_loss": 3.5105250148746276,
      "tokens_seen": 1527185408
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027137412236710127,
      "loss": 2.5976,
      "theoretical_loss": 3.510511778493088,
      "tokens_seen": 1527250944
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002713640922768305,
      "loss": 2.6662,
      "theoretical_loss": 3.510498542838552,
      "tokens_seen": 1527316480
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002713540621865597,
      "loss": 2.8509,
      "theoretical_loss": 3.5104853079109493,
      "tokens_seen": 1527382016
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027134403209628887,
      "loss": 2.8768,
      "theoretical_loss": 3.510472073710208,
      "tokens_seen": 1527447552
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027133400200601805,
      "loss": 3.05,
      "theoretical_loss": 3.510458840236258,
      "tokens_seen": 1527513088
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002713239719157473,
      "loss": 2.4847,
      "theoretical_loss": 3.510445607489028,
      "tokens_seen": 1527578624
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002713139418254764,
      "loss": 3.0303,
      "theoretical_loss": 3.5104323754684454,
      "tokens_seen": 1527644160
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027130391173520564,
      "loss": 2.8275,
      "theoretical_loss": 3.510419144174441,
      "tokens_seen": 1527709696
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027129388164493477,
      "loss": 2.8409,
      "theoretical_loss": 3.510405913606943,
      "tokens_seen": 1527775232
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.000271283851554664,
      "loss": 2.7793,
      "theoretical_loss": 3.51039268376588,
      "tokens_seen": 1527840768
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002712738214643932,
      "loss": 2.6961,
      "theoretical_loss": 3.510379454651182,
      "tokens_seen": 1527906304
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027126379137412237,
      "loss": 2.6636,
      "theoretical_loss": 3.510366226262777,
      "tokens_seen": 1527971840
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027125376128385155,
      "loss": 2.853,
      "theoretical_loss": 3.510352998600594,
      "tokens_seen": 1528037376
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027124373119358073,
      "loss": 2.6083,
      "theoretical_loss": 3.5103397716645626,
      "tokens_seen": 1528102912
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002712337011033099,
      "loss": 2.5927,
      "theoretical_loss": 3.510326545454611,
      "tokens_seen": 1528168448
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027122367101303915,
      "loss": 2.5378,
      "theoretical_loss": 3.510313319970669,
      "tokens_seen": 1528233984
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002712136409227683,
      "loss": 2.7252,
      "theoretical_loss": 3.5103000952126653,
      "tokens_seen": 1528299520
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002712036108324975,
      "loss": 2.7505,
      "theoretical_loss": 3.510286871180529,
      "tokens_seen": 1528365056
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002711935807422267,
      "loss": 2.7791,
      "theoretical_loss": 3.5102736478741887,
      "tokens_seen": 1528430592
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027118355065195587,
      "loss": 2.7496,
      "theoretical_loss": 3.5102604252935743,
      "tokens_seen": 1528496128
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027117352056168505,
      "loss": 2.7742,
      "theoretical_loss": 3.510247203438614,
      "tokens_seen": 1528561664
    },
    {
      "epoch": 0.46,
      "objective/train/advantage_avg": 0.4931037724018097,
      "objective/train/docs_used": 867112,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.928494691848755,
      "objective/train/original_loss": 2.928494453430176,
      "objective/train/theoretical_loss": 3.510233982309237,
      "objective/train/tokens_used": 1549087200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24325360357761383,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505468845367432,
      "objective/train/weighted_lm_loss": 3.0760929584503174,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0396549701690674,
      "theoretical_loss": 3.510233982309237,
      "tokens_seen": 1528627200
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027116349047141423,
      "loss": 2.8894,
      "theoretical_loss": 3.510233982309237,
      "tokens_seen": 1528627200
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002711534603811434,
      "loss": 2.7752,
      "theoretical_loss": 3.510220761905373,
      "tokens_seen": 1528692736
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027114343029087265,
      "loss": 2.8555,
      "theoretical_loss": 3.5102075422269507,
      "tokens_seen": 1528758272
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002711334002006018,
      "loss": 2.8524,
      "theoretical_loss": 3.510194323273899,
      "tokens_seen": 1528823808
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.000271123370110331,
      "loss": 2.7195,
      "theoretical_loss": 3.510181105046147,
      "tokens_seen": 1528889344
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027111334002006014,
      "loss": 2.9199,
      "theoretical_loss": 3.5101678875436244,
      "tokens_seen": 1528954880
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002711033099297894,
      "loss": 2.8745,
      "theoretical_loss": 3.51015467076626,
      "tokens_seen": 1529020416
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027109327983951856,
      "loss": 2.8429,
      "theoretical_loss": 3.5101414547139824,
      "tokens_seen": 1529085952
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027108324974924774,
      "loss": 2.5868,
      "theoretical_loss": 3.5101282393867215,
      "tokens_seen": 1529151488
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002710732196589769,
      "loss": 2.6664,
      "theoretical_loss": 3.510115024784406,
      "tokens_seen": 1529217024
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002710631895687061,
      "loss": 2.633,
      "theoretical_loss": 3.5101018109069653,
      "tokens_seen": 1529282560
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002710531594784353,
      "loss": 2.8508,
      "theoretical_loss": 3.5100885977543284,
      "tokens_seen": 1529348096
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002710431293881645,
      "loss": 2.9342,
      "theoretical_loss": 3.510075385326425,
      "tokens_seen": 1529413632
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027103309929789364,
      "loss": 2.6901,
      "theoretical_loss": 3.510062173623184,
      "tokens_seen": 1529479168
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002710230692076229,
      "loss": 2.675,
      "theoretical_loss": 3.5100489626445337,
      "tokens_seen": 1529544704
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002710130391173521,
      "loss": 2.6898,
      "theoretical_loss": 3.5100357523904044,
      "tokens_seen": 1529610240
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027100300902708124,
      "loss": 2.7782,
      "theoretical_loss": 3.5100225428607255,
      "tokens_seen": 1529675776
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002709929789368105,
      "loss": 2.5396,
      "theoretical_loss": 3.5100093340554253,
      "tokens_seen": 1529741312
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002709829488465396,
      "loss": 2.5137,
      "theoretical_loss": 3.5099961259744337,
      "tokens_seen": 1529806848
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027097291875626884,
      "loss": 2.6144,
      "theoretical_loss": 3.5099829186176796,
      "tokens_seen": 1529872384
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.000270962888665998,
      "loss": 2.8007,
      "theoretical_loss": 3.509969711985093,
      "tokens_seen": 1529937920
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002709528585757272,
      "loss": 2.495,
      "theoretical_loss": 3.509956506076602,
      "tokens_seen": 1530003456
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002709428284854564,
      "loss": 2.4996,
      "theoretical_loss": 3.509943300892137,
      "tokens_seen": 1530068992
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027093279839518556,
      "loss": 2.6847,
      "theoretical_loss": 3.5099300964316265,
      "tokens_seen": 1530134528
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027092276830491474,
      "loss": 2.9545,
      "theoretical_loss": 3.509916892695,
      "tokens_seen": 1530200064
    },
    {
      "epoch": 0.46,
      "objective/train/advantage_avg": 0.49444639682769775,
      "objective/train/docs_used": 868323,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.6093807220458984,
      "objective/train/original_loss": 2.6093804836273193,
      "objective/train/theoretical_loss": 3.5099036896821874,
      "objective/train/tokens_used": 1550725600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24455079436302185,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506876707077026,
      "objective/train/weighted_lm_loss": 2.7413952350616455,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0439622402191162,
      "theoretical_loss": 3.5099036896821874,
      "tokens_seen": 1530265600
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.000270912738214644,
      "loss": 2.6663,
      "theoretical_loss": 3.5099036896821874,
      "tokens_seen": 1530265600
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002709027081243731,
      "loss": 2.7607,
      "theoretical_loss": 3.509890487393117,
      "tokens_seen": 1530331136
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027089267803410234,
      "loss": 2.8967,
      "theoretical_loss": 3.509877285827719,
      "tokens_seen": 1530396672
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027088264794383147,
      "loss": 2.9353,
      "theoretical_loss": 3.5098640849859226,
      "tokens_seen": 1530462208
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002708726178535607,
      "loss": 2.9312,
      "theoretical_loss": 3.5098508848676566,
      "tokens_seen": 1530527744
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002708625877632899,
      "loss": 2.6215,
      "theoretical_loss": 3.509837685472851,
      "tokens_seen": 1530593280
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027085255767301907,
      "loss": 2.9547,
      "theoretical_loss": 3.5098244868014348,
      "tokens_seen": 1530658816
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027084252758274825,
      "loss": 2.6752,
      "theoretical_loss": 3.5098112888533373,
      "tokens_seen": 1530724352
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002708324974924775,
      "loss": 2.6664,
      "theoretical_loss": 3.5097980916284883,
      "tokens_seen": 1530789888
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002708224674022066,
      "loss": 2.7791,
      "theoretical_loss": 3.509784895126817,
      "tokens_seen": 1530855424
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027081243731193584,
      "loss": 2.5591,
      "theoretical_loss": 3.509771699348253,
      "tokens_seen": 1530920960
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027080240722166497,
      "loss": 2.9712,
      "theoretical_loss": 3.5097585042927255,
      "tokens_seen": 1530986496
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002707923771313942,
      "loss": 2.501,
      "theoretical_loss": 3.509745309960164,
      "tokens_seen": 1531052032
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002707823470411234,
      "loss": 2.445,
      "theoretical_loss": 3.5097321163504978,
      "tokens_seen": 1531117568
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027077231695085257,
      "loss": 2.6386,
      "theoretical_loss": 3.509718923463656,
      "tokens_seen": 1531183104
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027076228686058175,
      "loss": 2.9654,
      "theoretical_loss": 3.5097057312995696,
      "tokens_seen": 1531248640
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027075225677031093,
      "loss": 2.6512,
      "theoretical_loss": 3.5096925398581664,
      "tokens_seen": 1531314176
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002707422266800401,
      "loss": 2.6755,
      "theoretical_loss": 3.5096793491393763,
      "tokens_seen": 1531379712
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027073219658976935,
      "loss": 2.6705,
      "theoretical_loss": 3.5096661591431295,
      "tokens_seen": 1531445248
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002707221664994985,
      "loss": 2.5512,
      "theoretical_loss": 3.5096529698693546,
      "tokens_seen": 1531510784
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002707121364092277,
      "loss": 2.5688,
      "theoretical_loss": 3.509639781317982,
      "tokens_seen": 1531576320
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002707021063189569,
      "loss": 3.018,
      "theoretical_loss": 3.5096265934889406,
      "tokens_seen": 1531641856
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027069207622868607,
      "loss": 2.5688,
      "theoretical_loss": 3.50961340638216,
      "tokens_seen": 1531707392
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027068204613841525,
      "loss": 2.989,
      "theoretical_loss": 3.5096002199975693,
      "tokens_seen": 1531772928
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027067201604814443,
      "loss": 2.6093,
      "theoretical_loss": 3.509587034335099,
      "tokens_seen": 1531838464
    },
    {
      "epoch": 0.46,
      "objective/train/advantage_avg": 0.49288395047187805,
      "objective/train/docs_used": 869121,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.52876615524292,
      "objective/train/original_loss": 2.52876615524292,
      "objective/train/theoretical_loss": 3.5095738493946786,
      "objective/train/tokens_used": 1552364000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24302241206169128,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505237579345703,
      "objective/train/weighted_lm_loss": 2.656526803970337,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0440459251403809,
      "theoretical_loss": 3.5095738493946786,
      "tokens_seen": 1531904000
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002706619859578736,
      "loss": 2.6792,
      "theoretical_loss": 3.5095738493946786,
      "tokens_seen": 1531904000
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027065195586760285,
      "loss": 2.6765,
      "theoretical_loss": 3.5095606651762368,
      "tokens_seen": 1531969536
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.000270641925777332,
      "loss": 2.6379,
      "theoretical_loss": 3.509547481679704,
      "tokens_seen": 1532035072
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002706318956870612,
      "loss": 2.5328,
      "theoretical_loss": 3.509534298905009,
      "tokens_seen": 1532100608
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027062186559679034,
      "loss": 2.8122,
      "theoretical_loss": 3.5095211168520826,
      "tokens_seen": 1532166144
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002706118355065196,
      "loss": 2.7631,
      "theoretical_loss": 3.509507935520853,
      "tokens_seen": 1532231680
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027060180541624876,
      "loss": 2.6672,
      "theoretical_loss": 3.509494754911251,
      "tokens_seen": 1532297216
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027059177532597794,
      "loss": 2.6365,
      "theoretical_loss": 3.509481575023206,
      "tokens_seen": 1532362752
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002705817452357071,
      "loss": 2.6903,
      "theoretical_loss": 3.509468395856647,
      "tokens_seen": 1532428288
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002705717151454363,
      "loss": 2.6208,
      "theoretical_loss": 3.509455217411504,
      "tokens_seen": 1532493824
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002705616850551655,
      "loss": 2.8452,
      "theoretical_loss": 3.5094420396877064,
      "tokens_seen": 1532559360
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002705516549648947,
      "loss": 2.7211,
      "theoretical_loss": 3.509428862685185,
      "tokens_seen": 1532624896
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027054162487462384,
      "loss": 2.706,
      "theoretical_loss": 3.509415686403868,
      "tokens_seen": 1532690432
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002705315947843531,
      "loss": 2.7275,
      "theoretical_loss": 3.509402510843686,
      "tokens_seen": 1532755968
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027052156469408226,
      "loss": 2.9068,
      "theoretical_loss": 3.5093893360045687,
      "tokens_seen": 1532821504
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027051153460381144,
      "loss": 2.8715,
      "theoretical_loss": 3.509376161886445,
      "tokens_seen": 1532887040
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002705015045135406,
      "loss": 2.8191,
      "theoretical_loss": 3.5093629884892454,
      "tokens_seen": 1532952576
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002704914744232698,
      "loss": 2.9869,
      "theoretical_loss": 3.5093498158128997,
      "tokens_seen": 1533018112
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.000270481444332999,
      "loss": 2.7631,
      "theoretical_loss": 3.5093366438573375,
      "tokens_seen": 1533083648
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002704714142427282,
      "loss": 2.6198,
      "theoretical_loss": 3.509323472622488,
      "tokens_seen": 1533149184
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027046138415245735,
      "loss": 2.8188,
      "theoretical_loss": 3.5093103021082817,
      "tokens_seen": 1533214720
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002704513540621866,
      "loss": 2.9362,
      "theoretical_loss": 3.5092971323146473,
      "tokens_seen": 1533280256
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002704413239719157,
      "loss": 3.0984,
      "theoretical_loss": 3.509283963241516,
      "tokens_seen": 1533345792
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027043129388164494,
      "loss": 2.8465,
      "theoretical_loss": 3.509270794888817,
      "tokens_seen": 1533411328
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002704212637913741,
      "loss": 2.7026,
      "theoretical_loss": 3.5092576272564795,
      "tokens_seen": 1533476864
    },
    {
      "epoch": 0.46,
      "objective/train/advantage_avg": 0.49343141913414,
      "objective/train/docs_used": 869825,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.480644941329956,
      "objective/train/original_loss": 2.480644464492798,
      "objective/train/theoretical_loss": 3.5092444603444344,
      "objective/train/tokens_used": 1554002400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2435608208179474,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505812168121338,
      "objective/train/weighted_lm_loss": 2.606034278869629,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0338950157165527,
      "theoretical_loss": 3.5092444603444344,
      "tokens_seen": 1533542400
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002704112337011033,
      "loss": 2.8576,
      "theoretical_loss": 3.5092444603444344,
      "tokens_seen": 1533542400
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002704012036108325,
      "loss": 2.6863,
      "theoretical_loss": 3.5092312941526105,
      "tokens_seen": 1533607936
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027039117352056167,
      "loss": 2.8675,
      "theoretical_loss": 3.509218128680938,
      "tokens_seen": 1533673472
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027038114343029085,
      "loss": 2.6149,
      "theoretical_loss": 3.509204963929347,
      "tokens_seen": 1533739008
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002703711133400201,
      "loss": 2.8903,
      "theoretical_loss": 3.509191799897767,
      "tokens_seen": 1533804544
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002703610832497492,
      "loss": 2.8509,
      "theoretical_loss": 3.5091786365861286,
      "tokens_seen": 1533870080
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027035105315947845,
      "loss": 2.7084,
      "theoretical_loss": 3.5091654739943605,
      "tokens_seen": 1533935616
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027034102306920763,
      "loss": 2.6284,
      "theoretical_loss": 3.5091523121223935,
      "tokens_seen": 1534001152
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002703309929789368,
      "loss": 2.5079,
      "theoretical_loss": 3.509139150970157,
      "tokens_seen": 1534066688
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.000270320962888666,
      "loss": 2.6405,
      "theoretical_loss": 3.509125990537581,
      "tokens_seen": 1534132224
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027031093279839517,
      "loss": 2.6758,
      "theoretical_loss": 3.509112830824596,
      "tokens_seen": 1534197760
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027030090270812435,
      "loss": 2.6877,
      "theoretical_loss": 3.5090996718311307,
      "tokens_seen": 1534263296
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002702908726178536,
      "loss": 2.7612,
      "theoretical_loss": 3.509086513557116,
      "tokens_seen": 1534328832
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.0002702808425275827,
      "loss": 2.7129,
      "theoretical_loss": 3.5090733560024816,
      "tokens_seen": 1534394368
    },
    {
      "epoch": 0.46,
      "learning_rate": 0.00027027081243731195,
      "loss": 2.7953,
      "theoretical_loss": 3.5090601991671573,
      "tokens_seen": 1534459904
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00027026078234704113,
      "loss": 2.6156,
      "theoretical_loss": 3.5090470430510727,
      "tokens_seen": 1534525440
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002702507522567703,
      "loss": 2.6821,
      "theoretical_loss": 3.5090338876541587,
      "tokens_seen": 1534590976
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00027024072216649955,
      "loss": 2.8639,
      "theoretical_loss": 3.509020732976345,
      "tokens_seen": 1534656512
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002702306920762287,
      "loss": 2.8835,
      "theoretical_loss": 3.509007579017561,
      "tokens_seen": 1534722048
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002702206619859579,
      "loss": 2.6337,
      "theoretical_loss": 3.5089944257777375,
      "tokens_seen": 1534787584
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002702106318956871,
      "loss": 2.9053,
      "theoretical_loss": 3.5089812732568038,
      "tokens_seen": 1534853120
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00027020060180541627,
      "loss": 2.8122,
      "theoretical_loss": 3.50896812145469,
      "tokens_seen": 1534918656
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00027019057171514545,
      "loss": 2.6716,
      "theoretical_loss": 3.5089549703713265,
      "tokens_seen": 1534984192
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00027018054162487463,
      "loss": 2.7238,
      "theoretical_loss": 3.508941820006643,
      "tokens_seen": 1535049728
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002701705115346038,
      "loss": 3.1715,
      "theoretical_loss": 3.5089286703605698,
      "tokens_seen": 1535115264
    },
    {
      "epoch": 0.47,
      "objective/train/advantage_avg": 0.49521201848983765,
      "objective/train/docs_used": 871111,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.5455052852630615,
      "objective/train/original_loss": 2.5455050468444824,
      "objective/train/theoretical_loss": 3.508915521433037,
      "objective/train/tokens_used": 1555640800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24529336392879486,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507681369781494,
      "objective/train/weighted_lm_loss": 2.6747798919677734,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0454468727111816,
      "theoretical_loss": 3.508915521433037,
      "tokens_seen": 1535180800
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00027016048144433305,
      "loss": 2.5872,
      "theoretical_loss": 3.508915521433037,
      "tokens_seen": 1535180800
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002701504513540622,
      "loss": 2.7791,
      "theoretical_loss": 3.508902373223975,
      "tokens_seen": 1535246336
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002701404212637914,
      "loss": 2.7506,
      "theoretical_loss": 3.508889225733313,
      "tokens_seen": 1535311872
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00027013039117352054,
      "loss": 2.4522,
      "theoretical_loss": 3.5088760789609807,
      "tokens_seen": 1535377408
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002701203610832498,
      "loss": 2.9923,
      "theoretical_loss": 3.5088629329069096,
      "tokens_seen": 1535442944
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00027011033099297896,
      "loss": 2.7346,
      "theoretical_loss": 3.5088497875710294,
      "tokens_seen": 1535508480
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00027010030090270814,
      "loss": 2.8155,
      "theoretical_loss": 3.50883664295327,
      "tokens_seen": 1535574016
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002700902708124373,
      "loss": 2.5053,
      "theoretical_loss": 3.5088234990535616,
      "tokens_seen": 1535639552
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002700802407221665,
      "loss": 2.6671,
      "theoretical_loss": 3.5088103558718338,
      "tokens_seen": 1535705088
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002700702106318957,
      "loss": 2.9596,
      "theoretical_loss": 3.5087972134080174,
      "tokens_seen": 1535770624
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002700601805416249,
      "loss": 2.6473,
      "theoretical_loss": 3.5087840716620424,
      "tokens_seen": 1535836160
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00027005015045135404,
      "loss": 2.6309,
      "theoretical_loss": 3.508770930633839,
      "tokens_seen": 1535901696
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002700401203610833,
      "loss": 2.6483,
      "theoretical_loss": 3.508757790323337,
      "tokens_seen": 1535967232
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00027003009027081246,
      "loss": 2.8963,
      "theoretical_loss": 3.508744650730467,
      "tokens_seen": 1536032768
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00027002006018054164,
      "loss": 2.7316,
      "theoretical_loss": 3.508731511855159,
      "tokens_seen": 1536098304
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002700100300902708,
      "loss": 2.8497,
      "theoretical_loss": 3.5087183736973437,
      "tokens_seen": 1536163840
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00027,
      "loss": 2.6881,
      "theoretical_loss": 3.508705236256951,
      "tokens_seen": 1536229376
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002699899699097292,
      "loss": 2.7941,
      "theoretical_loss": 3.50869209953391,
      "tokens_seen": 1536294912
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002699799398194584,
      "loss": 2.7937,
      "theoretical_loss": 3.5086789635281526,
      "tokens_seen": 1536360448
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026996990972918755,
      "loss": 2.6767,
      "theoretical_loss": 3.508665828239608,
      "tokens_seen": 1536425984
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002699598796389168,
      "loss": 2.8023,
      "theoretical_loss": 3.508652693668207,
      "tokens_seen": 1536491520
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002699498495486459,
      "loss": 2.8238,
      "theoretical_loss": 3.50863955981388,
      "tokens_seen": 1536557056
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026993981945837514,
      "loss": 2.8672,
      "theoretical_loss": 3.5086264266765568,
      "tokens_seen": 1536622592
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002699297893681043,
      "loss": 2.6436,
      "theoretical_loss": 3.5086132942561674,
      "tokens_seen": 1536688128
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002699197592778335,
      "loss": 2.8302,
      "theoretical_loss": 3.508600162552643,
      "tokens_seen": 1536753664
    },
    {
      "epoch": 0.47,
      "objective/train/advantage_avg": 0.49465492367744446,
      "objective/train/docs_used": 871860,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.722501039505005,
      "objective/train/original_loss": 2.722501039505005,
      "objective/train/theoretical_loss": 3.5085870315659133,
      "objective/train/tokens_used": 1557279200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.244753897190094,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.05070960521698,
      "objective/train/weighted_lm_loss": 2.860301971435547,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 3.5085870315659133,
      "tokens_seen": 1536819200
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002699097291875627,
      "loss": 2.8448,
      "theoretical_loss": 3.5085870315659133,
      "tokens_seen": 1536819200
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026989969909729187,
      "loss": 2.6984,
      "theoretical_loss": 3.508573901295909,
      "tokens_seen": 1536884736
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026988966900702105,
      "loss": 2.863,
      "theoretical_loss": 3.5085607717425598,
      "tokens_seen": 1536950272
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002698796389167503,
      "loss": 2.7026,
      "theoretical_loss": 3.5085476429057962,
      "tokens_seen": 1537015808
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002698696088264794,
      "loss": 2.6229,
      "theoretical_loss": 3.5085345147855493,
      "tokens_seen": 1537081344
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026985957873620865,
      "loss": 2.6852,
      "theoretical_loss": 3.508521387381748,
      "tokens_seen": 1537146880
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026984954864593783,
      "loss": 2.8579,
      "theoretical_loss": 3.5085082606943243,
      "tokens_seen": 1537212416
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.000269839518555667,
      "loss": 2.6708,
      "theoretical_loss": 3.5084951347232076,
      "tokens_seen": 1537277952
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002698294884653962,
      "loss": 2.803,
      "theoretical_loss": 3.5084820094683282,
      "tokens_seen": 1537343488
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026981945837512537,
      "loss": 2.7798,
      "theoretical_loss": 3.5084688849296173,
      "tokens_seen": 1537409024
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026980942828485455,
      "loss": 2.7721,
      "theoretical_loss": 3.508455761107004,
      "tokens_seen": 1537474560
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002697993981945838,
      "loss": 2.7506,
      "theoretical_loss": 3.50844263800042,
      "tokens_seen": 1537540096
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002697893681043129,
      "loss": 2.4171,
      "theoretical_loss": 3.5084295156097953,
      "tokens_seen": 1537605632
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026977933801404215,
      "loss": 2.8223,
      "theoretical_loss": 3.50841639393506,
      "tokens_seen": 1537671168
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002697693079237713,
      "loss": 2.7295,
      "theoretical_loss": 3.5084032729761443,
      "tokens_seen": 1537736704
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002697592778335005,
      "loss": 2.9595,
      "theoretical_loss": 3.5083901527329795,
      "tokens_seen": 1537802240
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002697492477432297,
      "loss": 2.8088,
      "theoretical_loss": 3.508377033205496,
      "tokens_seen": 1537867776
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002697392176529589,
      "loss": 2.8534,
      "theoretical_loss": 3.508363914393623,
      "tokens_seen": 1537933312
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026972918756268806,
      "loss": 2.6523,
      "theoretical_loss": 3.508350796297293,
      "tokens_seen": 1537998848
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002697191574724173,
      "loss": 2.5448,
      "theoretical_loss": 3.508337678916434,
      "tokens_seen": 1538064384
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002697091273821464,
      "loss": 2.7444,
      "theoretical_loss": 3.508324562250979,
      "tokens_seen": 1538129920
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026969909729187565,
      "loss": 2.7418,
      "theoretical_loss": 3.508311446300857,
      "tokens_seen": 1538195456
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002696890672016048,
      "loss": 2.7181,
      "theoretical_loss": 3.508298331065999,
      "tokens_seen": 1538260992
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.000269679037111334,
      "loss": 2.8852,
      "theoretical_loss": 3.5082852165463354,
      "tokens_seen": 1538326528
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002696690070210632,
      "loss": 2.7167,
      "theoretical_loss": 3.5082721027417962,
      "tokens_seen": 1538392064
    },
    {
      "epoch": 0.47,
      "objective/train/advantage_avg": 0.4951346218585968,
      "objective/train/docs_used": 873345,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.694739818572998,
      "objective/train/original_loss": 2.69473934173584,
      "objective/train/theoretical_loss": 3.508258989652313,
      "objective/train/tokens_used": 1558917600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2452174574136734,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.05076003074646,
      "objective/train/weighted_lm_loss": 2.8313560485839844,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0432155132293701,
      "theoretical_loss": 3.508258989652313,
      "tokens_seen": 1538457600
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002696589769307924,
      "loss": 2.8556,
      "theoretical_loss": 3.508258989652313,
      "tokens_seen": 1538457600
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026964894684052156,
      "loss": 2.5493,
      "theoretical_loss": 3.508245877277816,
      "tokens_seen": 1538523136
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026963891675025074,
      "loss": 2.6807,
      "theoretical_loss": 3.508232765618235,
      "tokens_seen": 1538588672
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002696288866599799,
      "loss": 2.8276,
      "theoretical_loss": 3.5082196546735016,
      "tokens_seen": 1538654208
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026961885656970916,
      "loss": 2.6265,
      "theoretical_loss": 3.5082065444435457,
      "tokens_seen": 1538719744
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002696088264794383,
      "loss": 2.7926,
      "theoretical_loss": 3.508193434928298,
      "tokens_seen": 1538785280
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002695987963891675,
      "loss": 2.6822,
      "theoretical_loss": 3.5081803261276896,
      "tokens_seen": 1538850816
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026958876629889665,
      "loss": 2.853,
      "theoretical_loss": 3.5081672180416503,
      "tokens_seen": 1538916352
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002695787362086259,
      "loss": 2.6236,
      "theoretical_loss": 3.5081541106701115,
      "tokens_seen": 1538981888
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026956870611835506,
      "loss": 2.7607,
      "theoretical_loss": 3.5081410040130034,
      "tokens_seen": 1539047424
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026955867602808424,
      "loss": 2.7046,
      "theoretical_loss": 3.508127898070257,
      "tokens_seen": 1539112960
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002695486459378134,
      "loss": 2.987,
      "theoretical_loss": 3.5081147928418024,
      "tokens_seen": 1539178496
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026953861584754266,
      "loss": 2.8429,
      "theoretical_loss": 3.508101688327571,
      "tokens_seen": 1539244032
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002695285857572718,
      "loss": 2.6964,
      "theoretical_loss": 3.508088584527492,
      "tokens_seen": 1539309568
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.000269518555667001,
      "loss": 2.9577,
      "theoretical_loss": 3.508075481441498,
      "tokens_seen": 1539375104
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002695085255767302,
      "loss": 2.7996,
      "theoretical_loss": 3.5080623790695182,
      "tokens_seen": 1539440640
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002694984954864594,
      "loss": 2.8185,
      "theoretical_loss": 3.508049277411484,
      "tokens_seen": 1539506176
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002694884653961886,
      "loss": 2.713,
      "theoretical_loss": 3.5080361764673267,
      "tokens_seen": 1539571712
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026947843530591775,
      "loss": 2.9435,
      "theoretical_loss": 3.5080230762369755,
      "tokens_seen": 1539637248
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.000269468405215647,
      "loss": 2.6599,
      "theoretical_loss": 3.508009976720362,
      "tokens_seen": 1539702784
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002694583751253761,
      "loss": 2.8721,
      "theoretical_loss": 3.507996877917417,
      "tokens_seen": 1539768320
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026944834503510534,
      "loss": 2.7319,
      "theoretical_loss": 3.507983779828071,
      "tokens_seen": 1539833856
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002694383149448345,
      "loss": 2.6258,
      "theoretical_loss": 3.507970682452255,
      "tokens_seen": 1539899392
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002694282848545637,
      "loss": 2.8103,
      "theoretical_loss": 3.5079575857898995,
      "tokens_seen": 1539964928
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002694182547642929,
      "loss": 2.7682,
      "theoretical_loss": 3.507944489840936,
      "tokens_seen": 1540030464
    },
    {
      "epoch": 0.47,
      "objective/train/advantage_avg": 0.49572402238845825,
      "objective/train/docs_used": 874032,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.688263416290283,
      "objective/train/original_loss": 2.688263416290283,
      "objective/train/theoretical_loss": 3.507931394605294,
      "objective/train/tokens_used": 1560556000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24579523503780365,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0508219003677368,
      "objective/train/weighted_lm_loss": 2.824795961380005,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0422987937927246,
      "theoretical_loss": 3.507931394605294,
      "tokens_seen": 1540096000
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026940822467402207,
      "loss": 2.8468,
      "theoretical_loss": 3.507931394605294,
      "tokens_seen": 1540096000
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026939819458375125,
      "loss": 2.7675,
      "theoretical_loss": 3.5079183000829053,
      "tokens_seen": 1540161536
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002693881644934805,
      "loss": 2.7014,
      "theoretical_loss": 3.5079052062737004,
      "tokens_seen": 1540227072
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002693781344032096,
      "loss": 2.6028,
      "theoretical_loss": 3.5078921131776095,
      "tokens_seen": 1540292608
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026936810431293885,
      "loss": 2.9401,
      "theoretical_loss": 3.5078790207945647,
      "tokens_seen": 1540358144
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026935807422266803,
      "loss": 2.6886,
      "theoretical_loss": 3.5078659291244962,
      "tokens_seen": 1540423680
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002693480441323972,
      "loss": 3.0244,
      "theoretical_loss": 3.5078528381673344,
      "tokens_seen": 1540489216
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002693380140421264,
      "loss": 2.5456,
      "theoretical_loss": 3.507839747923011,
      "tokens_seen": 1540554752
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026932798395185557,
      "loss": 2.8635,
      "theoretical_loss": 3.5078266583914557,
      "tokens_seen": 1540620288
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026931795386158475,
      "loss": 2.6494,
      "theoretical_loss": 3.5078135695726007,
      "tokens_seen": 1540685824
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.000269307923771314,
      "loss": 2.9513,
      "theoretical_loss": 3.507800481466376,
      "tokens_seen": 1540751360
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002692978936810431,
      "loss": 2.6999,
      "theoretical_loss": 3.507787394072713,
      "tokens_seen": 1540816896
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026928786359077235,
      "loss": 2.5606,
      "theoretical_loss": 3.507774307391542,
      "tokens_seen": 1540882432
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002692778335005015,
      "loss": 2.5846,
      "theoretical_loss": 3.507761221422795,
      "tokens_seen": 1540947968
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002692678034102307,
      "loss": 2.7332,
      "theoretical_loss": 3.5077481361664016,
      "tokens_seen": 1541013504
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002692577733199599,
      "loss": 2.6955,
      "theoretical_loss": 3.5077350516222934,
      "tokens_seen": 1541079040
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002692477432296891,
      "loss": 2.8426,
      "theoretical_loss": 3.507721967790401,
      "tokens_seen": 1541144576
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026923771313941826,
      "loss": 2.9334,
      "theoretical_loss": 3.507708884670656,
      "tokens_seen": 1541210112
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002692276830491475,
      "loss": 2.6746,
      "theoretical_loss": 3.5076958022629885,
      "tokens_seen": 1541275648
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002692176529588766,
      "loss": 2.8482,
      "theoretical_loss": 3.5076827205673307,
      "tokens_seen": 1541341184
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026920762286860585,
      "loss": 2.7647,
      "theoretical_loss": 3.507669639583612,
      "tokens_seen": 1541406720
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.000269197592778335,
      "loss": 3.0205,
      "theoretical_loss": 3.507656559311765,
      "tokens_seen": 1541472256
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002691875626880642,
      "loss": 2.6954,
      "theoretical_loss": 3.507643479751719,
      "tokens_seen": 1541537792
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002691775325977934,
      "loss": 2.8139,
      "theoretical_loss": 3.5076304009034067,
      "tokens_seen": 1541603328
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002691675025075226,
      "loss": 2.6274,
      "theoretical_loss": 3.507617322766758,
      "tokens_seen": 1541668864
    },
    {
      "epoch": 0.47,
      "objective/train/advantage_avg": 0.4936749041080475,
      "objective/train/docs_used": 874753,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.542046070098877,
      "objective/train/original_loss": 2.542046308517456,
      "objective/train/theoretical_loss": 3.5076042453417045,
      "objective/train/tokens_used": 1562194400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24378274381160736,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506068468093872,
      "objective/train/weighted_lm_loss": 2.6704635620117188,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0432155132293701,
      "theoretical_loss": 3.5076042453417045,
      "tokens_seen": 1541734400
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026915747241725176,
      "loss": 2.7381,
      "theoretical_loss": 3.5076042453417045,
      "tokens_seen": 1541734400
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026914744232698094,
      "loss": 2.5566,
      "theoretical_loss": 3.507591168628177,
      "tokens_seen": 1541799936
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002691374122367101,
      "loss": 2.8027,
      "theoretical_loss": 3.507578092626106,
      "tokens_seen": 1541865472
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026912738214643936,
      "loss": 3.0015,
      "theoretical_loss": 3.5075650173354234,
      "tokens_seen": 1541931008
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002691173520561685,
      "loss": 2.8396,
      "theoretical_loss": 3.5075519427560597,
      "tokens_seen": 1541996544
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002691073219658977,
      "loss": 2.7994,
      "theoretical_loss": 3.5075388688879467,
      "tokens_seen": 1542062080
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026909729187562685,
      "loss": 2.897,
      "theoretical_loss": 3.5075257957310146,
      "tokens_seen": 1542127616
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002690872617853561,
      "loss": 2.7583,
      "theoretical_loss": 3.507512723285195,
      "tokens_seen": 1542193152
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026907723169508526,
      "loss": 2.7202,
      "theoretical_loss": 3.507499651550419,
      "tokens_seen": 1542258688
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026906720160481444,
      "loss": 2.8866,
      "theoretical_loss": 3.507486580526617,
      "tokens_seen": 1542324224
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002690571715145436,
      "loss": 2.7635,
      "theoretical_loss": 3.5074735102137216,
      "tokens_seen": 1542389760
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026904714142427286,
      "loss": 2.5949,
      "theoretical_loss": 3.5074604406116627,
      "tokens_seen": 1542455296
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.000269037111334002,
      "loss": 2.649,
      "theoretical_loss": 3.507447371720372,
      "tokens_seen": 1542520832
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002690270812437312,
      "loss": 2.7702,
      "theoretical_loss": 3.50743430353978,
      "tokens_seen": 1542586368
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026901705115346035,
      "loss": 2.7244,
      "theoretical_loss": 3.5074212360698187,
      "tokens_seen": 1542651904
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002690070210631896,
      "loss": 2.4958,
      "theoretical_loss": 3.507408169310419,
      "tokens_seen": 1542717440
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026899699097291877,
      "loss": 2.7958,
      "theoretical_loss": 3.507395103261511,
      "tokens_seen": 1542782976
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026898696088264795,
      "loss": 2.8161,
      "theoretical_loss": 3.507382037923028,
      "tokens_seen": 1542848512
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026897693079237713,
      "loss": 2.6702,
      "theoretical_loss": 3.5073689732948994,
      "tokens_seen": 1542914048
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002689669007021063,
      "loss": 2.6723,
      "theoretical_loss": 3.5073559093770577,
      "tokens_seen": 1542979584
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002689568706118355,
      "loss": 3.0172,
      "theoretical_loss": 3.5073428461694327,
      "tokens_seen": 1543045120
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002689468405215647,
      "loss": 2.732,
      "theoretical_loss": 3.507329783671957,
      "tokens_seen": 1543110656
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026893681043129385,
      "loss": 3.085,
      "theoretical_loss": 3.507316721884561,
      "tokens_seen": 1543176192
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002689267803410231,
      "loss": 2.6402,
      "theoretical_loss": 3.5073036608071764,
      "tokens_seen": 1543241728
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002689167502507522,
      "loss": 2.8375,
      "theoretical_loss": 3.5072906004397337,
      "tokens_seen": 1543307264
    },
    {
      "epoch": 0.47,
      "objective/train/advantage_avg": 0.4957728683948517,
      "objective/train/docs_used": 875990,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.8719193935394287,
      "objective/train/original_loss": 2.8719193935394287,
      "objective/train/theoretical_loss": 3.507277540782165,
      "objective/train/tokens_used": 1563832800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24584606289863586,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0508270263671875,
      "objective/train/weighted_lm_loss": 3.017897844314575,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 3.507277540782165,
      "tokens_seen": 1543372800
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026890672016048145,
      "loss": 2.8131,
      "theoretical_loss": 3.507277540782165,
      "tokens_seen": 1543372800
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026889669007021063,
      "loss": 3.1464,
      "theoretical_loss": 3.507264481834402,
      "tokens_seen": 1543438336
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002688866599799398,
      "loss": 2.8205,
      "theoretical_loss": 3.507251423596374,
      "tokens_seen": 1543503872
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.000268876629889669,
      "loss": 2.7858,
      "theoretical_loss": 3.507238366068014,
      "tokens_seen": 1543569408
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026886659979939823,
      "loss": 2.7025,
      "theoretical_loss": 3.507225309249254,
      "tokens_seen": 1543634944
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026885656970912736,
      "loss": 2.7094,
      "theoretical_loss": 3.5072122531400227,
      "tokens_seen": 1543700480
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002688465396188566,
      "loss": 2.6676,
      "theoretical_loss": 3.5071991977402535,
      "tokens_seen": 1543766016
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002688365095285857,
      "loss": 2.6237,
      "theoretical_loss": 3.5071861430498767,
      "tokens_seen": 1543831552
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026882647943831495,
      "loss": 2.8065,
      "theoretical_loss": 3.5071730890688246,
      "tokens_seen": 1543897088
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026881644934804413,
      "loss": 2.8624,
      "theoretical_loss": 3.5071600357970283,
      "tokens_seen": 1543962624
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002688064192577733,
      "loss": 2.7788,
      "theoretical_loss": 3.5071469832344184,
      "tokens_seen": 1544028160
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002687963891675025,
      "loss": 2.9634,
      "theoretical_loss": 3.5071339313809267,
      "tokens_seen": 1544093696
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002687863590772317,
      "loss": 2.7632,
      "theoretical_loss": 3.507120880236485,
      "tokens_seen": 1544159232
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026877632898696086,
      "loss": 2.7664,
      "theoretical_loss": 3.507107829801024,
      "tokens_seen": 1544224768
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002687662988966901,
      "loss": 2.8304,
      "theoretical_loss": 3.5070947800744756,
      "tokens_seen": 1544290304
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002687562688064193,
      "loss": 2.8079,
      "theoretical_loss": 3.507081731056771,
      "tokens_seen": 1544355840
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026874623871614846,
      "loss": 2.6982,
      "theoretical_loss": 3.5070686827478417,
      "tokens_seen": 1544421376
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002687362086258777,
      "loss": 2.6775,
      "theoretical_loss": 3.507055635147619,
      "tokens_seen": 1544486912
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002687261785356068,
      "loss": 2.8243,
      "theoretical_loss": 3.5070425882560343,
      "tokens_seen": 1544552448
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026871614844533605,
      "loss": 2.6997,
      "theoretical_loss": 3.5070295420730195,
      "tokens_seen": 1544617984
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002687061183550652,
      "loss": 2.7913,
      "theoretical_loss": 3.5070164965985056,
      "tokens_seen": 1544683520
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002686960882647944,
      "loss": 2.8683,
      "theoretical_loss": 3.507003451832424,
      "tokens_seen": 1544749056
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002686860581745236,
      "loss": 2.6354,
      "theoretical_loss": 3.5069904077747065,
      "tokens_seen": 1544814592
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002686760280842528,
      "loss": 2.8742,
      "theoretical_loss": 3.5069773644252846,
      "tokens_seen": 1544880128
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026866599799398196,
      "loss": 2.7293,
      "theoretical_loss": 3.50696432178409,
      "tokens_seen": 1544945664
    },
    {
      "epoch": 0.47,
      "objective/train/advantage_avg": 0.4951463043689728,
      "objective/train/docs_used": 876526,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.486497640609741,
      "objective/train/original_loss": 2.4864978790283203,
      "objective/train/theoretical_loss": 3.5069512798510534,
      "objective/train/tokens_used": 1565471200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24524113535881042,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050761342048645,
      "objective/train/weighted_lm_loss": 2.612543821334839,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0363596677780151,
      "theoretical_loss": 3.5069512798510534,
      "tokens_seen": 1545011200
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026865596790371114,
      "loss": 2.9172,
      "theoretical_loss": 3.5069512798510534,
      "tokens_seen": 1545011200
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002686459378134403,
      "loss": 2.6125,
      "theoretical_loss": 3.5069382386261068,
      "tokens_seen": 1545076736
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026863590772316956,
      "loss": 2.8979,
      "theoretical_loss": 3.5069251981091814,
      "tokens_seen": 1545142272
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002686258776328987,
      "loss": 2.8449,
      "theoretical_loss": 3.5069121583002096,
      "tokens_seen": 1545207808
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002686158475426279,
      "loss": 2.897,
      "theoretical_loss": 3.5068991191991223,
      "tokens_seen": 1545273344
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026860581745235705,
      "loss": 2.8079,
      "theoretical_loss": 3.5068860808058506,
      "tokens_seen": 1545338880
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002685957873620863,
      "loss": 2.4552,
      "theoretical_loss": 3.506873043120327,
      "tokens_seen": 1545404416
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026858575727181546,
      "loss": 2.5831,
      "theoretical_loss": 3.5068600061424826,
      "tokens_seen": 1545469952
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026857572718154464,
      "loss": 2.6324,
      "theoretical_loss": 3.5068469698722486,
      "tokens_seen": 1545535488
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002685656970912738,
      "loss": 2.6945,
      "theoretical_loss": 3.506833934309558,
      "tokens_seen": 1545601024
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026855566700100306,
      "loss": 2.6169,
      "theoretical_loss": 3.5068208994543406,
      "tokens_seen": 1545666560
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002685456369107322,
      "loss": 2.8543,
      "theoretical_loss": 3.5068078653065293,
      "tokens_seen": 1545732096
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002685356068204614,
      "loss": 2.8064,
      "theoretical_loss": 3.506794831866055,
      "tokens_seen": 1545797632
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026852557673019055,
      "loss": 2.9112,
      "theoretical_loss": 3.50678179913285,
      "tokens_seen": 1545863168
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002685155466399198,
      "loss": 2.6403,
      "theoretical_loss": 3.506768767106845,
      "tokens_seen": 1545928704
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026850551654964897,
      "loss": 2.531,
      "theoretical_loss": 3.506755735787972,
      "tokens_seen": 1545994240
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026849548645937815,
      "loss": 2.6998,
      "theoretical_loss": 3.5067427051761633,
      "tokens_seen": 1546059776
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026848545636910733,
      "loss": 2.7861,
      "theoretical_loss": 3.50672967527135,
      "tokens_seen": 1546125312
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002684754262788365,
      "loss": 2.8116,
      "theoretical_loss": 3.506716646073464,
      "tokens_seen": 1546190848
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002684653961885657,
      "loss": 2.6385,
      "theoretical_loss": 3.5067036175824366,
      "tokens_seen": 1546256384
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002684553660982949,
      "loss": 2.8196,
      "theoretical_loss": 3.5066905897981995,
      "tokens_seen": 1546321920
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026844533600802405,
      "loss": 2.5358,
      "theoretical_loss": 3.506677562720685,
      "tokens_seen": 1546387456
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002684353059177533,
      "loss": 2.5269,
      "theoretical_loss": 3.506664536349825,
      "tokens_seen": 1546452992
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002684252758274824,
      "loss": 2.5755,
      "theoretical_loss": 3.5066515106855496,
      "tokens_seen": 1546518528
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026841524573721165,
      "loss": 2.7938,
      "theoretical_loss": 3.5066384857277924,
      "tokens_seen": 1546584064
    },
    {
      "epoch": 0.47,
      "objective/train/advantage_avg": 0.49399855732917786,
      "objective/train/docs_used": 877732,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.6242401599884033,
      "objective/train/original_loss": 2.624239921569824,
      "objective/train/theoretical_loss": 3.5066254614764842,
      "objective/train/tokens_used": 1567109600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24412542581558228,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506409406661987,
      "objective/train/weighted_lm_loss": 2.757054567337036,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0396549701690674,
      "theoretical_loss": 3.5066254614764842,
      "tokens_seen": 1546649600
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026840521564694083,
      "loss": 2.6408,
      "theoretical_loss": 3.5066254614764842,
      "tokens_seen": 1546649600
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026839518555667,
      "loss": 2.7516,
      "theoretical_loss": 3.5066124379315564,
      "tokens_seen": 1546715136
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002683851554663992,
      "loss": 2.8366,
      "theoretical_loss": 3.506599415092942,
      "tokens_seen": 1546780672
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026837512537612843,
      "loss": 2.7767,
      "theoretical_loss": 3.506586392960572,
      "tokens_seen": 1546846208
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026836509528585756,
      "loss": 2.7079,
      "theoretical_loss": 3.506573371534378,
      "tokens_seen": 1546911744
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002683550651955868,
      "loss": 2.8665,
      "theoretical_loss": 3.5065603508142917,
      "tokens_seen": 1546977280
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002683450351053159,
      "loss": 2.8519,
      "theoretical_loss": 3.5065473308002453,
      "tokens_seen": 1547042816
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026833500501504515,
      "loss": 2.8059,
      "theoretical_loss": 3.506534311492171,
      "tokens_seen": 1547108352
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026832497492477433,
      "loss": 2.7126,
      "theoretical_loss": 3.5065212928899996,
      "tokens_seen": 1547173888
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002683149448345035,
      "loss": 2.9099,
      "theoretical_loss": 3.506508274993664,
      "tokens_seen": 1547239424
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002683049147442327,
      "loss": 2.8588,
      "theoretical_loss": 3.5064952578030955,
      "tokens_seen": 1547304960
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002682948846539619,
      "loss": 2.7539,
      "theoretical_loss": 3.5064822413182255,
      "tokens_seen": 1547370496
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026828485456369106,
      "loss": 2.8075,
      "theoretical_loss": 3.5064692255389867,
      "tokens_seen": 1547436032
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002682748244734203,
      "loss": 2.6912,
      "theoretical_loss": 3.50645621046531,
      "tokens_seen": 1547501568
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002682647943831494,
      "loss": 2.7028,
      "theoretical_loss": 3.5064431960971287,
      "tokens_seen": 1547567104
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026825476429287866,
      "loss": 2.7402,
      "theoretical_loss": 3.5064301824343733,
      "tokens_seen": 1547632640
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002682447342026078,
      "loss": 2.8191,
      "theoretical_loss": 3.5064171694769763,
      "tokens_seen": 1547698176
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.000268234704112337,
      "loss": 2.7275,
      "theoretical_loss": 3.5064041572248694,
      "tokens_seen": 1547763712
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002682246740220662,
      "loss": 2.9674,
      "theoretical_loss": 3.506391145677985,
      "tokens_seen": 1547829248
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002682146439317954,
      "loss": 2.5758,
      "theoretical_loss": 3.506378134836254,
      "tokens_seen": 1547894784
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026820461384152456,
      "loss": 2.93,
      "theoretical_loss": 3.5063651246996095,
      "tokens_seen": 1547960320
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002681945837512538,
      "loss": 2.9591,
      "theoretical_loss": 3.5063521152679833,
      "tokens_seen": 1548025856
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002681845536609829,
      "loss": 2.8436,
      "theoretical_loss": 3.506339106541306,
      "tokens_seen": 1548091392
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026817452357071216,
      "loss": 3.031,
      "theoretical_loss": 3.5063260985195113,
      "tokens_seen": 1548156928
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002681644934804413,
      "loss": 2.7547,
      "theoretical_loss": 3.5063130912025304,
      "tokens_seen": 1548222464
    },
    {
      "epoch": 0.47,
      "objective/train/advantage_avg": 0.49384692311286926,
      "objective/train/docs_used": 878530,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.8567655086517334,
      "objective/train/original_loss": 2.856765031814575,
      "objective/train/theoretical_loss": 3.506300084590295,
      "objective/train/tokens_used": 1568748000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2439466118812561,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506247282028198,
      "objective/train/weighted_lm_loss": 3.0014519691467285,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0363596677780151,
      "theoretical_loss": 3.506300084590295,
      "tokens_seen": 1548288000
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002681544633901705,
      "loss": 2.8219,
      "theoretical_loss": 3.506300084590295,
      "tokens_seen": 1548288000
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002681444332998997,
      "loss": 2.7722,
      "theoretical_loss": 3.5062870786827376,
      "tokens_seen": 1548353536
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002681344032096289,
      "loss": 2.8075,
      "theoretical_loss": 3.5062740734797897,
      "tokens_seen": 1548419072
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026812437311935807,
      "loss": 2.6022,
      "theoretical_loss": 3.506261068981384,
      "tokens_seen": 1548484608
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026811434302908725,
      "loss": 2.4955,
      "theoretical_loss": 3.506248065187452,
      "tokens_seen": 1548550144
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026810431293881643,
      "loss": 2.7508,
      "theoretical_loss": 3.5062350620979252,
      "tokens_seen": 1548615680
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026809428284854566,
      "loss": 2.5999,
      "theoretical_loss": 3.506222059712737,
      "tokens_seen": 1548681216
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002680842527582748,
      "loss": 2.9664,
      "theoretical_loss": 3.5062090580318186,
      "tokens_seen": 1548746752
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.000268074222668004,
      "loss": 2.6739,
      "theoretical_loss": 3.506196057055102,
      "tokens_seen": 1548812288
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026806419257773315,
      "loss": 2.6814,
      "theoretical_loss": 3.5061830567825196,
      "tokens_seen": 1548877824
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002680541624874624,
      "loss": 2.8507,
      "theoretical_loss": 3.5061700572140033,
      "tokens_seen": 1548943360
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026804413239719157,
      "loss": 2.4865,
      "theoretical_loss": 3.506157058349485,
      "tokens_seen": 1549008896
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026803410230692075,
      "loss": 2.8802,
      "theoretical_loss": 3.5061440601888973,
      "tokens_seen": 1549074432
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026802407221664993,
      "loss": 2.6964,
      "theoretical_loss": 3.5061310627321713,
      "tokens_seen": 1549139968
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026801404212637917,
      "loss": 2.7164,
      "theoretical_loss": 3.5061180659792406,
      "tokens_seen": 1549205504
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026800401203610835,
      "loss": 2.7677,
      "theoretical_loss": 3.506105069930036,
      "tokens_seen": 1549271040
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026799398194583753,
      "loss": 2.7534,
      "theoretical_loss": 3.5060920745844903,
      "tokens_seen": 1549336576
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002679839518555667,
      "loss": 2.8931,
      "theoretical_loss": 3.506079079942536,
      "tokens_seen": 1549402112
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002679739217652959,
      "loss": 2.9344,
      "theoretical_loss": 3.506066086004104,
      "tokens_seen": 1549467648
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002679638916750251,
      "loss": 2.8846,
      "theoretical_loss": 3.506053092769127,
      "tokens_seen": 1549533184
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026795386158475425,
      "loss": 2.8446,
      "theoretical_loss": 3.506040100237538,
      "tokens_seen": 1549598720
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002679438314944835,
      "loss": 2.8216,
      "theoretical_loss": 3.5060271084092682,
      "tokens_seen": 1549664256
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002679338014042126,
      "loss": 3.0462,
      "theoretical_loss": 3.5060141172842503,
      "tokens_seen": 1549729792
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026792377131394185,
      "loss": 2.978,
      "theoretical_loss": 3.506001126862416,
      "tokens_seen": 1549795328
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026791374122367103,
      "loss": 2.6899,
      "theoretical_loss": 3.505988137143698,
      "tokens_seen": 1549860864
    },
    {
      "epoch": 0.47,
      "objective/train/advantage_avg": 0.496010422706604,
      "objective/train/docs_used": 879841,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.9122796058654785,
      "objective/train/original_loss": 2.9122798442840576,
      "objective/train/theoretical_loss": 3.5059751481280284,
      "objective/train/tokens_used": 1570386400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24608877301216125,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0508521795272827,
      "objective/train/weighted_lm_loss": 3.060393810272217,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0396549701690674,
      "theoretical_loss": 3.5059751481280284,
      "tokens_seen": 1549926400
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002679037111334002,
      "loss": 2.8974,
      "theoretical_loss": 3.5059751481280284,
      "tokens_seen": 1549926400
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002678936810431294,
      "loss": 2.8389,
      "theoretical_loss": 3.5059621598153394,
      "tokens_seen": 1549991936
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026788365095285863,
      "loss": 2.7912,
      "theoretical_loss": 3.505949172205563,
      "tokens_seen": 1550057472
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026787362086258776,
      "loss": 2.7356,
      "theoretical_loss": 3.5059361852986317,
      "tokens_seen": 1550123008
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.000267863590772317,
      "loss": 2.7008,
      "theoretical_loss": 3.505923199094478,
      "tokens_seen": 1550188544
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002678535606820461,
      "loss": 3.0359,
      "theoretical_loss": 3.5059102135930336,
      "tokens_seen": 1550254080
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026784353059177535,
      "loss": 2.7024,
      "theoretical_loss": 3.5058972287942307,
      "tokens_seen": 1550319616
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026783350050150453,
      "loss": 2.8723,
      "theoretical_loss": 3.5058842446980023,
      "tokens_seen": 1550385152
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002678234704112337,
      "loss": 2.7377,
      "theoretical_loss": 3.5058712613042804,
      "tokens_seen": 1550450688
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002678134403209629,
      "loss": 2.7978,
      "theoretical_loss": 3.5058582786129966,
      "tokens_seen": 1550516224
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002678034102306921,
      "loss": 2.73,
      "theoretical_loss": 3.5058452966240843,
      "tokens_seen": 1550581760
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026779338014042126,
      "loss": 2.8204,
      "theoretical_loss": 3.505832315337475,
      "tokens_seen": 1550647296
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002677833500501505,
      "loss": 2.6224,
      "theoretical_loss": 3.5058193347531015,
      "tokens_seen": 1550712832
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002677733199598796,
      "loss": 2.8014,
      "theoretical_loss": 3.505806354870896,
      "tokens_seen": 1550778368
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026776328986960886,
      "loss": 2.6852,
      "theoretical_loss": 3.505793375690791,
      "tokens_seen": 1550843904
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.000267753259779338,
      "loss": 2.6975,
      "theoretical_loss": 3.5057803972127184,
      "tokens_seen": 1550909440
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002677432296890672,
      "loss": 2.9531,
      "theoretical_loss": 3.505767419436611,
      "tokens_seen": 1550974976
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002677331995987964,
      "loss": 2.8345,
      "theoretical_loss": 3.505754442362401,
      "tokens_seen": 1551040512
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002677231695085256,
      "loss": 2.6462,
      "theoretical_loss": 3.5057414659900203,
      "tokens_seen": 1551106048
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026771313941825476,
      "loss": 3.0516,
      "theoretical_loss": 3.5057284903194024,
      "tokens_seen": 1551171584
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.000267703109327984,
      "loss": 2.7014,
      "theoretical_loss": 3.505715515350479,
      "tokens_seen": 1551237120
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002676930792377131,
      "loss": 2.7237,
      "theoretical_loss": 3.505702541083182,
      "tokens_seen": 1551302656
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026768304914744236,
      "loss": 2.8998,
      "theoretical_loss": 3.5056895675174453,
      "tokens_seen": 1551368192
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002676730190571715,
      "loss": 2.7244,
      "theoretical_loss": 3.5056765946532003,
      "tokens_seen": 1551433728
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002676629889669007,
      "loss": 2.8779,
      "theoretical_loss": 3.5056636224903794,
      "tokens_seen": 1551499264
    },
    {
      "epoch": 0.47,
      "objective/train/advantage_avg": 0.49370697140693665,
      "objective/train/docs_used": 880523,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.6227729320526123,
      "objective/train/original_loss": 2.622772693634033,
      "objective/train/theoretical_loss": 3.5056506510289154,
      "objective/train/tokens_used": 1572024800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24381864070892334,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506101846694946,
      "objective/train/weighted_lm_loss": 2.755453109741211,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0402631759643555,
      "theoretical_loss": 3.5056506510289154,
      "tokens_seen": 1551564800
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002676529588766299,
      "loss": 2.6827,
      "theoretical_loss": 3.5056506510289154,
      "tokens_seen": 1551564800
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002676429287863591,
      "loss": 2.758,
      "theoretical_loss": 3.50563768026874,
      "tokens_seen": 1551630336
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026763289869608827,
      "loss": 2.752,
      "theoretical_loss": 3.505624710209787,
      "tokens_seen": 1551695872
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026762286860581745,
      "loss": 2.7241,
      "theoretical_loss": 3.505611740851988,
      "tokens_seen": 1551761408
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026761283851554663,
      "loss": 2.7695,
      "theoretical_loss": 3.505598772195275,
      "tokens_seen": 1551826944
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026760280842527586,
      "loss": 2.7091,
      "theoretical_loss": 3.5055858042395815,
      "tokens_seen": 1551892480
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.000267592778335005,
      "loss": 2.8316,
      "theoretical_loss": 3.50557283698484,
      "tokens_seen": 1551958016
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002675827482447342,
      "loss": 2.8213,
      "theoretical_loss": 3.5055598704309823,
      "tokens_seen": 1552023552
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026757271815446335,
      "loss": 2.7259,
      "theoretical_loss": 3.5055469045779413,
      "tokens_seen": 1552089088
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002675626880641926,
      "loss": 2.7399,
      "theoretical_loss": 3.50553393942565,
      "tokens_seen": 1552154624
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026755265797392177,
      "loss": 2.8498,
      "theoretical_loss": 3.50552097497404,
      "tokens_seen": 1552220160
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026754262788365095,
      "loss": 2.8446,
      "theoretical_loss": 3.505508011223044,
      "tokens_seen": 1552285696
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026753259779338013,
      "loss": 2.8718,
      "theoretical_loss": 3.5054950481725955,
      "tokens_seen": 1552351232
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026752256770310937,
      "loss": 2.7703,
      "theoretical_loss": 3.5054820858226257,
      "tokens_seen": 1552416768
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002675125376128385,
      "loss": 2.8214,
      "theoretical_loss": 3.505469124173068,
      "tokens_seen": 1552482304
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026750250752256773,
      "loss": 2.6602,
      "theoretical_loss": 3.5054561632238554,
      "tokens_seen": 1552547840
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026749247743229686,
      "loss": 2.9929,
      "theoretical_loss": 3.50544320297492,
      "tokens_seen": 1552613376
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002674824473420261,
      "loss": 2.7703,
      "theoretical_loss": 3.5054302434261944,
      "tokens_seen": 1552678912
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026747241725175527,
      "loss": 2.8477,
      "theoretical_loss": 3.505417284577611,
      "tokens_seen": 1552744448
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026746238716148445,
      "loss": 2.7878,
      "theoretical_loss": 3.5054043264291024,
      "tokens_seen": 1552809984
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026745235707121363,
      "loss": 2.8705,
      "theoretical_loss": 3.505391368980602,
      "tokens_seen": 1552875520
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002674423269809428,
      "loss": 2.9946,
      "theoretical_loss": 3.5053784122320417,
      "tokens_seen": 1552941056
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.000267432296890672,
      "loss": 2.7918,
      "theoretical_loss": 3.5053654561833545,
      "tokens_seen": 1553006592
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026742226680040123,
      "loss": 2.7448,
      "theoretical_loss": 3.505352500834473,
      "tokens_seen": 1553072128
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026741223671013036,
      "loss": 2.7739,
      "theoretical_loss": 3.505339546185329,
      "tokens_seen": 1553137664
    },
    {
      "epoch": 0.47,
      "objective/train/advantage_avg": 0.49552446603775024,
      "objective/train/docs_used": 881131,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.541545867919922,
      "objective/train/original_loss": 2.541545867919922,
      "objective/train/theoretical_loss": 3.505326592235857,
      "objective/train/tokens_used": 1573663200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24559901654720306,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0508010387420654,
      "objective/train/weighted_lm_loss": 2.6704764366149902,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 3.505326592235857,
      "tokens_seen": 1553203200
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002674022066198596,
      "loss": 2.7347,
      "theoretical_loss": 3.505326592235857,
      "tokens_seen": 1553203200
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002673921765295888,
      "loss": 2.6648,
      "theoretical_loss": 3.5053136389859882,
      "tokens_seen": 1553268736
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026738214643931796,
      "loss": 3.0938,
      "theoretical_loss": 3.505300686435656,
      "tokens_seen": 1553334272
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026737211634904714,
      "loss": 2.5485,
      "theoretical_loss": 3.505287734584793,
      "tokens_seen": 1553399808
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002673620862587763,
      "loss": 2.6308,
      "theoretical_loss": 3.5052747834333315,
      "tokens_seen": 1553465344
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002673520561685055,
      "loss": 2.7332,
      "theoretical_loss": 3.505261832981205,
      "tokens_seen": 1553530880
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026734202607823474,
      "loss": 2.7605,
      "theoretical_loss": 3.5052488832283455,
      "tokens_seen": 1553596416
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026733199598796386,
      "loss": 2.6974,
      "theoretical_loss": 3.505235934174686,
      "tokens_seen": 1553661952
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002673219658976931,
      "loss": 2.6669,
      "theoretical_loss": 3.5052229858201596,
      "tokens_seen": 1553727488
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002673119358074222,
      "loss": 2.8535,
      "theoretical_loss": 3.505210038164698,
      "tokens_seen": 1553793024
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026730190571715146,
      "loss": 2.9719,
      "theoretical_loss": 3.5051970912082355,
      "tokens_seen": 1553858560
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026729187562688064,
      "loss": 2.7542,
      "theoretical_loss": 3.5051841449507037,
      "tokens_seen": 1553924096
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002672818455366098,
      "loss": 2.9647,
      "theoretical_loss": 3.505171199392036,
      "tokens_seen": 1553989632
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.000267271815446339,
      "loss": 2.5889,
      "theoretical_loss": 3.505158254532165,
      "tokens_seen": 1554055168
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002672617853560682,
      "loss": 2.9112,
      "theoretical_loss": 3.5051453103710237,
      "tokens_seen": 1554120704
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002672517552657974,
      "loss": 2.843,
      "theoretical_loss": 3.505132366908545,
      "tokens_seen": 1554186240
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002672417251755266,
      "loss": 2.8641,
      "theoretical_loss": 3.505119424144661,
      "tokens_seen": 1554251776
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002672316950852558,
      "loss": 2.8902,
      "theoretical_loss": 3.505106482079305,
      "tokens_seen": 1554317312
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026722166499498496,
      "loss": 2.5789,
      "theoretical_loss": 3.50509354071241,
      "tokens_seen": 1554382848
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002672116349047142,
      "loss": 2.5286,
      "theoretical_loss": 3.5050806000439088,
      "tokens_seen": 1554448384
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002672016048144433,
      "loss": 2.6721,
      "theoretical_loss": 3.5050676600737343,
      "tokens_seen": 1554513920
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026719157472417256,
      "loss": 2.8976,
      "theoretical_loss": 3.505054720801819,
      "tokens_seen": 1554579456
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002671815446339017,
      "loss": 2.9102,
      "theoretical_loss": 3.505041782228096,
      "tokens_seen": 1554644992
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002671715145436309,
      "loss": 2.5033,
      "theoretical_loss": 3.505028844352499,
      "tokens_seen": 1554710528
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002671614844533601,
      "loss": 2.9594,
      "theoretical_loss": 3.5050159071749594,
      "tokens_seen": 1554776064
    },
    {
      "epoch": 0.47,
      "objective/train/advantage_avg": 0.4949467182159424,
      "objective/train/docs_used": 882511,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.465771198272705,
      "objective/train/original_loss": 2.465770959854126,
      "objective/train/theoretical_loss": 3.5050029706954113,
      "objective/train/tokens_used": 1575301600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24502575397491455,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507402420043945,
      "objective/train/weighted_lm_loss": 2.5906286239624023,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0454468727111816,
      "theoretical_loss": 3.5050029706954113,
      "tokens_seen": 1554841600
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002671514543630893,
      "loss": 2.6075,
      "theoretical_loss": 3.5050029706954113,
      "tokens_seen": 1554841600
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026714142427281847,
      "loss": 2.8586,
      "theoretical_loss": 3.504990034913787,
      "tokens_seen": 1554907136
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026713139418254765,
      "loss": 2.7063,
      "theoretical_loss": 3.504977099830019,
      "tokens_seen": 1554972672
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026712136409227683,
      "loss": 2.3911,
      "theoretical_loss": 3.504964165444042,
      "tokens_seen": 1555038208
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026711133400200606,
      "loss": 2.5588,
      "theoretical_loss": 3.5049512317557876,
      "tokens_seen": 1555103744
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002671013039117352,
      "loss": 2.5679,
      "theoretical_loss": 3.5049382987651887,
      "tokens_seen": 1555169280
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002670912738214644,
      "loss": 2.5431,
      "theoretical_loss": 3.504925366472179,
      "tokens_seen": 1555234816
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026708124373119355,
      "loss": 2.5654,
      "theoretical_loss": 3.5049124348766902,
      "tokens_seen": 1555300352
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002670712136409228,
      "loss": 2.9049,
      "theoretical_loss": 3.504899503978657,
      "tokens_seen": 1555365888
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026706118355065197,
      "loss": 2.6925,
      "theoretical_loss": 3.5048865737780113,
      "tokens_seen": 1555431424
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026705115346038115,
      "loss": 2.7832,
      "theoretical_loss": 3.504873644274686,
      "tokens_seen": 1555496960
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026704112337011033,
      "loss": 2.9144,
      "theoretical_loss": 3.504860715468615,
      "tokens_seen": 1555562496
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026703109327983957,
      "loss": 2.84,
      "theoretical_loss": 3.50484778735973,
      "tokens_seen": 1555628032
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002670210631895687,
      "loss": 2.4979,
      "theoretical_loss": 3.504834859947966,
      "tokens_seen": 1555693568
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026701103309929793,
      "loss": 2.7346,
      "theoretical_loss": 3.504821933233254,
      "tokens_seen": 1555759104
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026700100300902706,
      "loss": 3.0877,
      "theoretical_loss": 3.504809007215528,
      "tokens_seen": 1555824640
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002669909729187563,
      "loss": 2.7001,
      "theoretical_loss": 3.504796081894721,
      "tokens_seen": 1555890176
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026698094282848547,
      "loss": 2.935,
      "theoretical_loss": 3.5047831572707664,
      "tokens_seen": 1555955712
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026697091273821465,
      "loss": 2.6851,
      "theoretical_loss": 3.5047702333435966,
      "tokens_seen": 1556021248
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026696088264794383,
      "loss": 2.8362,
      "theoretical_loss": 3.504757310113145,
      "tokens_seen": 1556086784
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.000266950852557673,
      "loss": 2.8876,
      "theoretical_loss": 3.504744387579345,
      "tokens_seen": 1556152320
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002669408224674022,
      "loss": 2.9314,
      "theoretical_loss": 3.5047314657421293,
      "tokens_seen": 1556217856
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026693079237713143,
      "loss": 2.733,
      "theoretical_loss": 3.504718544601431,
      "tokens_seen": 1556283392
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026692076228686056,
      "loss": 2.9307,
      "theoretical_loss": 3.5047056241571837,
      "tokens_seen": 1556348928
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002669107321965898,
      "loss": 2.7292,
      "theoretical_loss": 3.5046927044093197,
      "tokens_seen": 1556414464
    },
    {
      "epoch": 0.47,
      "objective/train/advantage_avg": 0.49403539299964905,
      "objective/train/docs_used": 882978,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.6391587257385254,
      "objective/train/original_loss": 2.6391587257385254,
      "objective/train/theoretical_loss": 3.504679785357773,
      "objective/train/tokens_used": 1576940000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24414904415607452,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506445169448853,
      "objective/train/weighted_lm_loss": 2.772907257080078,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0373471975326538,
      "theoretical_loss": 3.504679785357773,
      "tokens_seen": 1556480000
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.000266900702106319,
      "loss": 2.7728,
      "theoretical_loss": 3.504679785357773,
      "tokens_seen": 1556480000
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026689067201604816,
      "loss": 2.9607,
      "theoretical_loss": 3.5046668670024768,
      "tokens_seen": 1556545536
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026688064192577734,
      "loss": 2.657,
      "theoretical_loss": 3.5046539493433633,
      "tokens_seen": 1556611072
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002668706118355065,
      "loss": 2.8384,
      "theoretical_loss": 3.5046410323803663,
      "tokens_seen": 1556676608
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002668605817452357,
      "loss": 2.8512,
      "theoretical_loss": 3.5046281161134196,
      "tokens_seen": 1556742144
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026685055165496494,
      "loss": 2.9916,
      "theoretical_loss": 3.504615200542455,
      "tokens_seen": 1556807680
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026684052156469406,
      "loss": 2.6984,
      "theoretical_loss": 3.504602285667407,
      "tokens_seen": 1556873216
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002668304914744233,
      "loss": 2.8274,
      "theoretical_loss": 3.5045893714882075,
      "tokens_seen": 1556938752
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002668204613841524,
      "loss": 2.7854,
      "theoretical_loss": 3.5045764580047916,
      "tokens_seen": 1557004288
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026681043129388166,
      "loss": 2.985,
      "theoretical_loss": 3.5045635452170902,
      "tokens_seen": 1557069824
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026680040120361084,
      "loss": 2.7981,
      "theoretical_loss": 3.5045506331250382,
      "tokens_seen": 1557135360
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026679037111334,
      "loss": 2.9296,
      "theoretical_loss": 3.5045377217285685,
      "tokens_seen": 1557200896
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002667803410230692,
      "loss": 3.0206,
      "theoretical_loss": 3.504524811027614,
      "tokens_seen": 1557266432
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002667703109327984,
      "loss": 2.8915,
      "theoretical_loss": 3.5045119010221084,
      "tokens_seen": 1557331968
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026676028084252757,
      "loss": 2.7137,
      "theoretical_loss": 3.504498991711985,
      "tokens_seen": 1557397504
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002667502507522568,
      "loss": 2.6999,
      "theoretical_loss": 3.5044860830971762,
      "tokens_seen": 1557463040
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026674022066198593,
      "loss": 2.6982,
      "theoretical_loss": 3.504473175177617,
      "tokens_seen": 1557528576
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026673019057171516,
      "loss": 2.9195,
      "theoretical_loss": 3.5044602679532386,
      "tokens_seen": 1557594112
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026672016048144434,
      "loss": 2.7709,
      "theoretical_loss": 3.504447361423976,
      "tokens_seen": 1557659648
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002667101303911735,
      "loss": 3.0057,
      "theoretical_loss": 3.5044344555897617,
      "tokens_seen": 1557725184
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002667001003009027,
      "loss": 2.7865,
      "theoretical_loss": 3.5044215504505294,
      "tokens_seen": 1557790720
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002666900702106319,
      "loss": 2.9553,
      "theoretical_loss": 3.504408646006212,
      "tokens_seen": 1557856256
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026668004012036107,
      "loss": 2.6976,
      "theoretical_loss": 3.504395742256743,
      "tokens_seen": 1557921792
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002666700100300903,
      "loss": 2.7508,
      "theoretical_loss": 3.504382839202056,
      "tokens_seen": 1557987328
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026665997993981943,
      "loss": 2.9821,
      "theoretical_loss": 3.5043699368420844,
      "tokens_seen": 1558052864
    },
    {
      "epoch": 0.47,
      "objective/train/advantage_avg": 0.49427396059036255,
      "objective/train/docs_used": 884109,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.7125518321990967,
      "objective/train/original_loss": 2.7125515937805176,
      "objective/train/theoretical_loss": 3.5043570351767617,
      "objective/train/tokens_used": 1578578400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24435773491859436,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050669550895691,
      "objective/train/weighted_lm_loss": 2.8499205112457275,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.04560387134552,
      "theoretical_loss": 3.5043570351767617,
      "tokens_seen": 1558118400
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026664994984954867,
      "loss": 2.8957,
      "theoretical_loss": 3.5043570351767617,
      "tokens_seen": 1558118400
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002666399197592778,
      "loss": 2.8687,
      "theoretical_loss": 3.50434413420602,
      "tokens_seen": 1558183936
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026662988966900703,
      "loss": 2.7194,
      "theoretical_loss": 3.5043312339297947,
      "tokens_seen": 1558249472
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002666198595787362,
      "loss": 2.7461,
      "theoretical_loss": 3.5043183343480178,
      "tokens_seen": 1558315008
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002666098294884654,
      "loss": 2.7559,
      "theoretical_loss": 3.504305435460623,
      "tokens_seen": 1558380544
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026659979939819457,
      "loss": 2.6643,
      "theoretical_loss": 3.504292537267544,
      "tokens_seen": 1558446080
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026658976930792375,
      "loss": 2.8409,
      "theoretical_loss": 3.504279639768714,
      "tokens_seen": 1558511616
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026657973921765293,
      "loss": 2.729,
      "theoretical_loss": 3.5042667429640666,
      "tokens_seen": 1558577152
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026656970912738217,
      "loss": 2.7888,
      "theoretical_loss": 3.504253846853535,
      "tokens_seen": 1558642688
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002665596790371113,
      "loss": 2.88,
      "theoretical_loss": 3.504240951437053,
      "tokens_seen": 1558708224
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026654964894684053,
      "loss": 2.7494,
      "theoretical_loss": 3.5042280567145543,
      "tokens_seen": 1558773760
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002665396188565697,
      "loss": 2.7973,
      "theoretical_loss": 3.504215162685971,
      "tokens_seen": 1558839296
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002665295887662989,
      "loss": 2.6476,
      "theoretical_loss": 3.5042022693512385,
      "tokens_seen": 1558904832
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002665195586760281,
      "loss": 3.0051,
      "theoretical_loss": 3.5041893767102885,
      "tokens_seen": 1558970368
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026650952858575726,
      "loss": 2.6887,
      "theoretical_loss": 3.5041764847630557,
      "tokens_seen": 1559035904
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002664994984954865,
      "loss": 2.9134,
      "theoretical_loss": 3.5041635935094737,
      "tokens_seen": 1559101440
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026648946840521567,
      "loss": 2.8318,
      "theoretical_loss": 3.5041507029494747,
      "tokens_seen": 1559166976
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026647943831494485,
      "loss": 2.7338,
      "theoretical_loss": 3.504137813082994,
      "tokens_seen": 1559232512
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026646940822467403,
      "loss": 2.7564,
      "theoretical_loss": 3.5041249239099637,
      "tokens_seen": 1559298048
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002664593781344032,
      "loss": 2.8178,
      "theoretical_loss": 3.5041120354303184,
      "tokens_seen": 1559363584
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002664493480441324,
      "loss": 2.8276,
      "theoretical_loss": 3.5040991476439904,
      "tokens_seen": 1559429120
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026643931795386163,
      "loss": 2.8405,
      "theoretical_loss": 3.5040862605509147,
      "tokens_seen": 1559494656
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026642928786359076,
      "loss": 2.7121,
      "theoretical_loss": 3.504073374151024,
      "tokens_seen": 1559560192
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026641925777332,
      "loss": 2.8631,
      "theoretical_loss": 3.504060488444252,
      "tokens_seen": 1559625728
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002664092276830492,
      "loss": 2.7866,
      "theoretical_loss": 3.5040476034305326,
      "tokens_seen": 1559691264
    },
    {
      "epoch": 0.47,
      "objective/train/advantage_avg": 0.4938380718231201,
      "objective/train/docs_used": 884823,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.599853038787842,
      "objective/train/original_loss": 2.5998528003692627,
      "objective/train/theoretical_loss": 3.504034719109799,
      "objective/train/tokens_used": 1580216800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24394448101520538,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506237745285034,
      "objective/train/weighted_lm_loss": 2.731429100036621,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0396549701690674,
      "theoretical_loss": 3.504034719109799,
      "tokens_seen": 1559756800
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026639919759277836,
      "loss": 2.9057,
      "theoretical_loss": 3.504034719109799,
      "tokens_seen": 1559756800
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026638916750250754,
      "loss": 2.6979,
      "theoretical_loss": 3.5040218354819856,
      "tokens_seen": 1559822336
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002663791374122367,
      "loss": 2.9647,
      "theoretical_loss": 3.504008952547025,
      "tokens_seen": 1559887872
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002663691073219659,
      "loss": 2.9745,
      "theoretical_loss": 3.5039960703048507,
      "tokens_seen": 1559953408
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026635907723169514,
      "loss": 2.9751,
      "theoretical_loss": 3.503983188755398,
      "tokens_seen": 1560018944
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026634904714142426,
      "loss": 2.7563,
      "theoretical_loss": 3.503970307898599,
      "tokens_seen": 1560084480
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002663390170511535,
      "loss": 2.5815,
      "theoretical_loss": 3.5039574277343877,
      "tokens_seen": 1560150016
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002663289869608826,
      "loss": 2.6918,
      "theoretical_loss": 3.503944548262698,
      "tokens_seen": 1560215552
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026631895687061186,
      "loss": 2.9213,
      "theoretical_loss": 3.5039316694834635,
      "tokens_seen": 1560281088
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026630892678034104,
      "loss": 2.5257,
      "theoretical_loss": 3.503918791396618,
      "tokens_seen": 1560346624
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002662988966900702,
      "loss": 2.8133,
      "theoretical_loss": 3.503905914002095,
      "tokens_seen": 1560412160
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002662888665997994,
      "loss": 2.7438,
      "theoretical_loss": 3.5038930372998283,
      "tokens_seen": 1560477696
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002662788365095286,
      "loss": 2.8933,
      "theoretical_loss": 3.503880161289751,
      "tokens_seen": 1560543232
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026626880641925777,
      "loss": 3.0289,
      "theoretical_loss": 3.5038672859717983,
      "tokens_seen": 1560608768
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.000266258776328987,
      "loss": 2.5571,
      "theoretical_loss": 3.5038544113459027,
      "tokens_seen": 1560674304
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026624874623871613,
      "loss": 2.8485,
      "theoretical_loss": 3.5038415374119984,
      "tokens_seen": 1560739840
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026623871614844536,
      "loss": 2.7289,
      "theoretical_loss": 3.503828664170019,
      "tokens_seen": 1560805376
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026622868605817454,
      "loss": 2.8753,
      "theoretical_loss": 3.503815791619898,
      "tokens_seen": 1560870912
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002662186559679037,
      "loss": 2.8492,
      "theoretical_loss": 3.50380291976157,
      "tokens_seen": 1560936448
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002662086258776329,
      "loss": 2.8686,
      "theoretical_loss": 3.5037900485949676,
      "tokens_seen": 1561001984
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002661985957873621,
      "loss": 2.5812,
      "theoretical_loss": 3.5037771781200258,
      "tokens_seen": 1561067520
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026618856569709127,
      "loss": 2.6843,
      "theoretical_loss": 3.5037643083366774,
      "tokens_seen": 1561133056
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002661785356068205,
      "loss": 2.8531,
      "theoretical_loss": 3.503751439244857,
      "tokens_seen": 1561198592
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026616850551654963,
      "loss": 2.7001,
      "theoretical_loss": 3.503738570844498,
      "tokens_seen": 1561264128
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026615847542627887,
      "loss": 2.8266,
      "theoretical_loss": 3.5037257031355344,
      "tokens_seen": 1561329664
    },
    {
      "epoch": 0.47,
      "objective/train/advantage_avg": 0.49459999799728394,
      "objective/train/docs_used": 885915,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.95155668258667,
      "objective/train/original_loss": 2.9515573978424072,
      "objective/train/theoretical_loss": 3.5037128361178995,
      "objective/train/tokens_used": 1581855200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24468904733657837,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507038831710815,
      "objective/train/weighted_lm_loss": 3.1012492179870605,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0442860126495361,
      "theoretical_loss": 3.5037128361178995,
      "tokens_seen": 1561395200
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.000266148445336008,
      "loss": 2.8743,
      "theoretical_loss": 3.5037128361178995,
      "tokens_seen": 1561395200
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026613841524573723,
      "loss": 2.6513,
      "theoretical_loss": 3.503699969791528,
      "tokens_seen": 1561460736
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002661283851554664,
      "loss": 2.7109,
      "theoretical_loss": 3.503687104156353,
      "tokens_seen": 1561526272
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002661183550651956,
      "loss": 2.8386,
      "theoretical_loss": 3.503674239212309,
      "tokens_seen": 1561591808
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026610832497492477,
      "loss": 2.8619,
      "theoretical_loss": 3.5036613749593295,
      "tokens_seen": 1561657344
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026609829488465395,
      "loss": 2.69,
      "theoretical_loss": 3.5036485113973486,
      "tokens_seen": 1561722880
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026608826479438313,
      "loss": 2.6647,
      "theoretical_loss": 3.5036356485262994,
      "tokens_seen": 1561788416
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026607823470411237,
      "loss": 2.8098,
      "theoretical_loss": 3.503622786346117,
      "tokens_seen": 1561853952
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002660682046138415,
      "loss": 2.8763,
      "theoretical_loss": 3.5036099248567343,
      "tokens_seen": 1561919488
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026605817452357073,
      "loss": 2.9672,
      "theoretical_loss": 3.503597064058086,
      "tokens_seen": 1561985024
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002660481444332999,
      "loss": 2.7539,
      "theoretical_loss": 3.5035842039501057,
      "tokens_seen": 1562050560
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002660381143430291,
      "loss": 2.9755,
      "theoretical_loss": 3.503571344532727,
      "tokens_seen": 1562116096
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002660280842527583,
      "loss": 2.6508,
      "theoretical_loss": 3.5035584858058844,
      "tokens_seen": 1562181632
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026601805416248746,
      "loss": 3.2273,
      "theoretical_loss": 3.503545627769511,
      "tokens_seen": 1562247168
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026600802407221664,
      "loss": 2.9346,
      "theoretical_loss": 3.5035327704235417,
      "tokens_seen": 1562312704
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026599799398194587,
      "loss": 2.7203,
      "theoretical_loss": 3.5035199137679105,
      "tokens_seen": 1562378240
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.000265987963891675,
      "loss": 2.859,
      "theoretical_loss": 3.5035070578025502,
      "tokens_seen": 1562443776
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026597793380140423,
      "loss": 2.902,
      "theoretical_loss": 3.5034942025273965,
      "tokens_seen": 1562509312
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026596790371113336,
      "loss": 2.7805,
      "theoretical_loss": 3.5034813479423823,
      "tokens_seen": 1562574848
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002659578736208626,
      "loss": 2.8749,
      "theoretical_loss": 3.5034684940474414,
      "tokens_seen": 1562640384
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002659478435305918,
      "loss": 2.7419,
      "theoretical_loss": 3.5034556408425086,
      "tokens_seen": 1562705920
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026593781344032096,
      "loss": 3.0917,
      "theoretical_loss": 3.503442788327517,
      "tokens_seen": 1562771456
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026592778335005014,
      "loss": 2.8193,
      "theoretical_loss": 3.5034299365024015,
      "tokens_seen": 1562836992
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002659177532597794,
      "loss": 3.0333,
      "theoretical_loss": 3.503417085367096,
      "tokens_seen": 1562902528
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002659077231695085,
      "loss": 2.3649,
      "theoretical_loss": 3.503404234921534,
      "tokens_seen": 1562968064
    },
    {
      "epoch": 0.47,
      "objective/train/advantage_avg": 0.49427852034568787,
      "objective/train/docs_used": 886454,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.6781134605407715,
      "objective/train/original_loss": 2.6781134605407715,
      "objective/train/theoretical_loss": 3.50339138516565,
      "objective/train/tokens_used": 1583493600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2443818897008896,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506700277328491,
      "objective/train/weighted_lm_loss": 2.8136398792266846,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.044721245765686,
      "theoretical_loss": 3.50339138516565,
      "tokens_seen": 1563033600
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026589769307923774,
      "loss": 2.6456,
      "theoretical_loss": 3.50339138516565,
      "tokens_seen": 1563033600
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026588766298896686,
      "loss": 2.8239,
      "theoretical_loss": 3.5033785360993783,
      "tokens_seen": 1563099136
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002658776328986961,
      "loss": 2.8144,
      "theoretical_loss": 3.5033656877226527,
      "tokens_seen": 1563164672
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002658676028084253,
      "loss": 2.9472,
      "theoretical_loss": 3.503352840035407,
      "tokens_seen": 1563230208
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026585757271815446,
      "loss": 2.8467,
      "theoretical_loss": 3.5033399930375757,
      "tokens_seen": 1563295744
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026584754262788364,
      "loss": 2.7919,
      "theoretical_loss": 3.5033271467290925,
      "tokens_seen": 1563361280
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002658375125376128,
      "loss": 2.7341,
      "theoretical_loss": 3.503314301109892,
      "tokens_seen": 1563426816
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.000265827482447342,
      "loss": 2.875,
      "theoretical_loss": 3.503301456179908,
      "tokens_seen": 1563492352
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026581745235707124,
      "loss": 3.0444,
      "theoretical_loss": 3.503288611939075,
      "tokens_seen": 1563557888
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026580742226680037,
      "loss": 2.716,
      "theoretical_loss": 3.5032757683873266,
      "tokens_seen": 1563623424
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002657973921765296,
      "loss": 2.7962,
      "theoretical_loss": 3.5032629255245977,
      "tokens_seen": 1563688960
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026578736208625873,
      "loss": 2.8036,
      "theoretical_loss": 3.5032500833508218,
      "tokens_seen": 1563754496
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026577733199598797,
      "loss": 2.822,
      "theoretical_loss": 3.503237241865933,
      "tokens_seen": 1563820032
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026576730190571715,
      "loss": 2.8463,
      "theoretical_loss": 3.5032244010698657,
      "tokens_seen": 1563885568
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026575727181544633,
      "loss": 2.7984,
      "theoretical_loss": 3.5032115609625545,
      "tokens_seen": 1563951104
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026574724172517556,
      "loss": 2.552,
      "theoretical_loss": 3.5031987215439333,
      "tokens_seen": 1564016640
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026573721163490474,
      "loss": 2.4885,
      "theoretical_loss": 3.503185882813936,
      "tokens_seen": 1564082176
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002657271815446339,
      "loss": 2.6868,
      "theoretical_loss": 3.503173044772497,
      "tokens_seen": 1564147712
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002657171514543631,
      "loss": 2.8627,
      "theoretical_loss": 3.5031602074195503,
      "tokens_seen": 1564213248
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002657071213640923,
      "loss": 2.8538,
      "theoretical_loss": 3.503147370755031,
      "tokens_seen": 1564278784
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026569709127382147,
      "loss": 2.6144,
      "theoretical_loss": 3.5031345347788725,
      "tokens_seen": 1564344320
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002656870611835507,
      "loss": 2.8417,
      "theoretical_loss": 3.5031216994910093,
      "tokens_seen": 1564409856
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026567703109327983,
      "loss": 2.8411,
      "theoretical_loss": 3.5031088648913755,
      "tokens_seen": 1564475392
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026566700100300907,
      "loss": 2.7425,
      "theoretical_loss": 3.503096030979906,
      "tokens_seen": 1564540928
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002656569709127382,
      "loss": 2.8839,
      "theoretical_loss": 3.503083197756534,
      "tokens_seen": 1564606464
    },
    {
      "epoch": 0.47,
      "objective/train/advantage_avg": 0.49306586384773254,
      "objective/train/docs_used": 887471,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.6218924522399902,
      "objective/train/original_loss": 2.6218926906585693,
      "objective/train/theoretical_loss": 3.5030703652211947,
      "objective/train/tokens_used": 1585132000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2432197481393814,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505428314208984,
      "objective/train/weighted_lm_loss": 2.7542436122894287,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0435127019882202,
      "theoretical_loss": 3.5030703652211947,
      "tokens_seen": 1564672000
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026564694082246743,
      "loss": 2.5688,
      "theoretical_loss": 3.5030703652211947,
      "tokens_seen": 1564672000
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002656369107321966,
      "loss": 2.758,
      "theoretical_loss": 3.5030575333738225,
      "tokens_seen": 1564737536
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002656268806419258,
      "loss": 2.8737,
      "theoretical_loss": 3.5030447022143507,
      "tokens_seen": 1564803072
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026561685055165497,
      "loss": 2.9586,
      "theoretical_loss": 3.5030318717427145,
      "tokens_seen": 1564868608
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026560682046138415,
      "loss": 2.9115,
      "theoretical_loss": 3.503019041958848,
      "tokens_seen": 1564934144
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026559679037111333,
      "loss": 2.6639,
      "theoretical_loss": 3.5030062128626853,
      "tokens_seen": 1564999680
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026558676028084257,
      "loss": 3.0222,
      "theoretical_loss": 3.502993384454161,
      "tokens_seen": 1565065216
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002655767301905717,
      "loss": 2.8532,
      "theoretical_loss": 3.5029805567332097,
      "tokens_seen": 1565130752
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026556670010030093,
      "loss": 2.8886,
      "theoretical_loss": 3.5029677296997646,
      "tokens_seen": 1565196288
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002655566700100301,
      "loss": 3.0883,
      "theoretical_loss": 3.5029549033537615,
      "tokens_seen": 1565261824
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002655466399197593,
      "loss": 2.8249,
      "theoretical_loss": 3.5029420776951343,
      "tokens_seen": 1565327360
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002655366098294885,
      "loss": 2.9179,
      "theoretical_loss": 3.502929252723817,
      "tokens_seen": 1565392896
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026552657973921766,
      "loss": 2.8736,
      "theoretical_loss": 3.502916428439744,
      "tokens_seen": 1565458432
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026551654964894684,
      "loss": 2.8173,
      "theoretical_loss": 3.5029036048428503,
      "tokens_seen": 1565523968
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026550651955867607,
      "loss": 2.9156,
      "theoretical_loss": 3.50289078193307,
      "tokens_seen": 1565589504
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002654964894684052,
      "loss": 2.6902,
      "theoretical_loss": 3.5028779597103377,
      "tokens_seen": 1565655040
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026548645937813443,
      "loss": 2.8582,
      "theoretical_loss": 3.5028651381745872,
      "tokens_seen": 1565720576
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026547642928786356,
      "loss": 2.9232,
      "theoretical_loss": 3.502852317325753,
      "tokens_seen": 1565786112
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002654663991975928,
      "loss": 2.648,
      "theoretical_loss": 3.502839497163771,
      "tokens_seen": 1565851648
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.000265456369107322,
      "loss": 2.7826,
      "theoretical_loss": 3.502826677688574,
      "tokens_seen": 1565917184
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026544633901705116,
      "loss": 2.7754,
      "theoretical_loss": 3.502813858900097,
      "tokens_seen": 1565982720
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026543630892678034,
      "loss": 2.6571,
      "theoretical_loss": 3.502801040798275,
      "tokens_seen": 1566048256
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002654262788365096,
      "loss": 3.0272,
      "theoretical_loss": 3.502788223383041,
      "tokens_seen": 1566113792
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002654162487462387,
      "loss": 2.6741,
      "theoretical_loss": 3.5027754066543313,
      "tokens_seen": 1566179328
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026540621865596794,
      "loss": 2.8277,
      "theoretical_loss": 3.502762590612079,
      "tokens_seen": 1566244864
    },
    {
      "epoch": 0.47,
      "objective/train/advantage_avg": 0.4937269985675812,
      "objective/train/docs_used": 888126,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.428346633911133,
      "objective/train/original_loss": 2.428346872329712,
      "objective/train/theoretical_loss": 3.5027497752562198,
      "objective/train/tokens_used": 1586770400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24383454024791718,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506123304367065,
      "objective/train/weighted_lm_loss": 2.5514943599700928,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0432155132293701,
      "theoretical_loss": 3.5027497752562198,
      "tokens_seen": 1566310400
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026539618856569707,
      "loss": 2.815,
      "theoretical_loss": 3.5027497752562198,
      "tokens_seen": 1566310400
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002653861584754263,
      "loss": 2.7654,
      "theoretical_loss": 3.502736960586687,
      "tokens_seen": 1566375936
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002653761283851555,
      "loss": 2.6805,
      "theoretical_loss": 3.5027241466034167,
      "tokens_seen": 1566441472
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026536609829488466,
      "loss": 2.4873,
      "theoretical_loss": 3.502711333306342,
      "tokens_seen": 1566507008
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026535606820461384,
      "loss": 2.7157,
      "theoretical_loss": 3.502698520695398,
      "tokens_seen": 1566572544
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.000265346038114343,
      "loss": 2.8324,
      "theoretical_loss": 3.502685708770519,
      "tokens_seen": 1566638080
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002653360080240722,
      "loss": 2.8537,
      "theoretical_loss": 3.50267289753164,
      "tokens_seen": 1566703616
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026532597793380144,
      "loss": 2.7406,
      "theoretical_loss": 3.5026600869786955,
      "tokens_seen": 1566769152
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026531594784353057,
      "loss": 2.4893,
      "theoretical_loss": 3.50264727711162,
      "tokens_seen": 1566834688
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002653059177532598,
      "loss": 2.927,
      "theoretical_loss": 3.502634467930348,
      "tokens_seen": 1566900224
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026529588766298893,
      "loss": 3.0878,
      "theoretical_loss": 3.502621659434814,
      "tokens_seen": 1566965760
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026528585757271817,
      "loss": 2.5918,
      "theoretical_loss": 3.5026088516249523,
      "tokens_seen": 1567031296
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026527582748244735,
      "loss": 2.9175,
      "theoretical_loss": 3.5025960445006987,
      "tokens_seen": 1567096832
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026526579739217653,
      "loss": 2.8233,
      "theoretical_loss": 3.5025832380619866,
      "tokens_seen": 1567162368
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002652557673019057,
      "loss": 2.7433,
      "theoretical_loss": 3.502570432308751,
      "tokens_seen": 1567227904
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026524573721163494,
      "loss": 2.554,
      "theoretical_loss": 3.502557627240927,
      "tokens_seen": 1567293440
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026523570712136407,
      "loss": 2.7938,
      "theoretical_loss": 3.5025448228584493,
      "tokens_seen": 1567358976
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.0002652256770310933,
      "loss": 2.6774,
      "theoretical_loss": 3.5025320191612517,
      "tokens_seen": 1567424512
    },
    {
      "epoch": 0.47,
      "learning_rate": 0.00026521564694082243,
      "loss": 2.8166,
      "theoretical_loss": 3.5025192161492695,
      "tokens_seen": 1567490048
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026520561685055167,
      "loss": 2.8588,
      "theoretical_loss": 3.502506413822437,
      "tokens_seen": 1567555584
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026519558676028085,
      "loss": 2.7759,
      "theoretical_loss": 3.5024936121806896,
      "tokens_seen": 1567621120
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026518555667001003,
      "loss": 2.8988,
      "theoretical_loss": 3.5024808112239616,
      "tokens_seen": 1567686656
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002651755265797392,
      "loss": 2.993,
      "theoretical_loss": 3.5024680109521875,
      "tokens_seen": 1567752192
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002651654964894684,
      "loss": 2.8237,
      "theoretical_loss": 3.5024552113653025,
      "tokens_seen": 1567817728
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002651554663991976,
      "loss": 2.5936,
      "theoretical_loss": 3.5024424124632403,
      "tokens_seen": 1567883264
    },
    {
      "epoch": 0.48,
      "objective/train/advantage_avg": 0.4939648509025574,
      "objective/train/docs_used": 889384,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.6358988285064697,
      "objective/train/original_loss": 2.6358983516693115,
      "objective/train/theoretical_loss": 3.502429614245937,
      "objective/train/tokens_used": 1588408800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24409084022045135,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506373643875122,
      "objective/train/weighted_lm_loss": 2.7694091796875,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0304546356201172,
      "theoretical_loss": 3.502429614245937,
      "tokens_seen": 1567948800
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002651454363089268,
      "loss": 2.8315,
      "theoretical_loss": 3.502429614245937,
      "tokens_seen": 1567948800
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026513540621865594,
      "loss": 2.772,
      "theoretical_loss": 3.5024168167133265,
      "tokens_seen": 1568014336
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026512537612838517,
      "loss": 2.9102,
      "theoretical_loss": 3.5024040198653434,
      "tokens_seen": 1568079872
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002651153460381143,
      "loss": 2.8116,
      "theoretical_loss": 3.502391223701923,
      "tokens_seen": 1568145408
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026510531594784353,
      "loss": 2.7152,
      "theoretical_loss": 3.5023784282230004,
      "tokens_seen": 1568210944
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002650952858575727,
      "loss": 2.9429,
      "theoretical_loss": 3.502365633428509,
      "tokens_seen": 1568276480
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002650852557673019,
      "loss": 2.6657,
      "theoretical_loss": 3.5023528393183847,
      "tokens_seen": 1568342016
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002650752256770311,
      "loss": 2.7296,
      "theoretical_loss": 3.502340045892563,
      "tokens_seen": 1568407552
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002650651955867603,
      "loss": 2.8244,
      "theoretical_loss": 3.502327253150977,
      "tokens_seen": 1568473088
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026505516549648944,
      "loss": 2.709,
      "theoretical_loss": 3.5023144610935617,
      "tokens_seen": 1568538624
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002650451354062187,
      "loss": 2.8802,
      "theoretical_loss": 3.5023016697202536,
      "tokens_seen": 1568604160
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002650351053159478,
      "loss": 2.8254,
      "theoretical_loss": 3.502288879030986,
      "tokens_seen": 1568669696
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026502507522567704,
      "loss": 2.7855,
      "theoretical_loss": 3.502276089025694,
      "tokens_seen": 1568735232
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002650150451354062,
      "loss": 2.9249,
      "theoretical_loss": 3.502263299704313,
      "tokens_seen": 1568800768
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002650050150451354,
      "loss": 2.6422,
      "theoretical_loss": 3.502250511066777,
      "tokens_seen": 1568866304
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026499498495486464,
      "loss": 2.8349,
      "theoretical_loss": 3.502237723113022,
      "tokens_seen": 1568931840
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026498495486459376,
      "loss": 2.7661,
      "theoretical_loss": 3.502224935842982,
      "tokens_seen": 1568997376
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.000264974924774323,
      "loss": 2.8824,
      "theoretical_loss": 3.502212149256592,
      "tokens_seen": 1569062912
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002649648946840522,
      "loss": 2.6762,
      "theoretical_loss": 3.502199363353787,
      "tokens_seen": 1569128448
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026495486459378136,
      "loss": 2.9074,
      "theoretical_loss": 3.502186578134502,
      "tokens_seen": 1569193984
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026494483450351054,
      "loss": 2.8605,
      "theoretical_loss": 3.502173793598672,
      "tokens_seen": 1569259520
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002649348044132398,
      "loss": 2.6518,
      "theoretical_loss": 3.5021610097462315,
      "tokens_seen": 1569325056
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002649247743229689,
      "loss": 2.7305,
      "theoretical_loss": 3.502148226577116,
      "tokens_seen": 1569390592
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026491474423269814,
      "loss": 2.9317,
      "theoretical_loss": 3.5021354440912607,
      "tokens_seen": 1569456128
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026490471414242727,
      "loss": 2.8895,
      "theoretical_loss": 3.502122662288599,
      "tokens_seen": 1569521664
    },
    {
      "epoch": 0.48,
      "objective/train/advantage_avg": 0.4939507842063904,
      "objective/train/docs_used": 890057,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.5073723793029785,
      "objective/train/original_loss": 2.5073721408843994,
      "objective/train/theoretical_loss": 3.5021098811690674,
      "objective/train/tokens_used": 1590047200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24407124519348145,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506356954574585,
      "objective/train/weighted_lm_loss": 2.634181499481201,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0425469875335693,
      "theoretical_loss": 3.5021098811690674,
      "tokens_seen": 1569587200
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002648946840521565,
      "loss": 2.7545,
      "theoretical_loss": 3.5021098811690674,
      "tokens_seen": 1569587200
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002648846539618857,
      "loss": 2.8783,
      "theoretical_loss": 3.5020971007326,
      "tokens_seen": 1569652736
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026487462387161486,
      "loss": 2.8964,
      "theoretical_loss": 3.5020843209791326,
      "tokens_seen": 1569718272
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026486459378134404,
      "loss": 2.8281,
      "theoretical_loss": 3.5020715419085997,
      "tokens_seen": 1569783808
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002648545636910732,
      "loss": 2.7298,
      "theoretical_loss": 3.502058763520936,
      "tokens_seen": 1569849344
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002648445336008024,
      "loss": 2.7237,
      "theoretical_loss": 3.5020459858160766,
      "tokens_seen": 1569914880
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026483450351053164,
      "loss": 2.7443,
      "theoretical_loss": 3.502033208793957,
      "tokens_seen": 1569980416
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026482447342026077,
      "loss": 2.6639,
      "theoretical_loss": 3.502020432454512,
      "tokens_seen": 1570045952
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026481444332999,
      "loss": 3.0875,
      "theoretical_loss": 3.5020076567976766,
      "tokens_seen": 1570111488
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026480441323971913,
      "loss": 2.9401,
      "theoretical_loss": 3.5019948818233857,
      "tokens_seen": 1570177024
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026479438314944837,
      "loss": 2.8445,
      "theoretical_loss": 3.5019821075315747,
      "tokens_seen": 1570242560
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026478435305917755,
      "loss": 3.0533,
      "theoretical_loss": 3.5019693339221787,
      "tokens_seen": 1570308096
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026477432296890673,
      "loss": 2.6452,
      "theoretical_loss": 3.501956560995132,
      "tokens_seen": 1570373632
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002647642928786359,
      "loss": 2.7981,
      "theoretical_loss": 3.5019437887503697,
      "tokens_seen": 1570439168
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026475426278836514,
      "loss": 2.5514,
      "theoretical_loss": 3.501931017187828,
      "tokens_seen": 1570504704
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026474423269809427,
      "loss": 2.5858,
      "theoretical_loss": 3.5019182463074414,
      "tokens_seen": 1570570240
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002647342026078235,
      "loss": 2.7969,
      "theoretical_loss": 3.501905476109145,
      "tokens_seen": 1570635776
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026472417251755263,
      "loss": 2.7558,
      "theoretical_loss": 3.5018927065928733,
      "tokens_seen": 1570701312
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026471414242728187,
      "loss": 2.8976,
      "theoretical_loss": 3.501879937758562,
      "tokens_seen": 1570766848
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026470411233701105,
      "loss": 3.0151,
      "theoretical_loss": 3.5018671696061467,
      "tokens_seen": 1570832384
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026469408224674023,
      "loss": 2.9097,
      "theoretical_loss": 3.501854402135562,
      "tokens_seen": 1570897920
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002646840521564694,
      "loss": 2.8551,
      "theoretical_loss": 3.501841635346743,
      "tokens_seen": 1570963456
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002646740220661986,
      "loss": 2.9124,
      "theoretical_loss": 3.501828869239625,
      "tokens_seen": 1571028992
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002646639919759278,
      "loss": 2.8846,
      "theoretical_loss": 3.501816103814143,
      "tokens_seen": 1571094528
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.000264653961885657,
      "loss": 2.9424,
      "theoretical_loss": 3.501803339070232,
      "tokens_seen": 1571160064
    },
    {
      "epoch": 0.48,
      "objective/train/advantage_avg": 0.49483010172843933,
      "objective/train/docs_used": 891281,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.7225375175476074,
      "objective/train/original_loss": 2.7225372791290283,
      "objective/train/theoretical_loss": 3.5017905750078278,
      "objective/train/tokens_used": 1591685600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24495181441307068,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507283210754395,
      "objective/train/weighted_lm_loss": 2.8605968952178955,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0253151655197144,
      "theoretical_loss": 3.5017905750078278,
      "tokens_seen": 1571225600
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026464393179538614,
      "loss": 3.0173,
      "theoretical_loss": 3.5017905750078278,
      "tokens_seen": 1571225600
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026463390170511537,
      "loss": 2.7226,
      "theoretical_loss": 3.501777811626865,
      "tokens_seen": 1571291136
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002646238716148445,
      "loss": 2.7892,
      "theoretical_loss": 3.50176504892728,
      "tokens_seen": 1571356672
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026461384152457373,
      "loss": 2.7,
      "theoretical_loss": 3.5017522869090056,
      "tokens_seen": 1571422208
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002646038114343029,
      "loss": 2.9245,
      "theoretical_loss": 3.5017395255719794,
      "tokens_seen": 1571487744
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002645937813440321,
      "loss": 2.7836,
      "theoretical_loss": 3.5017267649161354,
      "tokens_seen": 1571553280
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002645837512537613,
      "loss": 3.0433,
      "theoretical_loss": 3.5017140049414097,
      "tokens_seen": 1571618816
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002645737211634905,
      "loss": 2.6899,
      "theoretical_loss": 3.5017012456477365,
      "tokens_seen": 1571684352
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026456369107321964,
      "loss": 2.7934,
      "theoretical_loss": 3.5016884870350515,
      "tokens_seen": 1571749888
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002645536609829489,
      "loss": 2.8401,
      "theoretical_loss": 3.5016757291032903,
      "tokens_seen": 1571815424
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.000264543630892678,
      "loss": 3.012,
      "theoretical_loss": 3.5016629718523875,
      "tokens_seen": 1571880960
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026453360080240724,
      "loss": 3.0053,
      "theoretical_loss": 3.501650215282279,
      "tokens_seen": 1571946496
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002645235707121364,
      "loss": 2.692,
      "theoretical_loss": 3.5016374593928994,
      "tokens_seen": 1572012032
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002645135406218656,
      "loss": 2.9706,
      "theoretical_loss": 3.5016247041841853,
      "tokens_seen": 1572077568
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002645035105315948,
      "loss": 2.9775,
      "theoretical_loss": 3.5016119496560707,
      "tokens_seen": 1572143104
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026449348044132396,
      "loss": 2.6374,
      "theoretical_loss": 3.5015991958084913,
      "tokens_seen": 1572208640
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026448345035105314,
      "loss": 2.9692,
      "theoretical_loss": 3.5015864426413827,
      "tokens_seen": 1572274176
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002644734202607824,
      "loss": 2.9045,
      "theoretical_loss": 3.5015736901546797,
      "tokens_seen": 1572339712
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002644633901705115,
      "loss": 2.8063,
      "theoretical_loss": 3.501560938348318,
      "tokens_seen": 1572405248
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026445336008024074,
      "loss": 2.729,
      "theoretical_loss": 3.501548187222233,
      "tokens_seen": 1572470784
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026444332998996987,
      "loss": 2.6469,
      "theoretical_loss": 3.50153543677636,
      "tokens_seen": 1572536320
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002644332998996991,
      "loss": 2.8103,
      "theoretical_loss": 3.5015226870106346,
      "tokens_seen": 1572601856
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002644232698094283,
      "loss": 2.6699,
      "theoretical_loss": 3.5015099379249914,
      "tokens_seen": 1572667392
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026441323971915747,
      "loss": 2.6186,
      "theoretical_loss": 3.501497189519367,
      "tokens_seen": 1572732928
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026440320962888665,
      "loss": 2.9052,
      "theoretical_loss": 3.5014844417936954,
      "tokens_seen": 1572798464
    },
    {
      "epoch": 0.48,
      "objective/train/advantage_avg": 0.49396780133247375,
      "objective/train/docs_used": 891900,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.7724437713623047,
      "objective/train/original_loss": 2.7724437713623047,
      "objective/train/theoretical_loss": 3.501471694747913,
      "objective/train/tokens_used": 1593324000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24408069252967834,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506374835968018,
      "objective/train/weighted_lm_loss": 2.912747621536255,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0425469875335693,
      "theoretical_loss": 3.501471694747913,
      "tokens_seen": 1572864000
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002643931795386159,
      "loss": 2.7696,
      "theoretical_loss": 3.501471694747913,
      "tokens_seen": 1572864000
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.000264383149448345,
      "loss": 2.942,
      "theoretical_loss": 3.5014589483819547,
      "tokens_seen": 1572929536
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026437311935807424,
      "loss": 2.8552,
      "theoretical_loss": 3.5014462026957567,
      "tokens_seen": 1572995072
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026436308926780337,
      "loss": 2.943,
      "theoretical_loss": 3.501433457689253,
      "tokens_seen": 1573060608
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002643530591775326,
      "loss": 3.0133,
      "theoretical_loss": 3.5014207133623803,
      "tokens_seen": 1573126144
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002643430290872618,
      "loss": 2.8107,
      "theoretical_loss": 3.501407969715074,
      "tokens_seen": 1573191680
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026433299899699097,
      "loss": 2.91,
      "theoretical_loss": 3.501395226747269,
      "tokens_seen": 1573257216
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026432296890672015,
      "loss": 2.9576,
      "theoretical_loss": 3.501382484458901,
      "tokens_seen": 1573322752
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026431293881644933,
      "loss": 2.9492,
      "theoretical_loss": 3.501369742849905,
      "tokens_seen": 1573388288
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002643029087261785,
      "loss": 3.0978,
      "theoretical_loss": 3.5013570019202174,
      "tokens_seen": 1573453824
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026429287863590775,
      "loss": 2.7874,
      "theoretical_loss": 3.5013442616697734,
      "tokens_seen": 1573519360
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002642828485456369,
      "loss": 2.7086,
      "theoretical_loss": 3.501331522098508,
      "tokens_seen": 1573584896
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002642728184553661,
      "loss": 2.5908,
      "theoretical_loss": 3.501318783206357,
      "tokens_seen": 1573650432
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026426278836509534,
      "loss": 2.9128,
      "theoretical_loss": 3.501306044993256,
      "tokens_seen": 1573715968
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026425275827482447,
      "loss": 2.9022,
      "theoretical_loss": 3.5012933074591404,
      "tokens_seen": 1573781504
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002642427281845537,
      "loss": 2.8539,
      "theoretical_loss": 3.501280570603946,
      "tokens_seen": 1573847040
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026423269809428283,
      "loss": 2.8581,
      "theoretical_loss": 3.5012678344276082,
      "tokens_seen": 1573912576
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026422266800401207,
      "loss": 2.5501,
      "theoretical_loss": 3.501255098930062,
      "tokens_seen": 1573978112
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026421263791374125,
      "loss": 2.8622,
      "theoretical_loss": 3.501242364111244,
      "tokens_seen": 1574043648
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026420260782347043,
      "loss": 2.8715,
      "theoretical_loss": 3.501229629971089,
      "tokens_seen": 1574109184
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002641925777331996,
      "loss": 3.1704,
      "theoretical_loss": 3.5012168965095327,
      "tokens_seen": 1574174720
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002641825476429288,
      "loss": 2.7255,
      "theoretical_loss": 3.501204163726511,
      "tokens_seen": 1574240256
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.000264172517552658,
      "loss": 2.9545,
      "theoretical_loss": 3.5011914316219594,
      "tokens_seen": 1574305792
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002641624874623872,
      "loss": 2.8277,
      "theoretical_loss": 3.5011787001958137,
      "tokens_seen": 1574371328
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026415245737211634,
      "loss": 2.7539,
      "theoretical_loss": 3.5011659694480084,
      "tokens_seen": 1574436864
    },
    {
      "epoch": 0.48,
      "objective/train/advantage_avg": 0.49172261357307434,
      "objective/train/docs_used": 893256,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.734058141708374,
      "objective/train/original_loss": 2.734058141708374,
      "objective/train/theoretical_loss": 3.5011532393784806,
      "objective/train/tokens_used": 1594962400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24186155200004578,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0504016876220703,
      "objective/train/weighted_lm_loss": 2.8718817234039307,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.046430230140686,
      "theoretical_loss": 3.5011532393784806,
      "tokens_seen": 1574502400
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026414242728184557,
      "loss": 2.7843,
      "theoretical_loss": 3.5011532393784806,
      "tokens_seen": 1574502400
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002641323971915747,
      "loss": 2.8387,
      "theoretical_loss": 3.501140509987165,
      "tokens_seen": 1574567936
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026412236710130393,
      "loss": 2.6103,
      "theoretical_loss": 3.501127781273998,
      "tokens_seen": 1574633472
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002641123370110331,
      "loss": 2.8532,
      "theoretical_loss": 3.501115053238914,
      "tokens_seen": 1574699008
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002641023069207623,
      "loss": 2.8816,
      "theoretical_loss": 3.50110232588185,
      "tokens_seen": 1574764544
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002640922768304915,
      "loss": 2.8603,
      "theoretical_loss": 3.501089599202741,
      "tokens_seen": 1574830080
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002640822467402207,
      "loss": 2.8507,
      "theoretical_loss": 3.5010768732015225,
      "tokens_seen": 1574895616
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026407221664994984,
      "loss": 2.8435,
      "theoretical_loss": 3.5010641478781306,
      "tokens_seen": 1574961152
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002640621865596791,
      "loss": 2.8201,
      "theoretical_loss": 3.501051423232501,
      "tokens_seen": 1575026688
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002640521564694082,
      "loss": 2.7229,
      "theoretical_loss": 3.5010386992645692,
      "tokens_seen": 1575092224
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026404212637913744,
      "loss": 2.8365,
      "theoretical_loss": 3.5010259759742715,
      "tokens_seen": 1575157760
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002640320962888666,
      "loss": 2.7469,
      "theoretical_loss": 3.501013253361542,
      "tokens_seen": 1575223296
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002640220661985958,
      "loss": 2.9209,
      "theoretical_loss": 3.5010005314263184,
      "tokens_seen": 1575288832
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.000264012036108325,
      "loss": 2.6129,
      "theoretical_loss": 3.500987810168535,
      "tokens_seen": 1575354368
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026400200601805416,
      "loss": 2.7834,
      "theoretical_loss": 3.5009750895881284,
      "tokens_seen": 1575419904
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026399197592778334,
      "loss": 2.9052,
      "theoretical_loss": 3.500962369685034,
      "tokens_seen": 1575485440
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002639819458375126,
      "loss": 2.9733,
      "theoretical_loss": 3.5009496504591873,
      "tokens_seen": 1575550976
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002639719157472417,
      "loss": 2.73,
      "theoretical_loss": 3.500936931910525,
      "tokens_seen": 1575616512
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026396188565697094,
      "loss": 2.6905,
      "theoretical_loss": 3.500924214038982,
      "tokens_seen": 1575682048
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026395185556670007,
      "loss": 2.8583,
      "theoretical_loss": 3.5009114968444943,
      "tokens_seen": 1575747584
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002639418254764293,
      "loss": 3.0548,
      "theoretical_loss": 3.500898780326998,
      "tokens_seen": 1575813120
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002639317953861585,
      "loss": 2.9182,
      "theoretical_loss": 3.5008860644864286,
      "tokens_seen": 1575878656
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026392176529588767,
      "loss": 2.9454,
      "theoretical_loss": 3.500873349322722,
      "tokens_seen": 1575944192
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026391173520561685,
      "loss": 2.7509,
      "theoretical_loss": 3.5008606348358136,
      "tokens_seen": 1576009728
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002639017051153461,
      "loss": 2.9096,
      "theoretical_loss": 3.50084792102564,
      "tokens_seen": 1576075264
    },
    {
      "epoch": 0.48,
      "objective/train/advantage_avg": 0.4935598373413086,
      "objective/train/docs_used": 893728,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.9842207431793213,
      "objective/train/original_loss": 2.9842207431793213,
      "objective/train/theoretical_loss": 3.5008352078921368,
      "objective/train/tokens_used": 1596600800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24369187653064728,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505945682525635,
      "objective/train/weighted_lm_loss": 3.1351230144500732,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0428944826126099,
      "theoretical_loss": 3.5008352078921368,
      "tokens_seen": 1576140800
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002638916750250752,
      "loss": 2.9078,
      "theoretical_loss": 3.5008352078921368,
      "tokens_seen": 1576140800
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026388164493480444,
      "loss": 2.8313,
      "theoretical_loss": 3.50082249543524,
      "tokens_seen": 1576206336
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026387161484453357,
      "loss": 2.6554,
      "theoretical_loss": 3.5008097836548844,
      "tokens_seen": 1576271872
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002638615847542628,
      "loss": 2.7145,
      "theoretical_loss": 3.5007970725510074,
      "tokens_seen": 1576337408
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.000263851554663992,
      "loss": 2.877,
      "theoretical_loss": 3.500784362123544,
      "tokens_seen": 1576402944
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026384152457372117,
      "loss": 2.669,
      "theoretical_loss": 3.50077165237243,
      "tokens_seen": 1576468480
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026383149448345035,
      "loss": 2.8521,
      "theoretical_loss": 3.5007589432976016,
      "tokens_seen": 1576534016
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026382146439317953,
      "loss": 2.5829,
      "theoretical_loss": 3.5007462348989944,
      "tokens_seen": 1576599552
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002638114343029087,
      "loss": 2.7604,
      "theoretical_loss": 3.500733527176545,
      "tokens_seen": 1576665088
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026380140421263795,
      "loss": 2.6301,
      "theoretical_loss": 3.500720820130189,
      "tokens_seen": 1576730624
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002637913741223671,
      "loss": 2.8635,
      "theoretical_loss": 3.5007081137598615,
      "tokens_seen": 1576796160
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002637813440320963,
      "loss": 2.7384,
      "theoretical_loss": 3.5006954080655,
      "tokens_seen": 1576861696
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002637713139418255,
      "loss": 2.9328,
      "theoretical_loss": 3.5006827030470387,
      "tokens_seen": 1576927232
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026376128385155467,
      "loss": 2.8726,
      "theoretical_loss": 3.5006699987044154,
      "tokens_seen": 1576992768
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026375125376128385,
      "loss": 2.9255,
      "theoretical_loss": 3.5006572950375645,
      "tokens_seen": 1577058304
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026374122367101303,
      "loss": 2.8066,
      "theoretical_loss": 3.500644592046423,
      "tokens_seen": 1577123840
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002637311935807422,
      "loss": 3.0152,
      "theoretical_loss": 3.5006318897309256,
      "tokens_seen": 1577189376
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026372116349047145,
      "loss": 2.5771,
      "theoretical_loss": 3.50061918809101,
      "tokens_seen": 1577254912
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002637111334002006,
      "loss": 2.8589,
      "theoretical_loss": 3.5006064871266114,
      "tokens_seen": 1577320448
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002637011033099298,
      "loss": 2.9052,
      "theoretical_loss": 3.5005937868376655,
      "tokens_seen": 1577385984
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026369107321965894,
      "loss": 2.8293,
      "theoretical_loss": 3.5005810872241083,
      "tokens_seen": 1577451520
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002636810431293882,
      "loss": 2.7768,
      "theoretical_loss": 3.5005683882858767,
      "tokens_seen": 1577517056
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026367101303911736,
      "loss": 2.6865,
      "theoretical_loss": 3.500555690022906,
      "tokens_seen": 1577582592
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026366098294884654,
      "loss": 2.6685,
      "theoretical_loss": 3.5005429924351317,
      "tokens_seen": 1577648128
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002636509528585757,
      "loss": 3.0368,
      "theoretical_loss": 3.5005302955224913,
      "tokens_seen": 1577713664
    },
    {
      "epoch": 0.48,
      "objective/train/advantage_avg": 0.49495938420295715,
      "objective/train/docs_used": 894313,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.916358709335327,
      "objective/train/original_loss": 2.916358470916748,
      "objective/train/theoretical_loss": 3.50051759928492,
      "objective/train/tokens_used": 1598239200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24505329132080078,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507416725158691,
      "objective/train/weighted_lm_loss": 3.0642027854919434,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0432155132293701,
      "theoretical_loss": 3.50051759928492,
      "tokens_seen": 1577779200
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002636409227683049,
      "loss": 2.8919,
      "theoretical_loss": 3.50051759928492,
      "tokens_seen": 1577779200
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002636308926780341,
      "loss": 2.7786,
      "theoretical_loss": 3.5005049037223537,
      "tokens_seen": 1577844736
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002636208625877633,
      "loss": 3.0578,
      "theoretical_loss": 3.500492208834729,
      "tokens_seen": 1577910272
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026361083249749244,
      "loss": 2.9988,
      "theoretical_loss": 3.500479514621982,
      "tokens_seen": 1577975808
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002636008024072217,
      "loss": 2.9287,
      "theoretical_loss": 3.500466821084048,
      "tokens_seen": 1578041344
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026359077231695086,
      "loss": 2.8744,
      "theoretical_loss": 3.5004541282208637,
      "tokens_seen": 1578106880
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026358074222668004,
      "loss": 2.7125,
      "theoretical_loss": 3.5004414360323652,
      "tokens_seen": 1578172416
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002635707121364092,
      "loss": 3.0678,
      "theoretical_loss": 3.500428744518489,
      "tokens_seen": 1578237952
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002635606820461384,
      "loss": 2.6803,
      "theoretical_loss": 3.5004160536791704,
      "tokens_seen": 1578303488
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002635506519558676,
      "loss": 2.8139,
      "theoretical_loss": 3.500403363514346,
      "tokens_seen": 1578369024
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002635406218655968,
      "loss": 2.9895,
      "theoretical_loss": 3.500390674023952,
      "tokens_seen": 1578434560
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026353059177532595,
      "loss": 2.6059,
      "theoretical_loss": 3.5003779852079244,
      "tokens_seen": 1578500096
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002635205616850552,
      "loss": 2.8699,
      "theoretical_loss": 3.5003652970661996,
      "tokens_seen": 1578565632
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026351053159478436,
      "loss": 2.6361,
      "theoretical_loss": 3.5003526095987136,
      "tokens_seen": 1578631168
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026350050150451354,
      "loss": 2.6499,
      "theoretical_loss": 3.500339922805402,
      "tokens_seen": 1578696704
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002634904714142428,
      "loss": 2.8288,
      "theoretical_loss": 3.5003272366862026,
      "tokens_seen": 1578762240
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002634804413239719,
      "loss": 2.6951,
      "theoretical_loss": 3.5003145512410496,
      "tokens_seen": 1578827776
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026347041123370114,
      "loss": 2.9858,
      "theoretical_loss": 3.5003018664698806,
      "tokens_seen": 1578893312
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026346038114343027,
      "loss": 2.7909,
      "theoretical_loss": 3.500289182372631,
      "tokens_seen": 1578958848
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002634503510531595,
      "loss": 2.8626,
      "theoretical_loss": 3.5002764989492383,
      "tokens_seen": 1579024384
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002634403209628887,
      "loss": 3.0128,
      "theoretical_loss": 3.500263816199637,
      "tokens_seen": 1579089920
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026343029087261787,
      "loss": 2.8875,
      "theoretical_loss": 3.500251134123765,
      "tokens_seen": 1579155456
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026342026078234705,
      "loss": 2.8863,
      "theoretical_loss": 3.500238452721557,
      "tokens_seen": 1579220992
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002634102306920763,
      "loss": 2.8656,
      "theoretical_loss": 3.500225771992951,
      "tokens_seen": 1579286528
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002634002006018054,
      "loss": 2.6875,
      "theoretical_loss": 3.500213091937881,
      "tokens_seen": 1579352064
    },
    {
      "epoch": 0.48,
      "objective/train/advantage_avg": 0.49407002329826355,
      "objective/train/docs_used": 895615,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.885509729385376,
      "objective/train/original_loss": 2.885509729385376,
      "objective/train/theoretical_loss": 3.5002004125562856,
      "objective/train/tokens_used": 1599877600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2441597431898117,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506482124328613,
      "objective/train/weighted_lm_loss": 3.0316929817199707,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0455524921417236,
      "theoretical_loss": 3.5002004125562856,
      "tokens_seen": 1579417600
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026339017051153464,
      "loss": 2.7545,
      "theoretical_loss": 3.5002004125562856,
      "tokens_seen": 1579417600
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026338014042126377,
      "loss": 2.7113,
      "theoretical_loss": 3.500187733848099,
      "tokens_seen": 1579483136
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.000263370110330993,
      "loss": 3.0476,
      "theoretical_loss": 3.5001750558132594,
      "tokens_seen": 1579548672
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002633600802407222,
      "loss": 2.9937,
      "theoretical_loss": 3.5001623784517024,
      "tokens_seen": 1579614208
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026335005015045137,
      "loss": 2.8507,
      "theoretical_loss": 3.5001497017633634,
      "tokens_seen": 1579679744
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026334002006018055,
      "loss": 2.7523,
      "theoretical_loss": 3.50013702574818,
      "tokens_seen": 1579745280
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026332998996990973,
      "loss": 2.8482,
      "theoretical_loss": 3.5001243504060877,
      "tokens_seen": 1579810816
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002633199598796389,
      "loss": 2.9302,
      "theoretical_loss": 3.5001116757370236,
      "tokens_seen": 1579876352
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026330992978936815,
      "loss": 2.9274,
      "theoretical_loss": 3.5000990017409235,
      "tokens_seen": 1579941888
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002632998996990973,
      "loss": 2.7064,
      "theoretical_loss": 3.5000863284177237,
      "tokens_seen": 1580007424
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002632898696088265,
      "loss": 2.8225,
      "theoretical_loss": 3.500073655767361,
      "tokens_seen": 1580072960
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002632798395185557,
      "loss": 2.9246,
      "theoretical_loss": 3.500060983789771,
      "tokens_seen": 1580138496
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026326980942828487,
      "loss": 2.7444,
      "theoretical_loss": 3.500048312484891,
      "tokens_seen": 1580204032
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026325977933801405,
      "loss": 3.0824,
      "theoretical_loss": 3.500035641852657,
      "tokens_seen": 1580269568
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026324974924774323,
      "loss": 2.8488,
      "theoretical_loss": 3.5000229718930047,
      "tokens_seen": 1580335104
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002632397191574724,
      "loss": 2.8012,
      "theoretical_loss": 3.5000103026058715,
      "tokens_seen": 1580400640
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026322968906720165,
      "loss": 2.8116,
      "theoretical_loss": 3.4999976339911942,
      "tokens_seen": 1580466176
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002632196589769308,
      "loss": 2.8914,
      "theoretical_loss": 3.499984966048908,
      "tokens_seen": 1580531712
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026320962888666,
      "loss": 2.7748,
      "theoretical_loss": 3.4999722987789497,
      "tokens_seen": 1580597248
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026319959879638914,
      "loss": 2.7127,
      "theoretical_loss": 3.4999596321812563,
      "tokens_seen": 1580662784
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002631895687061184,
      "loss": 2.8671,
      "theoretical_loss": 3.4999469662557634,
      "tokens_seen": 1580728320
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026317953861584756,
      "loss": 3.2453,
      "theoretical_loss": 3.4999343010024084,
      "tokens_seen": 1580793856
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026316950852557674,
      "loss": 3.1117,
      "theoretical_loss": 3.499921636421127,
      "tokens_seen": 1580859392
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002631594784353059,
      "loss": 2.8039,
      "theoretical_loss": 3.499908972511856,
      "tokens_seen": 1580924928
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002631494483450351,
      "loss": 2.871,
      "theoretical_loss": 3.499896309274532,
      "tokens_seen": 1580990464
    },
    {
      "epoch": 0.48,
      "objective/train/advantage_avg": 0.49540871381759644,
      "objective/train/docs_used": 896135,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.2597808837890625,
      "objective/train/original_loss": 2.2597808837890625,
      "objective/train/theoretical_loss": 3.499883646709091,
      "objective/train/tokens_used": 1601516000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24547143280506134,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507886409759521,
      "objective/train/weighted_lm_loss": 2.374396800994873,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.04560387134552,
      "theoretical_loss": 3.499883646709091,
      "tokens_seen": 1581056000
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002631394182547643,
      "loss": 2.7884,
      "theoretical_loss": 3.499883646709091,
      "tokens_seen": 1581056000
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002631293881644935,
      "loss": 3.0542,
      "theoretical_loss": 3.4998709848154697,
      "tokens_seen": 1581121536
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026311935807422264,
      "loss": 2.9612,
      "theoretical_loss": 3.499858323593605,
      "tokens_seen": 1581187072
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002631093279839519,
      "loss": 2.9854,
      "theoretical_loss": 3.4998456630434336,
      "tokens_seen": 1581252608
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026309929789368106,
      "loss": 3.0675,
      "theoretical_loss": 3.499833003164891,
      "tokens_seen": 1581318144
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026308926780341024,
      "loss": 2.7417,
      "theoretical_loss": 3.4998203439579147,
      "tokens_seen": 1581383680
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002630792377131394,
      "loss": 2.7382,
      "theoretical_loss": 3.499807685422441,
      "tokens_seen": 1581449216
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002630692076228686,
      "loss": 2.9387,
      "theoretical_loss": 3.499795027558406,
      "tokens_seen": 1581514752
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002630591775325978,
      "loss": 2.8202,
      "theoretical_loss": 3.4997823703657467,
      "tokens_seen": 1581580288
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.000263049147442327,
      "loss": 3.3114,
      "theoretical_loss": 3.4997697138444,
      "tokens_seen": 1581645824
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026303911735205615,
      "loss": 2.9083,
      "theoretical_loss": 3.4997570579943016,
      "tokens_seen": 1581711360
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002630290872617854,
      "loss": 2.8396,
      "theoretical_loss": 3.499744402815389,
      "tokens_seen": 1581776896
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002630190571715145,
      "loss": 2.7175,
      "theoretical_loss": 3.499731748307598,
      "tokens_seen": 1581842432
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026300902708124374,
      "loss": 2.9362,
      "theoretical_loss": 3.4997190944708656,
      "tokens_seen": 1581907968
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002629989969909729,
      "loss": 2.8893,
      "theoretical_loss": 3.499706441305128,
      "tokens_seen": 1581973504
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002629889669007021,
      "loss": 2.855,
      "theoretical_loss": 3.4996937888103226,
      "tokens_seen": 1582039040
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002629789368104313,
      "loss": 2.6146,
      "theoretical_loss": 3.4996811369863856,
      "tokens_seen": 1582104576
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026296890672016047,
      "loss": 3.129,
      "theoretical_loss": 3.499668485833254,
      "tokens_seen": 1582170112
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026295887662988965,
      "loss": 2.8185,
      "theoretical_loss": 3.4996558353508638,
      "tokens_seen": 1582235648
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002629488465396189,
      "loss": 2.7319,
      "theoretical_loss": 3.499643185539152,
      "tokens_seen": 1582301184
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.000262938816449348,
      "loss": 2.8085,
      "theoretical_loss": 3.499630536398055,
      "tokens_seen": 1582366720
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026292878635907725,
      "loss": 2.8347,
      "theoretical_loss": 3.49961788792751,
      "tokens_seen": 1582432256
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026291875626880643,
      "loss": 2.9861,
      "theoretical_loss": 3.4996052401274538,
      "tokens_seen": 1582497792
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002629087261785356,
      "loss": 2.6706,
      "theoretical_loss": 3.499592592997822,
      "tokens_seen": 1582563328
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002628986960882648,
      "loss": 3.0574,
      "theoretical_loss": 3.499579946538552,
      "tokens_seen": 1582628864
    },
    {
      "epoch": 0.48,
      "objective/train/advantage_avg": 0.4934535324573517,
      "objective/train/docs_used": 897089,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.367469072341919,
      "objective/train/original_loss": 2.367469072341919,
      "objective/train/theoretical_loss": 3.4995673007495816,
      "objective/train/tokens_used": 1603154400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24356147646903992,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505833625793457,
      "objective/train/weighted_lm_loss": 2.4873950481414795,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0442860126495361,
      "theoretical_loss": 3.4995673007495816,
      "tokens_seen": 1582694400
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026288866599799397,
      "loss": 2.7954,
      "theoretical_loss": 3.4995673007495816,
      "tokens_seen": 1582694400
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026287863590772315,
      "loss": 3.0997,
      "theoretical_loss": 3.4995546556308454,
      "tokens_seen": 1582759936
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002628686058174524,
      "loss": 3.0769,
      "theoretical_loss": 3.4995420111822813,
      "tokens_seen": 1582825472
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002628585757271815,
      "loss": 2.7785,
      "theoretical_loss": 3.499529367403826,
      "tokens_seen": 1582891008
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026284854563691075,
      "loss": 2.8402,
      "theoretical_loss": 3.4995167242954164,
      "tokens_seen": 1582956544
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002628385155466399,
      "loss": 2.6725,
      "theoretical_loss": 3.4995040818569887,
      "tokens_seen": 1583022080
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002628284854563691,
      "loss": 2.8576,
      "theoretical_loss": 3.49949144008848,
      "tokens_seen": 1583087616
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002628184553660983,
      "loss": 2.803,
      "theoretical_loss": 3.499478798989827,
      "tokens_seen": 1583153152
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002628084252758275,
      "loss": 2.9429,
      "theoretical_loss": 3.499466158560966,
      "tokens_seen": 1583218688
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026279839518555666,
      "loss": 3.0206,
      "theoretical_loss": 3.499453518801835,
      "tokens_seen": 1583284224
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002627883650952859,
      "loss": 2.8843,
      "theoretical_loss": 3.4994408797123704,
      "tokens_seen": 1583349760
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.000262778335005015,
      "loss": 2.9472,
      "theoretical_loss": 3.4994282412925077,
      "tokens_seen": 1583415296
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026276830491474425,
      "loss": 2.9261,
      "theoretical_loss": 3.4994156035421855,
      "tokens_seen": 1583480832
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026275827482447343,
      "loss": 2.8803,
      "theoretical_loss": 3.4994029664613393,
      "tokens_seen": 1583546368
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002627482447342026,
      "loss": 2.8818,
      "theoretical_loss": 3.499390330049907,
      "tokens_seen": 1583611904
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026273821464393185,
      "loss": 2.818,
      "theoretical_loss": 3.4993776943078245,
      "tokens_seen": 1583677440
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.000262728184553661,
      "loss": 2.8624,
      "theoretical_loss": 3.4993650592350294,
      "tokens_seen": 1583742976
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002627181544633902,
      "loss": 2.9293,
      "theoretical_loss": 3.499352424831458,
      "tokens_seen": 1583808512
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026270812437311934,
      "loss": 3.0789,
      "theoretical_loss": 3.4993397910970474,
      "tokens_seen": 1583874048
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002626980942828486,
      "loss": 3.2509,
      "theoretical_loss": 3.4993271580317344,
      "tokens_seen": 1583939584
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026268806419257776,
      "loss": 2.8668,
      "theoretical_loss": 3.499314525635456,
      "tokens_seen": 1584005120
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026267803410230694,
      "loss": 2.9344,
      "theoretical_loss": 3.499301893908149,
      "tokens_seen": 1584070656
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002626680040120361,
      "loss": 3.046,
      "theoretical_loss": 3.49928926284975,
      "tokens_seen": 1584136192
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002626579739217653,
      "loss": 2.7266,
      "theoretical_loss": 3.4992766324601967,
      "tokens_seen": 1584201728
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002626479438314945,
      "loss": 2.8555,
      "theoretical_loss": 3.4992640027394253,
      "tokens_seen": 1584267264
    },
    {
      "epoch": 0.48,
      "objective/train/advantage_avg": 0.4938502907752991,
      "objective/train/docs_used": 897735,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.7925333976745605,
      "objective/train/original_loss": 2.7925333976745605,
      "objective/train/theoretical_loss": 3.499251373687373,
      "objective/train/tokens_used": 1604792800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2439507395029068,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506250858306885,
      "objective/train/weighted_lm_loss": 2.9339005947113037,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0417574644088745,
      "theoretical_loss": 3.499251373687373,
      "tokens_seen": 1584332800
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002626379137412237,
      "loss": 2.8552,
      "theoretical_loss": 3.499251373687373,
      "tokens_seen": 1584332800
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026262788365095284,
      "loss": 3.0731,
      "theoretical_loss": 3.499238745303977,
      "tokens_seen": 1584398336
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002626178535606821,
      "loss": 2.8742,
      "theoretical_loss": 3.4992261175891732,
      "tokens_seen": 1584463872
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026260782347041126,
      "loss": 2.8219,
      "theoretical_loss": 3.4992134905428998,
      "tokens_seen": 1584529408
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026259779338014044,
      "loss": 2.8506,
      "theoretical_loss": 3.4992008641650933,
      "tokens_seen": 1584594944
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002625877632898696,
      "loss": 2.928,
      "theoretical_loss": 3.49918823845569,
      "tokens_seen": 1584660480
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002625777331995988,
      "loss": 2.8264,
      "theoretical_loss": 3.4991756134146286,
      "tokens_seen": 1584726016
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.000262567703109328,
      "loss": 3.114,
      "theoretical_loss": 3.499162989041844,
      "tokens_seen": 1584791552
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002625576730190572,
      "loss": 2.7806,
      "theoretical_loss": 3.4991503653372744,
      "tokens_seen": 1584857088
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026254764292878635,
      "loss": 2.8279,
      "theoretical_loss": 3.4991377423008565,
      "tokens_seen": 1584922624
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002625376128385156,
      "loss": 2.8043,
      "theoretical_loss": 3.4991251199325277,
      "tokens_seen": 1584988160
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002625275827482447,
      "loss": 2.8724,
      "theoretical_loss": 3.4991124982322246,
      "tokens_seen": 1585053696
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026251755265797394,
      "loss": 3.0771,
      "theoretical_loss": 3.4990998771998845,
      "tokens_seen": 1585119232
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002625075225677031,
      "loss": 2.9268,
      "theoretical_loss": 3.499087256835444,
      "tokens_seen": 1585184768
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002624974924774323,
      "loss": 2.8485,
      "theoretical_loss": 3.49907463713884,
      "tokens_seen": 1585250304
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002624874623871615,
      "loss": 2.8124,
      "theoretical_loss": 3.4990620181100107,
      "tokens_seen": 1585315840
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026247743229689067,
      "loss": 3.0308,
      "theoretical_loss": 3.499049399748892,
      "tokens_seen": 1585381376
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026246740220661985,
      "loss": 3.0367,
      "theoretical_loss": 3.4990367820554216,
      "tokens_seen": 1585446912
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002624573721163491,
      "loss": 2.5958,
      "theoretical_loss": 3.499024165029536,
      "tokens_seen": 1585512448
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002624473420260782,
      "loss": 2.913,
      "theoretical_loss": 3.499011548671173,
      "tokens_seen": 1585577984
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026243731193580745,
      "loss": 2.809,
      "theoretical_loss": 3.4989989329802693,
      "tokens_seen": 1585643520
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026242728184553663,
      "loss": 2.7163,
      "theoretical_loss": 3.498986317956762,
      "tokens_seen": 1585709056
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002624172517552658,
      "loss": 2.8926,
      "theoretical_loss": 3.4989737036005883,
      "tokens_seen": 1585774592
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.000262407221664995,
      "loss": 3.0791,
      "theoretical_loss": 3.4989610899116856,
      "tokens_seen": 1585840128
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026239719157472417,
      "loss": 2.8887,
      "theoretical_loss": 3.49894847688999,
      "tokens_seen": 1585905664
    },
    {
      "epoch": 0.48,
      "objective/train/advantage_avg": 0.49525678157806396,
      "objective/train/docs_used": 899130,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.373932361602783,
      "objective/train/original_loss": 2.373932361602783,
      "objective/train/theoretical_loss": 3.49893586453544,
      "objective/train/tokens_used": 1606431200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24532446265220642,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507729053497314,
      "objective/train/weighted_lm_loss": 2.494605541229248,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0253151655197144,
      "theoretical_loss": 3.49893586453544,
      "tokens_seen": 1585971200
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026238716148445335,
      "loss": 2.5782,
      "theoretical_loss": 3.49893586453544,
      "tokens_seen": 1585971200
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002623771313941826,
      "loss": 2.8381,
      "theoretical_loss": 3.498923252847972,
      "tokens_seen": 1586036736
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002623671013039117,
      "loss": 2.8429,
      "theoretical_loss": 3.498910641827523,
      "tokens_seen": 1586102272
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026235707121364095,
      "loss": 2.5936,
      "theoretical_loss": 3.4988980314740306,
      "tokens_seen": 1586167808
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002623470411233701,
      "loss": 3.0366,
      "theoretical_loss": 3.4988854217874317,
      "tokens_seen": 1586233344
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002623370110330993,
      "loss": 2.9121,
      "theoretical_loss": 3.4988728127676634,
      "tokens_seen": 1586298880
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002623269809428285,
      "loss": 3.1108,
      "theoretical_loss": 3.4988602044146635,
      "tokens_seen": 1586364416
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002623169508525577,
      "loss": 2.6803,
      "theoretical_loss": 3.4988475967283685,
      "tokens_seen": 1586429952
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026230692076228686,
      "loss": 2.6899,
      "theoretical_loss": 3.498834989708716,
      "tokens_seen": 1586495488
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002622968906720161,
      "loss": 2.7859,
      "theoretical_loss": 3.498822383355643,
      "tokens_seen": 1586561024
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002622868605817452,
      "loss": 2.6119,
      "theoretical_loss": 3.498809777669087,
      "tokens_seen": 1586626560
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026227683049147445,
      "loss": 2.7841,
      "theoretical_loss": 3.4987971726489846,
      "tokens_seen": 1586692096
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002622668004012036,
      "loss": 2.8826,
      "theoretical_loss": 3.4987845682952736,
      "tokens_seen": 1586757632
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002622567703109328,
      "loss": 2.7503,
      "theoretical_loss": 3.4987719646078914,
      "tokens_seen": 1586823168
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.000262246740220662,
      "loss": 2.8562,
      "theoretical_loss": 3.4987593615867745,
      "tokens_seen": 1586888704
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002622367101303912,
      "loss": 3.0237,
      "theoretical_loss": 3.498746759231861,
      "tokens_seen": 1586954240
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026222668004012036,
      "loss": 2.944,
      "theoretical_loss": 3.498734157543088,
      "tokens_seen": 1587019776
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026221664994984954,
      "loss": 2.9247,
      "theoretical_loss": 3.498721556520392,
      "tokens_seen": 1587085312
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002622066198595787,
      "loss": 2.6606,
      "theoretical_loss": 3.4987089561637115,
      "tokens_seen": 1587150848
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026219658976930796,
      "loss": 2.6882,
      "theoretical_loss": 3.4986963564729834,
      "tokens_seen": 1587216384
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002621865596790371,
      "loss": 2.7408,
      "theoretical_loss": 3.4986837574481444,
      "tokens_seen": 1587281920
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002621765295887663,
      "loss": 2.7941,
      "theoretical_loss": 3.498671159089132,
      "tokens_seen": 1587347456
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026216649949849545,
      "loss": 2.6281,
      "theoretical_loss": 3.4986585613958843,
      "tokens_seen": 1587412992
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002621564694082247,
      "loss": 2.8131,
      "theoretical_loss": 3.498645964368338,
      "tokens_seen": 1587478528
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026214643931795386,
      "loss": 2.686,
      "theoretical_loss": 3.4986333680064297,
      "tokens_seen": 1587544064
    },
    {
      "epoch": 0.48,
      "objective/train/advantage_avg": 0.49308791756629944,
      "objective/train/docs_used": 899656,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.2280139923095703,
      "objective/train/original_loss": 3.2280142307281494,
      "objective/train/theoretical_loss": 3.4986207723100984,
      "objective/train/tokens_used": 1608069600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2432619035243988,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505452156066895,
      "objective/train/weighted_lm_loss": 3.3910324573516846,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0435127019882202,
      "theoretical_loss": 3.4986207723100984,
      "tokens_seen": 1587609600
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026213640922768304,
      "loss": 3.0032,
      "theoretical_loss": 3.4986207723100984,
      "tokens_seen": 1587609600
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002621263791374122,
      "loss": 2.6491,
      "theoretical_loss": 3.49860817727928,
      "tokens_seen": 1587675136
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026211634904714146,
      "loss": 2.6079,
      "theoretical_loss": 3.4985955829139135,
      "tokens_seen": 1587740672
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002621063189568706,
      "loss": 2.9165,
      "theoretical_loss": 3.4985829892139346,
      "tokens_seen": 1587806208
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002620962888665998,
      "loss": 2.8654,
      "theoretical_loss": 3.4985703961792813,
      "tokens_seen": 1587871744
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026208625877632895,
      "loss": 2.7604,
      "theoretical_loss": 3.4985578038098915,
      "tokens_seen": 1587937280
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002620762286860582,
      "loss": 2.7887,
      "theoretical_loss": 3.498545212105702,
      "tokens_seen": 1588002816
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026206619859578737,
      "loss": 2.8189,
      "theoretical_loss": 3.4985326210666496,
      "tokens_seen": 1588068352
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026205616850551655,
      "loss": 2.5508,
      "theoretical_loss": 3.4985200306926734,
      "tokens_seen": 1588133888
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026204613841524573,
      "loss": 3.0641,
      "theoretical_loss": 3.4985074409837096,
      "tokens_seen": 1588199424
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002620361083249749,
      "loss": 2.8618,
      "theoretical_loss": 3.4984948519396957,
      "tokens_seen": 1588264960
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002620260782347041,
      "loss": 2.7621,
      "theoretical_loss": 3.49848226356057,
      "tokens_seen": 1588330496
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002620160481444333,
      "loss": 2.8943,
      "theoretical_loss": 3.498469675846269,
      "tokens_seen": 1588396032
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002620060180541625,
      "loss": 2.9809,
      "theoretical_loss": 3.49845708879673,
      "tokens_seen": 1588461568
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002619959879638917,
      "loss": 2.9081,
      "theoretical_loss": 3.4984445024118918,
      "tokens_seen": 1588527104
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026198595787362087,
      "loss": 2.9986,
      "theoretical_loss": 3.4984319166916906,
      "tokens_seen": 1588592640
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026197592778335005,
      "loss": 2.8064,
      "theoretical_loss": 3.4984193316360646,
      "tokens_seen": 1588658176
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002619658976930793,
      "loss": 3.0521,
      "theoretical_loss": 3.4984067472449514,
      "tokens_seen": 1588723712
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002619558676028084,
      "loss": 2.8654,
      "theoretical_loss": 3.4983941635182876,
      "tokens_seen": 1588789248
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026194583751253765,
      "loss": 2.7086,
      "theoretical_loss": 3.498381580456011,
      "tokens_seen": 1588854784
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026193580742226683,
      "loss": 2.7293,
      "theoretical_loss": 3.49836899805806,
      "tokens_seen": 1588920320
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.000261925777331996,
      "loss": 2.8069,
      "theoretical_loss": 3.4983564163243708,
      "tokens_seen": 1588985856
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002619157472417252,
      "loss": 2.9557,
      "theoretical_loss": 3.4983438352548824,
      "tokens_seen": 1589051392
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026190571715145437,
      "loss": 2.9165,
      "theoretical_loss": 3.498331254849531,
      "tokens_seen": 1589116928
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026189568706118355,
      "loss": 2.9598,
      "theoretical_loss": 3.498318675108255,
      "tokens_seen": 1589182464
    },
    {
      "epoch": 0.48,
      "objective/train/advantage_avg": 0.4931066036224365,
      "objective/train/docs_used": 900745,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.6543920040130615,
      "objective/train/original_loss": 2.6543920040130615,
      "objective/train/theoretical_loss": 3.4983060960309915,
      "objective/train/tokens_used": 1609708000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2432296872138977,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505470037460327,
      "objective/train/weighted_lm_loss": 2.7887027263641357,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 3.4983060960309915,
      "tokens_seen": 1589248000
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002618856569709128,
      "loss": 2.9863,
      "theoretical_loss": 3.4983060960309915,
      "tokens_seen": 1589248000
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002618756268806419,
      "loss": 2.8879,
      "theoretical_loss": 3.4982935176176784,
      "tokens_seen": 1589313536
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026186559679037115,
      "loss": 2.7929,
      "theoretical_loss": 3.4982809398682533,
      "tokens_seen": 1589379072
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002618555667001003,
      "loss": 2.8229,
      "theoretical_loss": 3.498268362782653,
      "tokens_seen": 1589444608
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002618455366098295,
      "loss": 2.924,
      "theoretical_loss": 3.498255786360817,
      "tokens_seen": 1589510144
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002618355065195587,
      "loss": 2.9092,
      "theoretical_loss": 3.4982432106026806,
      "tokens_seen": 1589575680
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002618254764292879,
      "loss": 2.8962,
      "theoretical_loss": 3.4982306355081825,
      "tokens_seen": 1589641216
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026181544633901706,
      "loss": 2.7235,
      "theoretical_loss": 3.49821806107726,
      "tokens_seen": 1589706752
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002618054162487463,
      "loss": 2.7123,
      "theoretical_loss": 3.4982054873098516,
      "tokens_seen": 1589772288
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002617953861584754,
      "loss": 2.7616,
      "theoretical_loss": 3.4981929142058945,
      "tokens_seen": 1589837824
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026178535606820465,
      "loss": 2.6367,
      "theoretical_loss": 3.498180341765326,
      "tokens_seen": 1589903360
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002617753259779338,
      "loss": 2.7467,
      "theoretical_loss": 3.4981677699880835,
      "tokens_seen": 1589968896
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.000261765295887663,
      "loss": 2.5749,
      "theoretical_loss": 3.4981551988741053,
      "tokens_seen": 1590034432
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002617552657973922,
      "loss": 2.8265,
      "theoretical_loss": 3.498142628423329,
      "tokens_seen": 1590099968
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002617452357071214,
      "loss": 2.7973,
      "theoretical_loss": 3.4981300586356925,
      "tokens_seen": 1590165504
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026173520561685056,
      "loss": 2.7943,
      "theoretical_loss": 3.4981174895111327,
      "tokens_seen": 1590231040
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026172517552657974,
      "loss": 2.6112,
      "theoretical_loss": 3.498104921049588,
      "tokens_seen": 1590296576
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002617151454363089,
      "loss": 2.8072,
      "theoretical_loss": 3.4980923532509953,
      "tokens_seen": 1590362112
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026170511534603816,
      "loss": 2.86,
      "theoretical_loss": 3.4980797861152935,
      "tokens_seen": 1590427648
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002616950852557673,
      "loss": 2.7403,
      "theoretical_loss": 3.4980672196424196,
      "tokens_seen": 1590493184
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002616850551654965,
      "loss": 2.9893,
      "theoretical_loss": 3.4980546538323107,
      "tokens_seen": 1590558720
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026167502507522565,
      "loss": 2.655,
      "theoretical_loss": 3.498042088684906,
      "tokens_seen": 1590624256
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002616649949849549,
      "loss": 2.7751,
      "theoretical_loss": 3.4980295242001422,
      "tokens_seen": 1590689792
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026165496489468406,
      "loss": 2.8454,
      "theoretical_loss": 3.4980169603779574,
      "tokens_seen": 1590755328
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026164493480441324,
      "loss": 2.7883,
      "theoretical_loss": 3.4980043972182897,
      "tokens_seen": 1590820864
    },
    {
      "epoch": 0.48,
      "objective/train/advantage_avg": 0.4922775328159332,
      "objective/train/docs_used": 902130,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.2942659854888916,
      "objective/train/original_loss": 2.2942659854888916,
      "objective/train/theoretical_loss": 3.497991834721076,
      "objective/train/tokens_used": 1611346400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24245460331439972,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0504602193832397,
      "objective/train/weighted_lm_loss": 2.410090208053589,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0460278987884521,
      "theoretical_loss": 3.497991834721076,
      "tokens_seen": 1590886400
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002616349047141424,
      "loss": 2.7551,
      "theoretical_loss": 3.497991834721076,
      "tokens_seen": 1590886400
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026162487462387166,
      "loss": 2.9071,
      "theoretical_loss": 3.497979272886255,
      "tokens_seen": 1590951936
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002616148445336008,
      "loss": 2.7781,
      "theoretical_loss": 3.4979667117137634,
      "tokens_seen": 1591017472
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026160481444333,
      "loss": 2.8072,
      "theoretical_loss": 3.4979541512035404,
      "tokens_seen": 1591083008
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026159478435305915,
      "loss": 2.9097,
      "theoretical_loss": 3.4979415913555227,
      "tokens_seen": 1591148544
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002615847542627884,
      "loss": 2.7248,
      "theoretical_loss": 3.4979290321696483,
      "tokens_seen": 1591214080
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026157472417251757,
      "loss": 2.6751,
      "theoretical_loss": 3.4979164736458555,
      "tokens_seen": 1591279616
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026156469408224675,
      "loss": 3.1109,
      "theoretical_loss": 3.497903915784082,
      "tokens_seen": 1591345152
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026155466399197593,
      "loss": 2.9627,
      "theoretical_loss": 3.497891358584265,
      "tokens_seen": 1591410688
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002615446339017051,
      "loss": 2.6227,
      "theoretical_loss": 3.497878802046343,
      "tokens_seen": 1591476224
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002615346038114343,
      "loss": 2.7613,
      "theoretical_loss": 3.497866246170254,
      "tokens_seen": 1591541760
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002615245737211635,
      "loss": 2.9137,
      "theoretical_loss": 3.4978536909559352,
      "tokens_seen": 1591607296
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026151454363089265,
      "loss": 2.7275,
      "theoretical_loss": 3.497841136403325,
      "tokens_seen": 1591672832
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002615045135406219,
      "loss": 2.8894,
      "theoretical_loss": 3.497828582512361,
      "tokens_seen": 1591738368
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.000261494483450351,
      "loss": 2.8653,
      "theoretical_loss": 3.4978160292829816,
      "tokens_seen": 1591803904
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026148445336008025,
      "loss": 2.8417,
      "theoretical_loss": 3.4978034767151236,
      "tokens_seen": 1591869440
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026147442326980943,
      "loss": 2.6117,
      "theoretical_loss": 3.4977909248087258,
      "tokens_seen": 1591934976
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002614643931795386,
      "loss": 2.954,
      "theoretical_loss": 3.497778373563726,
      "tokens_seen": 1592000512
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002614543630892678,
      "loss": 2.7751,
      "theoretical_loss": 3.497765822980062,
      "tokens_seen": 1592066048
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026144433299899703,
      "loss": 2.7404,
      "theoretical_loss": 3.497753273057672,
      "tokens_seen": 1592131584
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026143430290872616,
      "loss": 2.7089,
      "theoretical_loss": 3.497740723796494,
      "tokens_seen": 1592197120
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002614242728184554,
      "loss": 2.8612,
      "theoretical_loss": 3.4977281751964653,
      "tokens_seen": 1592262656
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002614142427281845,
      "loss": 2.8265,
      "theoretical_loss": 3.4977156272575245,
      "tokens_seen": 1592328192
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026140421263791375,
      "loss": 3.0212,
      "theoretical_loss": 3.4977030799796087,
      "tokens_seen": 1592393728
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026139418254764293,
      "loss": 2.859,
      "theoretical_loss": 3.4976905333626567,
      "tokens_seen": 1592459264
    },
    {
      "epoch": 0.48,
      "objective/train/advantage_avg": 0.4940631687641144,
      "objective/train/docs_used": 902721,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.5459694862365723,
      "objective/train/original_loss": 2.5459694862365723,
      "objective/train/theoretical_loss": 3.4976779874066066,
      "objective/train/tokens_used": 1612984800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24420106410980225,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506476163864136,
      "objective/train/weighted_lm_loss": 2.6745710372924805,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0338950157165527,
      "theoretical_loss": 3.4976779874066066,
      "tokens_seen": 1592524800
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002613841524573721,
      "loss": 2.7406,
      "theoretical_loss": 3.4976779874066066,
      "tokens_seen": 1592524800
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002613741223671013,
      "loss": 2.8076,
      "theoretical_loss": 3.4976654421113955,
      "tokens_seen": 1592590336
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002613640922768305,
      "loss": 2.8347,
      "theoretical_loss": 3.497652897476962,
      "tokens_seen": 1592655872
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026135406218655966,
      "loss": 2.6939,
      "theoretical_loss": 3.497640353503244,
      "tokens_seen": 1592721408
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002613440320962889,
      "loss": 2.6544,
      "theoretical_loss": 3.4976278101901803,
      "tokens_seen": 1592786944
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.000261334002006018,
      "loss": 2.7638,
      "theoretical_loss": 3.4976152675377077,
      "tokens_seen": 1592852480
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026132397191574726,
      "loss": 2.8431,
      "theoretical_loss": 3.4976027255457645,
      "tokens_seen": 1592918016
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002613139418254764,
      "loss": 2.9303,
      "theoretical_loss": 3.4975901842142894,
      "tokens_seen": 1592983552
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002613039117352056,
      "loss": 2.7597,
      "theoretical_loss": 3.4975776435432198,
      "tokens_seen": 1593049088
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002612938816449348,
      "loss": 2.6905,
      "theoretical_loss": 3.497565103532494,
      "tokens_seen": 1593114624
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.000261283851554664,
      "loss": 2.7007,
      "theoretical_loss": 3.49755256418205,
      "tokens_seen": 1593180160
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026127382146439316,
      "loss": 2.8661,
      "theoretical_loss": 3.497540025491826,
      "tokens_seen": 1593245696
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002612637913741224,
      "loss": 3.0636,
      "theoretical_loss": 3.4975274874617597,
      "tokens_seen": 1593311232
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002612537612838516,
      "loss": 2.7335,
      "theoretical_loss": 3.49751495009179,
      "tokens_seen": 1593376768
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026124373119358076,
      "loss": 2.6553,
      "theoretical_loss": 3.497502413381854,
      "tokens_seen": 1593442304
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026123370110330994,
      "loss": 2.7599,
      "theoretical_loss": 3.4974898773318905,
      "tokens_seen": 1593507840
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002612236710130391,
      "loss": 2.9597,
      "theoretical_loss": 3.497477341941837,
      "tokens_seen": 1593573376
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026121364092276836,
      "loss": 2.9085,
      "theoretical_loss": 3.4974648072116326,
      "tokens_seen": 1593638912
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002612036108324975,
      "loss": 2.9158,
      "theoretical_loss": 3.4974522731412145,
      "tokens_seen": 1593704448
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002611935807422267,
      "loss": 2.5252,
      "theoretical_loss": 3.4974397397305217,
      "tokens_seen": 1593769984
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026118355065195585,
      "loss": 2.8673,
      "theoretical_loss": 3.4974272069794914,
      "tokens_seen": 1593835520
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002611735205616851,
      "loss": 2.9015,
      "theoretical_loss": 3.4974146748880623,
      "tokens_seen": 1593901056
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026116349047141426,
      "loss": 2.9447,
      "theoretical_loss": 3.4974021434561724,
      "tokens_seen": 1593966592
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026115346038114344,
      "loss": 2.9126,
      "theoretical_loss": 3.4973896126837603,
      "tokens_seen": 1594032128
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002611434302908726,
      "loss": 2.7898,
      "theoretical_loss": 3.4973770825707637,
      "tokens_seen": 1594097664
    },
    {
      "epoch": 0.48,
      "objective/train/advantage_avg": 0.4945322573184967,
      "objective/train/docs_used": 903500,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.7500357627868652,
      "objective/train/original_loss": 2.7500360012054443,
      "objective/train/theoretical_loss": 3.4973645531171207,
      "objective/train/tokens_used": 1614623200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2446369081735611,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506967306137085,
      "objective/train/weighted_lm_loss": 2.889171600341797,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0417574644088745,
      "theoretical_loss": 3.4973645531171207,
      "tokens_seen": 1594163200
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026113340020060186,
      "loss": 2.6693,
      "theoretical_loss": 3.4973645531171207,
      "tokens_seen": 1594163200
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.000261123370110331,
      "loss": 2.7963,
      "theoretical_loss": 3.49735202432277,
      "tokens_seen": 1594228736
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002611133400200602,
      "loss": 2.8347,
      "theoretical_loss": 3.4973394961876494,
      "tokens_seen": 1594294272
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026110330992978935,
      "loss": 2.7185,
      "theoretical_loss": 3.4973269687116977,
      "tokens_seen": 1594359808
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002610932798395186,
      "loss": 2.9633,
      "theoretical_loss": 3.497314441894852,
      "tokens_seen": 1594425344
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026108324974924777,
      "loss": 3.0225,
      "theoretical_loss": 3.4973019157370517,
      "tokens_seen": 1594490880
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026107321965897695,
      "loss": 3.0439,
      "theoretical_loss": 3.497289390238234,
      "tokens_seen": 1594556416
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026106318956870613,
      "loss": 2.9219,
      "theoretical_loss": 3.497276865398338,
      "tokens_seen": 1594621952
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002610531594784353,
      "loss": 2.7769,
      "theoretical_loss": 3.497264341217302,
      "tokens_seen": 1594687488
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002610431293881645,
      "loss": 2.8566,
      "theoretical_loss": 3.497251817695064,
      "tokens_seen": 1594753024
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002610330992978937,
      "loss": 2.7915,
      "theoretical_loss": 3.4972392948315614,
      "tokens_seen": 1594818560
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026102306920762285,
      "loss": 2.8972,
      "theoretical_loss": 3.4972267726267336,
      "tokens_seen": 1594884096
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002610130391173521,
      "loss": 2.7809,
      "theoretical_loss": 3.497214251080519,
      "tokens_seen": 1594949632
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002610030090270812,
      "loss": 2.8982,
      "theoretical_loss": 3.497201730192855,
      "tokens_seen": 1595015168
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026099297893681045,
      "loss": 2.939,
      "theoretical_loss": 3.4971892099636808,
      "tokens_seen": 1595080704
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026098294884653963,
      "loss": 2.7545,
      "theoretical_loss": 3.4971766903929344,
      "tokens_seen": 1595146240
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002609729187562688,
      "loss": 2.7462,
      "theoretical_loss": 3.497164171480554,
      "tokens_seen": 1595211776
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.000260962888665998,
      "loss": 2.813,
      "theoretical_loss": 3.4971516532264775,
      "tokens_seen": 1595277312
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026095285857572723,
      "loss": 2.8007,
      "theoretical_loss": 3.4971391356306443,
      "tokens_seen": 1595342848
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026094282848545636,
      "loss": 2.7514,
      "theoretical_loss": 3.4971266186929917,
      "tokens_seen": 1595408384
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002609327983951856,
      "loss": 2.6949,
      "theoretical_loss": 3.497114102413458,
      "tokens_seen": 1595473920
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002609227683049147,
      "loss": 2.7787,
      "theoretical_loss": 3.497101586791983,
      "tokens_seen": 1595539456
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026091273821464395,
      "loss": 2.7732,
      "theoretical_loss": 3.497089071828504,
      "tokens_seen": 1595604992
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026090270812437313,
      "loss": 3.1412,
      "theoretical_loss": 3.497076557522959,
      "tokens_seen": 1595670528
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002608926780341023,
      "loss": 2.8881,
      "theoretical_loss": 3.4970640438752874,
      "tokens_seen": 1595736064
    },
    {
      "epoch": 0.48,
      "objective/train/advantage_avg": 0.49458837509155273,
      "objective/train/docs_used": 904004,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.0312812328338623,
      "objective/train/original_loss": 2.031280755996704,
      "objective/train/theoretical_loss": 3.497051530885427,
      "objective/train/tokens_used": 1616261600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24465246498584747,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050702452659607,
      "objective/train/weighted_lm_loss": 2.1342062950134277,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0465034246444702,
      "theoretical_loss": 3.497051530885427,
      "tokens_seen": 1595801600
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002608826479438315,
      "loss": 2.6745,
      "theoretical_loss": 3.497051530885427,
      "tokens_seen": 1595801600
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002608726178535607,
      "loss": 2.6301,
      "theoretical_loss": 3.4970390185533167,
      "tokens_seen": 1595867136
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026086258776328986,
      "loss": 2.899,
      "theoretical_loss": 3.4970265068788944,
      "tokens_seen": 1595932672
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002608525576730191,
      "loss": 2.6128,
      "theoretical_loss": 3.4970139958620985,
      "tokens_seen": 1595998208
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002608425275827482,
      "loss": 2.7079,
      "theoretical_loss": 3.4970014855028673,
      "tokens_seen": 1596063744
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026083249749247746,
      "loss": 2.9338,
      "theoretical_loss": 3.49698897580114,
      "tokens_seen": 1596129280
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002608224674022066,
      "loss": 2.6492,
      "theoretical_loss": 3.4969764667568546,
      "tokens_seen": 1596194816
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002608124373119358,
      "loss": 2.9372,
      "theoretical_loss": 3.496963958369949,
      "tokens_seen": 1596260352
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.000260802407221665,
      "loss": 2.6915,
      "theoretical_loss": 3.496951450640363,
      "tokens_seen": 1596325888
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002607923771313942,
      "loss": 2.9516,
      "theoretical_loss": 3.4969389435680336,
      "tokens_seen": 1596391424
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026078234704112336,
      "loss": 2.9112,
      "theoretical_loss": 3.4969264371529007,
      "tokens_seen": 1596456960
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002607723169508526,
      "loss": 2.5519,
      "theoretical_loss": 3.496913931394902,
      "tokens_seen": 1596522496
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002607622868605817,
      "loss": 3.0034,
      "theoretical_loss": 3.4969014262939755,
      "tokens_seen": 1596588032
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026075225677031096,
      "loss": 2.8128,
      "theoretical_loss": 3.4968889218500605,
      "tokens_seen": 1596653568
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002607422266800401,
      "loss": 2.8084,
      "theoretical_loss": 3.496876418063095,
      "tokens_seen": 1596719104
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002607321965897693,
      "loss": 2.7923,
      "theoretical_loss": 3.4968639149330185,
      "tokens_seen": 1596784640
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002607221664994985,
      "loss": 2.8857,
      "theoretical_loss": 3.496851412459768,
      "tokens_seen": 1596850176
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002607121364092277,
      "loss": 2.982,
      "theoretical_loss": 3.496838910643284,
      "tokens_seen": 1596915712
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026070210631895687,
      "loss": 2.7313,
      "theoretical_loss": 3.4968264094835027,
      "tokens_seen": 1596981248
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026069207622868605,
      "loss": 2.7545,
      "theoretical_loss": 3.4968139089803647,
      "tokens_seen": 1597046784
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026068204613841523,
      "loss": 2.8158,
      "theoretical_loss": 3.4968014091338073,
      "tokens_seen": 1597112320
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026067201604814446,
      "loss": 2.7897,
      "theoretical_loss": 3.49678890994377,
      "tokens_seen": 1597177856
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002606619859578736,
      "loss": 2.9471,
      "theoretical_loss": 3.496776411410191,
      "tokens_seen": 1597243392
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002606519558676028,
      "loss": 2.4987,
      "theoretical_loss": 3.496763913533008,
      "tokens_seen": 1597308928
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026064192577733195,
      "loss": 2.7817,
      "theoretical_loss": 3.4967514163121614,
      "tokens_seen": 1597374464
    },
    {
      "epoch": 0.48,
      "objective/train/advantage_avg": 0.49230799078941345,
      "objective/train/docs_used": 904558,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.8820090293884277,
      "objective/train/original_loss": 2.8820090293884277,
      "objective/train/theoretical_loss": 3.496738919747588,
      "objective/train/tokens_used": 1617900000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.242469921708107,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0504631996154785,
      "objective/train/weighted_lm_loss": 3.0272767543792725,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0363596677780151,
      "theoretical_loss": 3.496738919747588,
      "tokens_seen": 1597440000
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002606318956870612,
      "loss": 3.1497,
      "theoretical_loss": 3.496738919747588,
      "tokens_seen": 1597440000
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026062186559679037,
      "loss": 2.8452,
      "theoretical_loss": 3.4967264238392275,
      "tokens_seen": 1597505536
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026061183550651955,
      "loss": 2.9866,
      "theoretical_loss": 3.4967139285870186,
      "tokens_seen": 1597571072
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026060180541624873,
      "loss": 2.8625,
      "theoretical_loss": 3.496701433990899,
      "tokens_seen": 1597636608
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026059177532597797,
      "loss": 2.8606,
      "theoretical_loss": 3.496688940050808,
      "tokens_seen": 1597702144
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002605817452357071,
      "loss": 3.0972,
      "theoretical_loss": 3.4966764467666844,
      "tokens_seen": 1597767680
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026057171514543633,
      "loss": 2.731,
      "theoretical_loss": 3.4966639541384668,
      "tokens_seen": 1597833216
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026056168505516546,
      "loss": 2.8384,
      "theoretical_loss": 3.4966514621660933,
      "tokens_seen": 1597898752
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002605516549648947,
      "loss": 3.0062,
      "theoretical_loss": 3.496638970849503,
      "tokens_seen": 1597964288
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026054162487462387,
      "loss": 2.8692,
      "theoretical_loss": 3.4966264801886346,
      "tokens_seen": 1598029824
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026053159478435305,
      "loss": 3.027,
      "theoretical_loss": 3.496613990183427,
      "tokens_seen": 1598095360
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026052156469408223,
      "loss": 2.975,
      "theoretical_loss": 3.4966015008338185,
      "tokens_seen": 1598160896
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002605115346038114,
      "loss": 2.8106,
      "theoretical_loss": 3.4965890121397476,
      "tokens_seen": 1598226432
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026050150451354065,
      "loss": 2.7236,
      "theoretical_loss": 3.4965765241011537,
      "tokens_seen": 1598291968
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026049147442326983,
      "loss": 3.1128,
      "theoretical_loss": 3.496564036717975,
      "tokens_seen": 1598357504
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.000260481444332999,
      "loss": 2.6542,
      "theoretical_loss": 3.4965515499901505,
      "tokens_seen": 1598423040
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002604714142427282,
      "loss": 2.8895,
      "theoretical_loss": 3.4965390639176186,
      "tokens_seen": 1598488576
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026046138415245743,
      "loss": 2.8421,
      "theoretical_loss": 3.4965265785003186,
      "tokens_seen": 1598554112
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026045135406218656,
      "loss": 2.6912,
      "theoretical_loss": 3.4965140937381887,
      "tokens_seen": 1598619648
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002604413239719158,
      "loss": 2.9671,
      "theoretical_loss": 3.496501609631168,
      "tokens_seen": 1598685184
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002604312938816449,
      "loss": 2.9016,
      "theoretical_loss": 3.4964891261791955,
      "tokens_seen": 1598750720
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026042126379137415,
      "loss": 2.8195,
      "theoretical_loss": 3.496476643382209,
      "tokens_seen": 1598816256
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026041123370110333,
      "loss": 2.8758,
      "theoretical_loss": 3.4964641612401484,
      "tokens_seen": 1598881792
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002604012036108325,
      "loss": 2.5937,
      "theoretical_loss": 3.4964516797529517,
      "tokens_seen": 1598947328
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002603911735205617,
      "loss": 2.9199,
      "theoretical_loss": 3.496439198920558,
      "tokens_seen": 1599012864
    },
    {
      "epoch": 0.48,
      "objective/train/advantage_avg": 0.49550822377204895,
      "objective/train/docs_used": 905731,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.069395065307617,
      "objective/train/original_loss": 3.069395065307617,
      "objective/train/theoretical_loss": 3.4964267187429066,
      "objective/train/tokens_used": 1619538400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24558760225772858,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507993698120117,
      "objective/train/weighted_lm_loss": 3.2253777980804443,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 3.4964267187429066,
      "tokens_seen": 1599078400
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002603811434302909,
      "loss": 2.9065,
      "theoretical_loss": 3.4964267187429066,
      "tokens_seen": 1599078400
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026037111334002006,
      "loss": 2.9866,
      "theoretical_loss": 3.4964142392199355,
      "tokens_seen": 1599143936
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002603610832497493,
      "loss": 2.7704,
      "theoretical_loss": 3.496401760351584,
      "tokens_seen": 1599209472
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002603510531594784,
      "loss": 2.612,
      "theoretical_loss": 3.496389282137791,
      "tokens_seen": 1599275008
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026034102306920766,
      "loss": 2.9865,
      "theoretical_loss": 3.496376804578495,
      "tokens_seen": 1599340544
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002603309929789368,
      "loss": 2.5102,
      "theoretical_loss": 3.4963643276736356,
      "tokens_seen": 1599406080
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.000260320962888666,
      "loss": 2.9907,
      "theoretical_loss": 3.4963518514231504,
      "tokens_seen": 1599471616
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002603109327983952,
      "loss": 2.7932,
      "theoretical_loss": 3.496339375826979,
      "tokens_seen": 1599537152
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002603009027081244,
      "loss": 2.6981,
      "theoretical_loss": 3.4963269008850606,
      "tokens_seen": 1599602688
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026029087261785356,
      "loss": 2.8412,
      "theoretical_loss": 3.4963144265973334,
      "tokens_seen": 1599668224
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002602808425275828,
      "loss": 2.7882,
      "theoretical_loss": 3.4963019529637367,
      "tokens_seen": 1599733760
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002602708124373119,
      "loss": 3.0849,
      "theoretical_loss": 3.4962894799842097,
      "tokens_seen": 1599799296
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026026078234704116,
      "loss": 2.8228,
      "theoretical_loss": 3.4962770076586906,
      "tokens_seen": 1599864832
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002602507522567703,
      "loss": 2.6902,
      "theoretical_loss": 3.4962645359871187,
      "tokens_seen": 1599930368
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002602407221664995,
      "loss": 2.9196,
      "theoretical_loss": 3.4962520649694326,
      "tokens_seen": 1599995904
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002602306920762287,
      "loss": 2.9309,
      "theoretical_loss": 3.496239594605572,
      "tokens_seen": 1600061440
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002602206619859579,
      "loss": 2.8406,
      "theoretical_loss": 3.4962271248954755,
      "tokens_seen": 1600126976
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026021063189568707,
      "loss": 2.7638,
      "theoretical_loss": 3.4962146558390814,
      "tokens_seen": 1600192512
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026020060180541625,
      "loss": 2.7991,
      "theoretical_loss": 3.4962021874363294,
      "tokens_seen": 1600258048
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026019057171514543,
      "loss": 2.6862,
      "theoretical_loss": 3.496189719687158,
      "tokens_seen": 1600323584
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.00026018054162487466,
      "loss": 2.8858,
      "theoretical_loss": 3.496177252591507,
      "tokens_seen": 1600389120
    },
    {
      "epoch": 0.48,
      "learning_rate": 0.0002601705115346038,
      "loss": 2.7345,
      "theoretical_loss": 3.4961647861493144,
      "tokens_seen": 1600454656
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.000260160481444333,
      "loss": 2.7678,
      "theoretical_loss": 3.49615232036052,
      "tokens_seen": 1600520192
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00026015045135406215,
      "loss": 2.8694,
      "theoretical_loss": 3.4961398552250618,
      "tokens_seen": 1600585728
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002601404212637914,
      "loss": 2.8809,
      "theoretical_loss": 3.49612739074288,
      "tokens_seen": 1600651264
    },
    {
      "epoch": 0.49,
      "objective/train/advantage_avg": 0.4962941110134125,
      "objective/train/docs_used": 906398,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.751817464828491,
      "objective/train/original_loss": 2.751817226409912,
      "objective/train/theoretical_loss": 3.4961149269139127,
      "objective/train/tokens_used": 1621176800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24633201956748962,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0508817434310913,
      "objective/train/weighted_lm_loss": 2.8918333053588867,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0460278987884521,
      "theoretical_loss": 3.4961149269139127,
      "tokens_seen": 1600716800
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00026013039117352057,
      "loss": 2.7373,
      "theoretical_loss": 3.4961149269139127,
      "tokens_seen": 1600716800
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00026012036108324975,
      "loss": 2.5975,
      "theoretical_loss": 3.496102463738099,
      "tokens_seen": 1600782336
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00026011033099297893,
      "loss": 2.7987,
      "theoretical_loss": 3.4960900012153786,
      "tokens_seen": 1600847872
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00026010030090270817,
      "loss": 2.9153,
      "theoretical_loss": 3.4960775393456904,
      "tokens_seen": 1600913408
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002600902708124373,
      "loss": 2.7754,
      "theoretical_loss": 3.4960650781289724,
      "tokens_seen": 1600978944
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00026008024072216653,
      "loss": 2.6685,
      "theoretical_loss": 3.496052617565165,
      "tokens_seen": 1601044480
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00026007021063189566,
      "loss": 2.9782,
      "theoretical_loss": 3.496040157654207,
      "tokens_seen": 1601110016
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002600601805416249,
      "loss": 2.8395,
      "theoretical_loss": 3.4960276983960368,
      "tokens_seen": 1601175552
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00026005015045135407,
      "loss": 3.1498,
      "theoretical_loss": 3.4960152397905935,
      "tokens_seen": 1601241088
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00026004012036108325,
      "loss": 2.7247,
      "theoretical_loss": 3.4960027818378174,
      "tokens_seen": 1601306624
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00026003009027081243,
      "loss": 2.8808,
      "theoretical_loss": 3.495990324537646,
      "tokens_seen": 1601372160
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002600200601805416,
      "loss": 2.6324,
      "theoretical_loss": 3.4959778678900193,
      "tokens_seen": 1601437696
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002600100300902708,
      "loss": 2.7911,
      "theoretical_loss": 3.4959654118948764,
      "tokens_seen": 1601503232
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00026000000000000003,
      "loss": 2.8007,
      "theoretical_loss": 3.4959529565521565,
      "tokens_seen": 1601568768
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025998996990972916,
      "loss": 2.8542,
      "theoretical_loss": 3.4959405018617984,
      "tokens_seen": 1601634304
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002599799398194584,
      "loss": 2.9984,
      "theoretical_loss": 3.4959280478237416,
      "tokens_seen": 1601699840
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002599699097291876,
      "loss": 2.8471,
      "theoretical_loss": 3.4959155944379248,
      "tokens_seen": 1601765376
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025995987963891676,
      "loss": 3.029,
      "theoretical_loss": 3.4959031417042876,
      "tokens_seen": 1601830912
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025994984954864594,
      "loss": 2.7734,
      "theoretical_loss": 3.495890689622769,
      "tokens_seen": 1601896448
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002599398194583751,
      "loss": 3.0283,
      "theoretical_loss": 3.4958782381933076,
      "tokens_seen": 1601961984
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002599297893681043,
      "loss": 2.9341,
      "theoretical_loss": 3.4958657874158434,
      "tokens_seen": 1602027520
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025991975927783353,
      "loss": 2.9197,
      "theoretical_loss": 3.495853337290315,
      "tokens_seen": 1602093056
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025990972918756266,
      "loss": 2.6788,
      "theoretical_loss": 3.4958408878166622,
      "tokens_seen": 1602158592
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002598996990972919,
      "loss": 2.7129,
      "theoretical_loss": 3.495828438994824,
      "tokens_seen": 1602224128
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.000259889669007021,
      "loss": 2.9041,
      "theoretical_loss": 3.49581599082474,
      "tokens_seen": 1602289664
    },
    {
      "epoch": 0.49,
      "objective/train/advantage_avg": 0.49459558725357056,
      "objective/train/docs_used": 907690,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.8360955715179443,
      "objective/train/original_loss": 2.8360953330993652,
      "objective/train/theoretical_loss": 3.495803543306348,
      "objective/train/tokens_used": 1622815200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2446819692850113,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507032871246338,
      "objective/train/weighted_lm_loss": 2.9796206951141357,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0432155132293701,
      "theoretical_loss": 3.495803543306348,
      "tokens_seen": 1602355200
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025987963891675026,
      "loss": 3.0137,
      "theoretical_loss": 3.495803543306348,
      "tokens_seen": 1602355200
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025986960882647944,
      "loss": 2.821,
      "theoretical_loss": 3.4957910964395884,
      "tokens_seen": 1602420736
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002598595787362086,
      "loss": 2.6639,
      "theoretical_loss": 3.4957786502244,
      "tokens_seen": 1602486272
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002598495486459378,
      "loss": 3.0111,
      "theoretical_loss": 3.4957662046607227,
      "tokens_seen": 1602551808
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.000259839518555667,
      "loss": 2.6993,
      "theoretical_loss": 3.495753759748496,
      "tokens_seen": 1602617344
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025982948846539616,
      "loss": 2.8286,
      "theoretical_loss": 3.495741315487657,
      "tokens_seen": 1602682880
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002598194583751254,
      "loss": 2.6278,
      "theoretical_loss": 3.495728871878147,
      "tokens_seen": 1602748416
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002598094282848545,
      "loss": 2.8917,
      "theoretical_loss": 3.495716428919905,
      "tokens_seen": 1602813952
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025979939819458376,
      "loss": 2.9421,
      "theoretical_loss": 3.4957039866128703,
      "tokens_seen": 1602879488
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025978936810431294,
      "loss": 2.7989,
      "theoretical_loss": 3.4956915449569808,
      "tokens_seen": 1602945024
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002597793380140421,
      "loss": 3.0284,
      "theoretical_loss": 3.495679103952178,
      "tokens_seen": 1603010560
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002597693079237713,
      "loss": 2.8663,
      "theoretical_loss": 3.4956666635983993,
      "tokens_seen": 1603076096
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002597592778335005,
      "loss": 2.6567,
      "theoretical_loss": 3.4956542238955857,
      "tokens_seen": 1603141632
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002597492477432297,
      "loss": 2.8781,
      "theoretical_loss": 3.495641784843675,
      "tokens_seen": 1603207168
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002597392176529589,
      "loss": 2.9767,
      "theoretical_loss": 3.495629346442607,
      "tokens_seen": 1603272704
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002597291875626881,
      "loss": 2.6671,
      "theoretical_loss": 3.495616908692322,
      "tokens_seen": 1603338240
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025971915747241727,
      "loss": 2.7599,
      "theoretical_loss": 3.495604471592758,
      "tokens_seen": 1603403776
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025970912738214645,
      "loss": 2.8439,
      "theoretical_loss": 3.495592035143855,
      "tokens_seen": 1603469312
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025969909729187563,
      "loss": 2.8043,
      "theoretical_loss": 3.495579599345553,
      "tokens_seen": 1603534848
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025968906720160486,
      "loss": 3.0939,
      "theoretical_loss": 3.49556716419779,
      "tokens_seen": 1603600384
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.000259679037111334,
      "loss": 2.7682,
      "theoretical_loss": 3.495554729700506,
      "tokens_seen": 1603665920
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002596690070210632,
      "loss": 2.5959,
      "theoretical_loss": 3.4955422958536406,
      "tokens_seen": 1603731456
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025965897693079235,
      "loss": 2.7687,
      "theoretical_loss": 3.4955298626571336,
      "tokens_seen": 1603796992
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002596489468405216,
      "loss": 2.7751,
      "theoretical_loss": 3.4955174301109233,
      "tokens_seen": 1603862528
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025963891675025077,
      "loss": 3.033,
      "theoretical_loss": 3.49550499821495,
      "tokens_seen": 1603928064
    },
    {
      "epoch": 0.49,
      "objective/train/advantage_avg": 0.4935416579246521,
      "objective/train/docs_used": 908320,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.140263319015503,
      "objective/train/original_loss": 3.140263319015503,
      "objective/train/theoretical_loss": 3.495492566969153,
      "objective/train/tokens_used": 1624453600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2437097430229187,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505928993225098,
      "objective/train/weighted_lm_loss": 3.2991995811462402,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0363596677780151,
      "theoretical_loss": 3.495492566969153,
      "tokens_seen": 1603993600
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025962888665997995,
      "loss": 3.0998,
      "theoretical_loss": 3.495492566969153,
      "tokens_seen": 1603993600
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025961885656970913,
      "loss": 2.7581,
      "theoretical_loss": 3.4954801363734713,
      "tokens_seen": 1604059136
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025960882647943837,
      "loss": 2.9976,
      "theoretical_loss": 3.4954677064278448,
      "tokens_seen": 1604124672
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002595987963891675,
      "loss": 2.7258,
      "theoretical_loss": 3.495455277132213,
      "tokens_seen": 1604190208
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025958876629889673,
      "loss": 2.6877,
      "theoretical_loss": 3.4954428484865145,
      "tokens_seen": 1604255744
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025957873620862586,
      "loss": 2.7493,
      "theoretical_loss": 3.4954304204906896,
      "tokens_seen": 1604321280
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002595687061183551,
      "loss": 2.7506,
      "theoretical_loss": 3.4954179931446774,
      "tokens_seen": 1604386816
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025955867602808427,
      "loss": 2.8022,
      "theoretical_loss": 3.495405566448418,
      "tokens_seen": 1604452352
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025954864593781345,
      "loss": 2.9056,
      "theoretical_loss": 3.49539314040185,
      "tokens_seen": 1604517888
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025953861584754263,
      "loss": 2.7164,
      "theoretical_loss": 3.495380715004914,
      "tokens_seen": 1604583424
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002595285857572718,
      "loss": 2.8613,
      "theoretical_loss": 3.495368290257548,
      "tokens_seen": 1604648960
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.000259518555667001,
      "loss": 2.717,
      "theoretical_loss": 3.4953558661596924,
      "tokens_seen": 1604714496
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025950852557673023,
      "loss": 3.0688,
      "theoretical_loss": 3.4953434427112873,
      "tokens_seen": 1604780032
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025949849548645936,
      "loss": 2.9925,
      "theoretical_loss": 3.495331019912271,
      "tokens_seen": 1604845568
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002594884653961886,
      "loss": 2.804,
      "theoretical_loss": 3.495318597762584,
      "tokens_seen": 1604911104
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002594784353059178,
      "loss": 2.7877,
      "theoretical_loss": 3.4953061762621656,
      "tokens_seen": 1604976640
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025946840521564696,
      "loss": 2.7044,
      "theoretical_loss": 3.495293755410955,
      "tokens_seen": 1605042176
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025945837512537614,
      "loss": 2.8623,
      "theoretical_loss": 3.495281335208892,
      "tokens_seen": 1605107712
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002594483450351053,
      "loss": 2.8306,
      "theoretical_loss": 3.4952689156559162,
      "tokens_seen": 1605173248
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002594383149448345,
      "loss": 2.9949,
      "theoretical_loss": 3.495256496751967,
      "tokens_seen": 1605238784
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025942828485456373,
      "loss": 2.7953,
      "theoretical_loss": 3.4952440784969845,
      "tokens_seen": 1605304320
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025941825476429286,
      "loss": 2.7047,
      "theoretical_loss": 3.4952316608909078,
      "tokens_seen": 1605369856
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002594082246740221,
      "loss": 2.8032,
      "theoretical_loss": 3.4952192439336764,
      "tokens_seen": 1605435392
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002593981945837512,
      "loss": 2.9535,
      "theoretical_loss": 3.4952068276252306,
      "tokens_seen": 1605500928
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025938816449348046,
      "loss": 2.9056,
      "theoretical_loss": 3.4951944119655094,
      "tokens_seen": 1605566464
    },
    {
      "epoch": 0.49,
      "objective/train/advantage_avg": 0.4946036636829376,
      "objective/train/docs_used": 909776,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.4154129028320312,
      "objective/train/original_loss": 2.4154131412506104,
      "objective/train/theoretical_loss": 3.495181996954453,
      "objective/train/tokens_used": 1626092000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24467384815216064,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507041215896606,
      "objective/train/weighted_lm_loss": 2.537731647491455,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0432155132293701,
      "theoretical_loss": 3.495181996954453,
      "tokens_seen": 1605632000
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025937813440320964,
      "loss": 2.8352,
      "theoretical_loss": 3.495181996954453,
      "tokens_seen": 1605632000
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002593681043129388,
      "loss": 2.8392,
      "theoretical_loss": 3.495169582592,
      "tokens_seen": 1605697536
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.000259358074222668,
      "loss": 3.0746,
      "theoretical_loss": 3.495157168878091,
      "tokens_seen": 1605763072
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002593480441323972,
      "loss": 2.6401,
      "theoretical_loss": 3.4951447558126647,
      "tokens_seen": 1605828608
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025933801404212636,
      "loss": 2.4781,
      "theoretical_loss": 3.495132343395662,
      "tokens_seen": 1605894144
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002593279839518556,
      "loss": 2.624,
      "theoretical_loss": 3.495119931627022,
      "tokens_seen": 1605959680
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002593179538615847,
      "loss": 2.6882,
      "theoretical_loss": 3.4951075205066844,
      "tokens_seen": 1606025216
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025930792377131396,
      "loss": 2.7858,
      "theoretical_loss": 3.4950951100345886,
      "tokens_seen": 1606090752
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025929789368104314,
      "loss": 2.8968,
      "theoretical_loss": 3.4950827002106744,
      "tokens_seen": 1606156288
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002592878635907723,
      "loss": 2.7364,
      "theoretical_loss": 3.495070291034882,
      "tokens_seen": 1606221824
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002592778335005015,
      "loss": 2.7575,
      "theoretical_loss": 3.495057882507151,
      "tokens_seen": 1606287360
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002592678034102307,
      "loss": 2.7439,
      "theoretical_loss": 3.49504547462742,
      "tokens_seen": 1606352896
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025925777331995987,
      "loss": 2.6104,
      "theoretical_loss": 3.49503306739563,
      "tokens_seen": 1606418432
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002592477432296891,
      "loss": 2.8372,
      "theoretical_loss": 3.4950206608117202,
      "tokens_seen": 1606483968
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025923771313941823,
      "loss": 2.9417,
      "theoretical_loss": 3.4950082548756307,
      "tokens_seen": 1606549504
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025922768304914747,
      "loss": 2.7828,
      "theoretical_loss": 3.494995849587301,
      "tokens_seen": 1606615040
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002592176529588766,
      "loss": 2.7643,
      "theoretical_loss": 3.4949834449466706,
      "tokens_seen": 1606680576
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025920762286860583,
      "loss": 2.8452,
      "theoretical_loss": 3.4949710409536796,
      "tokens_seen": 1606746112
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.000259197592778335,
      "loss": 2.6541,
      "theoretical_loss": 3.4949586376082675,
      "tokens_seen": 1606811648
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002591875626880642,
      "loss": 2.8801,
      "theoretical_loss": 3.4949462349103744,
      "tokens_seen": 1606877184
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025917753259779337,
      "loss": 2.7558,
      "theoretical_loss": 3.49493383285994,
      "tokens_seen": 1606942720
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025916750250752255,
      "loss": 2.9613,
      "theoretical_loss": 3.4949214314569046,
      "tokens_seen": 1607008256
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025915747241725173,
      "loss": 2.8405,
      "theoretical_loss": 3.4949090307012067,
      "tokens_seen": 1607073792
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025914744232698097,
      "loss": 2.7931,
      "theoretical_loss": 3.494896630592787,
      "tokens_seen": 1607139328
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002591374122367101,
      "loss": 2.8203,
      "theoretical_loss": 3.4948842311315853,
      "tokens_seen": 1607204864
    },
    {
      "epoch": 0.49,
      "objective/train/advantage_avg": 0.4931439459323883,
      "objective/train/docs_used": 910493,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.563403367996216,
      "objective/train/original_loss": 2.563403367996216,
      "objective/train/theoretical_loss": 3.4948718323175414,
      "objective/train/tokens_used": 1627730400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2432613968849182,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050550937652588,
      "objective/train/weighted_lm_loss": 2.6928162574768066,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0417574644088745,
      "theoretical_loss": 3.4948718323175414,
      "tokens_seen": 1607270400
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025912738214643933,
      "loss": 3.1038,
      "theoretical_loss": 3.4948718323175414,
      "tokens_seen": 1607270400
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002591173520561685,
      "loss": 3.0252,
      "theoretical_loss": 3.4948594341505954,
      "tokens_seen": 1607335936
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002591073219658977,
      "loss": 2.8183,
      "theoretical_loss": 3.494847036630686,
      "tokens_seen": 1607401472
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002590972918756269,
      "loss": 2.8474,
      "theoretical_loss": 3.4948346397577543,
      "tokens_seen": 1607467008
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025908726178535606,
      "loss": 2.9901,
      "theoretical_loss": 3.4948222435317398,
      "tokens_seen": 1607532544
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025907723169508524,
      "loss": 2.8837,
      "theoretical_loss": 3.4948098479525824,
      "tokens_seen": 1607598080
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025906720160481447,
      "loss": 2.742,
      "theoretical_loss": 3.4947974530202215,
      "tokens_seen": 1607663616
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002590571715145436,
      "loss": 2.6898,
      "theoretical_loss": 3.494785058734598,
      "tokens_seen": 1607729152
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025904714142427283,
      "loss": 3.0195,
      "theoretical_loss": 3.4947726650956508,
      "tokens_seen": 1607794688
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025903711133400196,
      "loss": 2.8331,
      "theoretical_loss": 3.4947602721033197,
      "tokens_seen": 1607860224
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002590270812437312,
      "loss": 3.0705,
      "theoretical_loss": 3.4947478797575453,
      "tokens_seen": 1607925760
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002590170511534604,
      "loss": 2.4255,
      "theoretical_loss": 3.4947354880582675,
      "tokens_seen": 1607991296
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025900702106318956,
      "loss": 2.738,
      "theoretical_loss": 3.4947230970054264,
      "tokens_seen": 1608056832
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002589969909729188,
      "loss": 2.6955,
      "theoretical_loss": 3.4947107065989607,
      "tokens_seen": 1608122368
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.000258986960882648,
      "loss": 2.9167,
      "theoretical_loss": 3.4946983168388117,
      "tokens_seen": 1608187904
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025897693079237716,
      "loss": 2.6132,
      "theoretical_loss": 3.4946859277249187,
      "tokens_seen": 1608253440
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025896690070210634,
      "loss": 2.8631,
      "theoretical_loss": 3.4946735392572217,
      "tokens_seen": 1608318976
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002589568706118355,
      "loss": 2.7363,
      "theoretical_loss": 3.494661151435661,
      "tokens_seen": 1608384512
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002589468405215647,
      "loss": 2.9309,
      "theoretical_loss": 3.4946487642601762,
      "tokens_seen": 1608450048
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025893681043129393,
      "loss": 2.8239,
      "theoretical_loss": 3.4946363777307075,
      "tokens_seen": 1608515584
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025892678034102306,
      "loss": 2.6542,
      "theoretical_loss": 3.494623991847195,
      "tokens_seen": 1608581120
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002589167502507523,
      "loss": 2.8011,
      "theoretical_loss": 3.494611606609578,
      "tokens_seen": 1608646656
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002589067201604814,
      "loss": 2.6457,
      "theoretical_loss": 3.494599222017797,
      "tokens_seen": 1608712192
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025889669007021066,
      "loss": 2.8045,
      "theoretical_loss": 3.4945868380717924,
      "tokens_seen": 1608777728
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025888665997993984,
      "loss": 2.8836,
      "theoretical_loss": 3.494574454771504,
      "tokens_seen": 1608843264
    },
    {
      "epoch": 0.49,
      "objective/train/advantage_avg": 0.49185243248939514,
      "objective/train/docs_used": 911872,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.649402141571045,
      "objective/train/original_loss": 2.649402379989624,
      "objective/train/theoretical_loss": 3.4945620721168713,
      "objective/train/tokens_used": 1629368800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24202117323875427,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0504153966903687,
      "objective/train/weighted_lm_loss": 2.7831664085388184,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0396549701690674,
      "theoretical_loss": 3.4945620721168713,
      "tokens_seen": 1608908800
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.000258876629889669,
      "loss": 2.8805,
      "theoretical_loss": 3.4945620721168713,
      "tokens_seen": 1608908800
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002588665997993982,
      "loss": 2.825,
      "theoretical_loss": 3.494549690107835,
      "tokens_seen": 1608974336
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002588565697091274,
      "loss": 2.6336,
      "theoretical_loss": 3.4945373087443348,
      "tokens_seen": 1609039872
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025884653961885657,
      "loss": 2.8066,
      "theoretical_loss": 3.4945249280263107,
      "tokens_seen": 1609105408
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002588365095285858,
      "loss": 3.0048,
      "theoretical_loss": 3.494512547953703,
      "tokens_seen": 1609170944
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025882647943831493,
      "loss": 2.7291,
      "theoretical_loss": 3.4945001685264514,
      "tokens_seen": 1609236480
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025881644934804416,
      "loss": 2.8584,
      "theoretical_loss": 3.494487789744497,
      "tokens_seen": 1609302016
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025880641925777334,
      "loss": 2.7276,
      "theoretical_loss": 3.494475411607778,
      "tokens_seen": 1609367552
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002587963891675025,
      "loss": 2.6129,
      "theoretical_loss": 3.494463034116236,
      "tokens_seen": 1609433088
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002587863590772317,
      "loss": 2.7443,
      "theoretical_loss": 3.494450657269811,
      "tokens_seen": 1609498624
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002587763289869609,
      "loss": 2.7265,
      "theoretical_loss": 3.494438281068443,
      "tokens_seen": 1609564160
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025876629889669007,
      "loss": 2.9198,
      "theoretical_loss": 3.494425905512071,
      "tokens_seen": 1609629696
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002587562688064193,
      "loss": 2.6057,
      "theoretical_loss": 3.494413530600637,
      "tokens_seen": 1609695232
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025874623871614843,
      "loss": 2.8766,
      "theoretical_loss": 3.4944011563340798,
      "tokens_seen": 1609760768
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025873620862587767,
      "loss": 2.8593,
      "theoretical_loss": 3.49438878271234,
      "tokens_seen": 1609826304
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002587261785356068,
      "loss": 2.9655,
      "theoretical_loss": 3.494376409735358,
      "tokens_seen": 1609891840
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025871614844533603,
      "loss": 2.5411,
      "theoretical_loss": 3.4943640374030736,
      "tokens_seen": 1609957376
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002587061183550652,
      "loss": 2.8728,
      "theoretical_loss": 3.4943516657154268,
      "tokens_seen": 1610022912
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002586960882647944,
      "loss": 2.8494,
      "theoretical_loss": 3.4943392946723577,
      "tokens_seen": 1610088448
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025868605817452357,
      "loss": 2.7246,
      "theoretical_loss": 3.4943269242738073,
      "tokens_seen": 1610153984
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025867602808425275,
      "loss": 2.8394,
      "theoretical_loss": 3.494314554519715,
      "tokens_seen": 1610219520
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025866599799398193,
      "loss": 2.7818,
      "theoretical_loss": 3.494302185410022,
      "tokens_seen": 1610285056
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025865596790371117,
      "loss": 2.9862,
      "theoretical_loss": 3.494289816944667,
      "tokens_seen": 1610350592
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002586459378134403,
      "loss": 2.697,
      "theoretical_loss": 3.4942774491235906,
      "tokens_seen": 1610416128
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025863590772316953,
      "loss": 2.9064,
      "theoretical_loss": 3.4942650819467342,
      "tokens_seen": 1610481664
    },
    {
      "epoch": 0.49,
      "objective/train/advantage_avg": 0.49366283416748047,
      "objective/train/docs_used": 912654,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.744392156600952,
      "objective/train/original_loss": 2.744391918182373,
      "objective/train/theoretical_loss": 3.494252715414037,
      "objective/train/tokens_used": 1631007200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24381709098815918,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506056547164917,
      "objective/train/weighted_lm_loss": 2.8832502365112305,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0253151655197144,
      "theoretical_loss": 3.494252715414037,
      "tokens_seen": 1610547200
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002586258776328987,
      "loss": 2.9101,
      "theoretical_loss": 3.494252715414037,
      "tokens_seen": 1610547200
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002586158475426279,
      "loss": 2.7365,
      "theoretical_loss": 3.4942403495254393,
      "tokens_seen": 1610612736
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002586058174523571,
      "loss": 2.7998,
      "theoretical_loss": 3.4942279842808817,
      "tokens_seen": 1610678272
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025859578736208626,
      "loss": 2.7146,
      "theoretical_loss": 3.494215619680304,
      "tokens_seen": 1610743808
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025858575727181544,
      "loss": 3.0399,
      "theoretical_loss": 3.4942032557236473,
      "tokens_seen": 1610809344
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025857572718154467,
      "loss": 2.5764,
      "theoretical_loss": 3.4941908924108507,
      "tokens_seen": 1610874880
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002585656970912738,
      "loss": 2.7812,
      "theoretical_loss": 3.4941785297418555,
      "tokens_seen": 1610940416
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025855566700100303,
      "loss": 2.8923,
      "theoretical_loss": 3.4941661677166014,
      "tokens_seen": 1611005952
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025854563691073216,
      "loss": 2.6656,
      "theoretical_loss": 3.494153806335029,
      "tokens_seen": 1611071488
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002585356068204614,
      "loss": 2.7509,
      "theoretical_loss": 3.4941414455970787,
      "tokens_seen": 1611137024
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002585255767301906,
      "loss": 2.6818,
      "theoretical_loss": 3.4941290855026903,
      "tokens_seen": 1611202560
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025851554663991976,
      "loss": 2.9486,
      "theoretical_loss": 3.4941167260518045,
      "tokens_seen": 1611268096
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025850551654964894,
      "loss": 2.8422,
      "theoretical_loss": 3.4941043672443612,
      "tokens_seen": 1611333632
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002584954864593782,
      "loss": 2.9363,
      "theoretical_loss": 3.494092009080301,
      "tokens_seen": 1611399168
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002584854563691073,
      "loss": 2.8955,
      "theoretical_loss": 3.4940796515595647,
      "tokens_seen": 1611464704
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025847542627883654,
      "loss": 3.0255,
      "theoretical_loss": 3.4940672946820923,
      "tokens_seen": 1611530240
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025846539618856566,
      "loss": 2.7276,
      "theoretical_loss": 3.494054938447824,
      "tokens_seen": 1611595776
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002584553660982949,
      "loss": 2.6977,
      "theoretical_loss": 3.4940425828567,
      "tokens_seen": 1611661312
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002584453360080241,
      "loss": 2.6877,
      "theoretical_loss": 3.4940302279086612,
      "tokens_seen": 1611726848
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025843530591775326,
      "loss": 2.7266,
      "theoretical_loss": 3.494017873603648,
      "tokens_seen": 1611792384
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025842527582748244,
      "loss": 2.7422,
      "theoretical_loss": 3.4940055199416005,
      "tokens_seen": 1611857920
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002584152457372116,
      "loss": 2.7313,
      "theoretical_loss": 3.4939931669224586,
      "tokens_seen": 1611923456
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002584052156469408,
      "loss": 2.5691,
      "theoretical_loss": 3.4939808145461635,
      "tokens_seen": 1611988992
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025839518555667004,
      "loss": 2.8801,
      "theoretical_loss": 3.4939684628126555,
      "tokens_seen": 1612054528
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025838515546639917,
      "loss": 2.8189,
      "theoretical_loss": 3.4939561117218743,
      "tokens_seen": 1612120064
    },
    {
      "epoch": 0.49,
      "objective/train/advantage_avg": 0.49334225058555603,
      "objective/train/docs_used": 913987,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.9730091094970703,
      "objective/train/original_loss": 2.9730093479156494,
      "objective/train/theoretical_loss": 3.493943761273761,
      "objective/train/tokens_used": 1632645600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24346016347408295,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505716800689697,
      "objective/train/weighted_lm_loss": 3.123471260070801,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0425469875335693,
      "theoretical_loss": 3.493943761273761,
      "tokens_seen": 1612185600
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002583751253761284,
      "loss": 2.875,
      "theoretical_loss": 3.493943761273761,
      "tokens_seen": 1612185600
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025836509528585753,
      "loss": 2.7828,
      "theoretical_loss": 3.4939314114682567,
      "tokens_seen": 1612251136
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025835506519558677,
      "loss": 2.5482,
      "theoretical_loss": 3.4939190623053005,
      "tokens_seen": 1612316672
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025834503510531595,
      "loss": 2.7942,
      "theoretical_loss": 3.4939067137848334,
      "tokens_seen": 1612382208
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025833500501504513,
      "loss": 2.794,
      "theoretical_loss": 3.493894365906796,
      "tokens_seen": 1612447744
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002583249749247743,
      "loss": 2.8645,
      "theoretical_loss": 3.493882018671129,
      "tokens_seen": 1612513280
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025831494483450354,
      "loss": 2.9497,
      "theoretical_loss": 3.4938696720777718,
      "tokens_seen": 1612578816
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025830491474423267,
      "loss": 2.6097,
      "theoretical_loss": 3.493857326126666,
      "tokens_seen": 1612644352
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002582948846539619,
      "loss": 2.9714,
      "theoretical_loss": 3.4938449808177516,
      "tokens_seen": 1612709888
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025828485456369103,
      "loss": 2.9049,
      "theoretical_loss": 3.493832636150969,
      "tokens_seen": 1612775424
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025827482447342027,
      "loss": 2.8572,
      "theoretical_loss": 3.4938202921262596,
      "tokens_seen": 1612840960
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025826479438314945,
      "loss": 2.8005,
      "theoretical_loss": 3.4938079487435623,
      "tokens_seen": 1612906496
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025825476429287863,
      "loss": 2.8619,
      "theoretical_loss": 3.493795606002819,
      "tokens_seen": 1612972032
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025824473420260787,
      "loss": 2.882,
      "theoretical_loss": 3.4937832639039703,
      "tokens_seen": 1613037568
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.000258234704112337,
      "loss": 2.6001,
      "theoretical_loss": 3.4937709224469558,
      "tokens_seen": 1613103104
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025822467402206623,
      "loss": 2.7274,
      "theoretical_loss": 3.4937585816317167,
      "tokens_seen": 1613168640
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002582146439317954,
      "loss": 2.4011,
      "theoretical_loss": 3.493746241458193,
      "tokens_seen": 1613234176
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002582046138415246,
      "loss": 2.8015,
      "theoretical_loss": 3.4937339019263254,
      "tokens_seen": 1613299712
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025819458375125377,
      "loss": 2.9455,
      "theoretical_loss": 3.493721563036055,
      "tokens_seen": 1613365248
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025818455366098295,
      "loss": 2.6707,
      "theoretical_loss": 3.493709224787322,
      "tokens_seen": 1613430784
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025817452357071213,
      "loss": 2.7566,
      "theoretical_loss": 3.493696887180067,
      "tokens_seen": 1613496320
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025816449348044137,
      "loss": 2.7435,
      "theoretical_loss": 3.4936845502142306,
      "tokens_seen": 1613561856
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002581544633901705,
      "loss": 2.8745,
      "theoretical_loss": 3.493672213889754,
      "tokens_seen": 1613627392
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025814443329989973,
      "loss": 2.8637,
      "theoretical_loss": 3.493659878206577,
      "tokens_seen": 1613692928
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002581344032096289,
      "loss": 2.7429,
      "theoretical_loss": 3.4936475431646397,
      "tokens_seen": 1613758464
    },
    {
      "epoch": 0.49,
      "objective/train/advantage_avg": 0.49470868706703186,
      "objective/train/docs_used": 914763,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.87833571434021,
      "objective/train/original_loss": 2.87833571434021,
      "objective/train/theoretical_loss": 3.4936352087638842,
      "objective/train/tokens_used": 1634284000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24480891227722168,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507153272628784,
      "objective/train/weighted_lm_loss": 3.024430274963379,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0338950157165527,
      "theoretical_loss": 3.4936352087638842,
      "tokens_seen": 1613824000
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002581243731193581,
      "loss": 2.7131,
      "theoretical_loss": 3.4936352087638842,
      "tokens_seen": 1613824000
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002581143430290873,
      "loss": 2.7574,
      "theoretical_loss": 3.49362287500425,
      "tokens_seen": 1613889536
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025810431293881646,
      "loss": 2.8656,
      "theoretical_loss": 3.493610541885679,
      "tokens_seen": 1613955072
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025809428284854564,
      "loss": 3.0227,
      "theoretical_loss": 3.4935982094081104,
      "tokens_seen": 1614020608
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025808425275827487,
      "loss": 2.9088,
      "theoretical_loss": 3.493585877571485,
      "tokens_seen": 1614086144
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.000258074222668004,
      "loss": 2.6512,
      "theoretical_loss": 3.493573546375745,
      "tokens_seen": 1614151680
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025806419257773323,
      "loss": 2.7208,
      "theoretical_loss": 3.4935612158208293,
      "tokens_seen": 1614217216
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025805416248746236,
      "loss": 2.9435,
      "theoretical_loss": 3.49354888590668,
      "tokens_seen": 1614282752
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002580441323971916,
      "loss": 2.6564,
      "theoretical_loss": 3.493536556633236,
      "tokens_seen": 1614348288
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002580341023069208,
      "loss": 2.8579,
      "theoretical_loss": 3.49352422800044,
      "tokens_seen": 1614413824
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025802407221664996,
      "loss": 2.773,
      "theoretical_loss": 3.4935119000082318,
      "tokens_seen": 1614479360
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025801404212637914,
      "loss": 2.8472,
      "theoretical_loss": 3.493499572656552,
      "tokens_seen": 1614544896
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002580040120361084,
      "loss": 2.5992,
      "theoretical_loss": 3.4934872459453414,
      "tokens_seen": 1614610432
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002579939819458375,
      "loss": 2.7839,
      "theoretical_loss": 3.4934749198745405,
      "tokens_seen": 1614675968
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025798395185556674,
      "loss": 2.7013,
      "theoretical_loss": 3.4934625944440905,
      "tokens_seen": 1614741504
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025797392176529586,
      "loss": 2.9891,
      "theoretical_loss": 3.4934502696539322,
      "tokens_seen": 1614807040
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002579638916750251,
      "loss": 2.648,
      "theoretical_loss": 3.493437945504006,
      "tokens_seen": 1614872576
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002579538615847543,
      "loss": 2.7064,
      "theoretical_loss": 3.4934256219942523,
      "tokens_seen": 1614938112
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025794383149448346,
      "loss": 2.7373,
      "theoretical_loss": 3.4934132991246125,
      "tokens_seen": 1615003648
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025793380140421264,
      "loss": 2.8411,
      "theoretical_loss": 3.4934009768950274,
      "tokens_seen": 1615069184
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002579237713139418,
      "loss": 2.7414,
      "theoretical_loss": 3.493388655305438,
      "tokens_seen": 1615134720
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.000257913741223671,
      "loss": 2.6034,
      "theoretical_loss": 3.4933763343557844,
      "tokens_seen": 1615200256
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025790371113340024,
      "loss": 2.7647,
      "theoretical_loss": 3.4933640140460076,
      "tokens_seen": 1615265792
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025789368104312937,
      "loss": 2.7451,
      "theoretical_loss": 3.4933516943760488,
      "tokens_seen": 1615331328
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002578836509528586,
      "loss": 2.8049,
      "theoretical_loss": 3.493339375345848,
      "tokens_seen": 1615396864
    },
    {
      "epoch": 0.49,
      "objective/train/advantage_avg": 0.49277687072753906,
      "objective/train/docs_used": 916021,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.4977214336395264,
      "objective/train/original_loss": 2.4977214336395264,
      "objective/train/theoretical_loss": 3.493327056955347,
      "objective/train/tokens_used": 1635922400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24292872846126556,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505125522613525,
      "objective/train/weighted_lm_loss": 2.6239001750946045,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0304546356201172,
      "theoretical_loss": 3.493327056955347,
      "tokens_seen": 1615462400
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025787362086258773,
      "loss": 2.732,
      "theoretical_loss": 3.493327056955347,
      "tokens_seen": 1615462400
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025786359077231697,
      "loss": 2.8794,
      "theoretical_loss": 3.493314739204486,
      "tokens_seen": 1615527936
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025785356068204615,
      "loss": 2.8455,
      "theoretical_loss": 3.4933024220932056,
      "tokens_seen": 1615593472
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025784353059177533,
      "loss": 2.9549,
      "theoretical_loss": 3.4932901056214476,
      "tokens_seen": 1615659008
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002578335005015045,
      "loss": 2.682,
      "theoretical_loss": 3.4932777897891523,
      "tokens_seen": 1615724544
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025782347041123374,
      "loss": 2.9016,
      "theoretical_loss": 3.49326547459626,
      "tokens_seen": 1615790080
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025781344032096287,
      "loss": 2.6986,
      "theoretical_loss": 3.493253160042713,
      "tokens_seen": 1615855616
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002578034102306921,
      "loss": 2.6259,
      "theoretical_loss": 3.493240846128451,
      "tokens_seen": 1615921152
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025779338014042123,
      "loss": 2.6378,
      "theoretical_loss": 3.493228532853415,
      "tokens_seen": 1615986688
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025778335005015047,
      "loss": 2.9956,
      "theoretical_loss": 3.4932162202175467,
      "tokens_seen": 1616052224
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025777331995987965,
      "loss": 3.0558,
      "theoretical_loss": 3.493203908220786,
      "tokens_seen": 1616117760
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025776328986960883,
      "loss": 3.0608,
      "theoretical_loss": 3.493191596863074,
      "tokens_seen": 1616183296
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.000257753259779338,
      "loss": 2.9525,
      "theoretical_loss": 3.4931792861443522,
      "tokens_seen": 1616248832
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002577432296890672,
      "loss": 2.9678,
      "theoretical_loss": 3.493166976064561,
      "tokens_seen": 1616314368
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002577331995987964,
      "loss": 2.9539,
      "theoretical_loss": 3.493154666623642,
      "tokens_seen": 1616379904
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002577231695085256,
      "loss": 2.6571,
      "theoretical_loss": 3.4931423578215357,
      "tokens_seen": 1616445440
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025771313941825474,
      "loss": 2.9141,
      "theoretical_loss": 3.4931300496581823,
      "tokens_seen": 1616510976
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025770310932798397,
      "loss": 2.6498,
      "theoretical_loss": 3.4931177421335238,
      "tokens_seen": 1616576512
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002576930792377131,
      "loss": 2.7708,
      "theoretical_loss": 3.493105435247501,
      "tokens_seen": 1616642048
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025768304914744233,
      "loss": 2.7981,
      "theoretical_loss": 3.4930931290000546,
      "tokens_seen": 1616707584
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002576730190571715,
      "loss": 2.7518,
      "theoretical_loss": 3.4930808233911255,
      "tokens_seen": 1616773120
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002576629889669007,
      "loss": 2.8722,
      "theoretical_loss": 3.493068518420655,
      "tokens_seen": 1616838656
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002576529588766299,
      "loss": 2.647,
      "theoretical_loss": 3.4930562140885844,
      "tokens_seen": 1616904192
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002576429287863591,
      "loss": 2.791,
      "theoretical_loss": 3.493043910394854,
      "tokens_seen": 1616969728
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025763289869608824,
      "loss": 2.6594,
      "theoretical_loss": 3.4930316073394048,
      "tokens_seen": 1617035264
    },
    {
      "epoch": 0.49,
      "objective/train/advantage_avg": 0.49574023485183716,
      "objective/train/docs_used": 916782,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.395338296890259,
      "objective/train/original_loss": 2.395338296890259,
      "objective/train/theoretical_loss": 3.4930193049221785,
      "objective/train/tokens_used": 1637560800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2457919865846634,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050823450088501,
      "objective/train/weighted_lm_loss": 2.516979694366455,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0432155132293701,
      "theoretical_loss": 3.4930193049221785,
      "tokens_seen": 1617100800
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002576228686058175,
      "loss": 2.7092,
      "theoretical_loss": 3.4930193049221785,
      "tokens_seen": 1617100800
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002576128385155466,
      "loss": 2.8599,
      "theoretical_loss": 3.4930070031431155,
      "tokens_seen": 1617166336
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025760280842527584,
      "loss": 2.6776,
      "theoretical_loss": 3.4929947020021572,
      "tokens_seen": 1617231872
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.000257592778335005,
      "loss": 2.5802,
      "theoretical_loss": 3.4929824014992445,
      "tokens_seen": 1617297408
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002575827482447342,
      "loss": 2.6626,
      "theoretical_loss": 3.4929701016343184,
      "tokens_seen": 1617362944
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002575727181544634,
      "loss": 2.8837,
      "theoretical_loss": 3.4929578024073202,
      "tokens_seen": 1617428480
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025756268806419256,
      "loss": 2.728,
      "theoretical_loss": 3.492945503818191,
      "tokens_seen": 1617494016
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025755265797392174,
      "loss": 2.7819,
      "theoretical_loss": 3.492933205866871,
      "tokens_seen": 1617559552
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.000257542627883651,
      "loss": 2.6634,
      "theoretical_loss": 3.4929209085533026,
      "tokens_seen": 1617625088
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002575325977933801,
      "loss": 2.7699,
      "theoretical_loss": 3.492908611877426,
      "tokens_seen": 1617690624
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025752256770310934,
      "loss": 2.691,
      "theoretical_loss": 3.492896315839183,
      "tokens_seen": 1617756160
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025751253761283847,
      "loss": 2.8079,
      "theoretical_loss": 3.492884020438514,
      "tokens_seen": 1617821696
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002575025075225677,
      "loss": 2.8716,
      "theoretical_loss": 3.4928717256753603,
      "tokens_seen": 1617887232
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025749247743229694,
      "loss": 2.7543,
      "theoretical_loss": 3.492859431549663,
      "tokens_seen": 1617952768
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025748244734202606,
      "loss": 3.0544,
      "theoretical_loss": 3.4928471380613635,
      "tokens_seen": 1618018304
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002574724172517553,
      "loss": 2.6401,
      "theoretical_loss": 3.4928348452104028,
      "tokens_seen": 1618083840
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002574623871614845,
      "loss": 2.859,
      "theoretical_loss": 3.492822552996722,
      "tokens_seen": 1618149376
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025745235707121366,
      "loss": 2.8934,
      "theoretical_loss": 3.4928102614202623,
      "tokens_seen": 1618214912
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025744232698094284,
      "loss": 2.742,
      "theoretical_loss": 3.492797970480965,
      "tokens_seen": 1618280448
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.000257432296890672,
      "loss": 2.6658,
      "theoretical_loss": 3.4927856801787707,
      "tokens_seen": 1618345984
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002574222668004012,
      "loss": 2.6475,
      "theoretical_loss": 3.492773390513621,
      "tokens_seen": 1618411520
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025741223671013044,
      "loss": 2.5797,
      "theoretical_loss": 3.492761101485457,
      "tokens_seen": 1618477056
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025740220661985957,
      "loss": 2.6056,
      "theoretical_loss": 3.492748813094221,
      "tokens_seen": 1618542592
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002573921765295888,
      "loss": 2.9918,
      "theoretical_loss": 3.4927365253398523,
      "tokens_seen": 1618608128
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025738214643931793,
      "loss": 2.8786,
      "theoretical_loss": 3.492724238222293,
      "tokens_seen": 1618673664
    },
    {
      "epoch": 0.49,
      "objective/train/advantage_avg": 0.493866890668869,
      "objective/train/docs_used": 918149,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.328277349472046,
      "objective/train/original_loss": 2.328277349472046,
      "objective/train/theoretical_loss": 3.4927119517414846,
      "objective/train/tokens_used": 1639199200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24397164583206177,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506268739700317,
      "objective/train/weighted_lm_loss": 2.446045398712158,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0396549701690674,
      "theoretical_loss": 3.4927119517414846,
      "tokens_seen": 1618739200
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025737211634904717,
      "loss": 2.7562,
      "theoretical_loss": 3.4927119517414846,
      "tokens_seen": 1618739200
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025736208625877635,
      "loss": 2.9047,
      "theoretical_loss": 3.4926996658973675,
      "tokens_seen": 1618804736
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025735205616850553,
      "loss": 2.8113,
      "theoretical_loss": 3.492687380689884,
      "tokens_seen": 1618870272
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002573420260782347,
      "loss": 2.5867,
      "theoretical_loss": 3.4926750961189743,
      "tokens_seen": 1618935808
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025733199598796394,
      "loss": 2.5445,
      "theoretical_loss": 3.49266281218458,
      "tokens_seen": 1619001344
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025732196589769307,
      "loss": 2.9013,
      "theoretical_loss": 3.492650528886643,
      "tokens_seen": 1619066880
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002573119358074223,
      "loss": 2.8065,
      "theoretical_loss": 3.4926382462251047,
      "tokens_seen": 1619132416
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025730190571715143,
      "loss": 2.8249,
      "theoretical_loss": 3.4926259641999047,
      "tokens_seen": 1619197952
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025729187562688067,
      "loss": 2.6308,
      "theoretical_loss": 3.4926136828109855,
      "tokens_seen": 1619263488
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025728184553660985,
      "loss": 2.9143,
      "theoretical_loss": 3.4926014020582885,
      "tokens_seen": 1619329024
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025727181544633903,
      "loss": 2.6465,
      "theoretical_loss": 3.492589121941754,
      "tokens_seen": 1619394560
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002572617853560682,
      "loss": 2.9576,
      "theoretical_loss": 3.492576842461325,
      "tokens_seen": 1619460096
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002572517552657974,
      "loss": 3.0916,
      "theoretical_loss": 3.4925645636169413,
      "tokens_seen": 1619525632
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002572417251755266,
      "loss": 2.432,
      "theoretical_loss": 3.4925522854085447,
      "tokens_seen": 1619591168
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002572316950852558,
      "loss": 2.6734,
      "theoretical_loss": 3.4925400078360767,
      "tokens_seen": 1619656704
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025722166499498494,
      "loss": 2.7875,
      "theoretical_loss": 3.492527730899478,
      "tokens_seen": 1619722240
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025721163490471417,
      "loss": 2.8689,
      "theoretical_loss": 3.492515454598691,
      "tokens_seen": 1619787776
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002572016048144433,
      "loss": 2.6936,
      "theoretical_loss": 3.492503178933656,
      "tokens_seen": 1619853312
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025719157472417253,
      "loss": 2.6958,
      "theoretical_loss": 3.492490903904315,
      "tokens_seen": 1619918848
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002571815446339017,
      "loss": 2.7069,
      "theoretical_loss": 3.4924786295106096,
      "tokens_seen": 1619984384
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002571715145436309,
      "loss": 2.7746,
      "theoretical_loss": 3.49246635575248,
      "tokens_seen": 1620049920
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002571614844533601,
      "loss": 2.8034,
      "theoretical_loss": 3.492454082629869,
      "tokens_seen": 1620115456
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002571514543630893,
      "loss": 2.6393,
      "theoretical_loss": 3.4924418101427173,
      "tokens_seen": 1620180992
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025714142427281844,
      "loss": 2.6786,
      "theoretical_loss": 3.492429538290966,
      "tokens_seen": 1620246528
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002571313941825477,
      "loss": 2.8278,
      "theoretical_loss": 3.492417267074557,
      "tokens_seen": 1620312064
    },
    {
      "epoch": 0.49,
      "objective/train/advantage_avg": 0.4920796751976013,
      "objective/train/docs_used": 918561,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.250178098678589,
      "objective/train/original_loss": 2.2501778602600098,
      "objective/train/theoretical_loss": 3.4924049964934314,
      "objective/train/tokens_used": 1640837600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2422625571489334,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050439476966858,
      "objective/train/weighted_lm_loss": 2.3636348247528076,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0338950157165527,
      "theoretical_loss": 3.4924049964934314,
      "tokens_seen": 1620377600
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002571213640922768,
      "loss": 2.819,
      "theoretical_loss": 3.4924049964934314,
      "tokens_seen": 1620377600
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025711133400200604,
      "loss": 2.762,
      "theoretical_loss": 3.492392726547531,
      "tokens_seen": 1620443136
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002571013039117352,
      "loss": 2.6045,
      "theoretical_loss": 3.492380457236797,
      "tokens_seen": 1620508672
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002570912738214644,
      "loss": 2.8524,
      "theoretical_loss": 3.4923681885611706,
      "tokens_seen": 1620574208
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002570812437311936,
      "loss": 2.7637,
      "theoretical_loss": 3.492355920520594,
      "tokens_seen": 1620639744
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025707121364092276,
      "loss": 2.8538,
      "theoretical_loss": 3.4923436531150074,
      "tokens_seen": 1620705280
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025706118355065194,
      "loss": 2.8906,
      "theoretical_loss": 3.4923313863443535,
      "tokens_seen": 1620770816
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002570511534603812,
      "loss": 2.7881,
      "theoretical_loss": 3.492319120208573,
      "tokens_seen": 1620836352
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002570411233701103,
      "loss": 2.7838,
      "theoretical_loss": 3.4923068547076075,
      "tokens_seen": 1620901888
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025703109327983954,
      "loss": 2.8098,
      "theoretical_loss": 3.492294589841399,
      "tokens_seen": 1620967424
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025702106318956867,
      "loss": 2.8172,
      "theoretical_loss": 3.4922823256098883,
      "tokens_seen": 1621032960
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002570110330992979,
      "loss": 2.7256,
      "theoretical_loss": 3.4922700620130174,
      "tokens_seen": 1621098496
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002570010030090271,
      "loss": 2.855,
      "theoretical_loss": 3.4922577990507273,
      "tokens_seen": 1621164032
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025699097291875626,
      "loss": 2.7884,
      "theoretical_loss": 3.49224553672296,
      "tokens_seen": 1621229568
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025698094282848545,
      "loss": 2.818,
      "theoretical_loss": 3.4922332750296565,
      "tokens_seen": 1621295104
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002569709127382147,
      "loss": 2.8694,
      "theoretical_loss": 3.492221013970759,
      "tokens_seen": 1621360640
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002569608826479438,
      "loss": 2.773,
      "theoretical_loss": 3.4922087535462083,
      "tokens_seen": 1621426176
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025695085255767304,
      "loss": 2.6898,
      "theoretical_loss": 3.492196493755947,
      "tokens_seen": 1621491712
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025694082246740217,
      "loss": 2.6348,
      "theoretical_loss": 3.4921842345999154,
      "tokens_seen": 1621557248
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002569307923771314,
      "loss": 2.6066,
      "theoretical_loss": 3.4921719760780556,
      "tokens_seen": 1621622784
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002569207622868606,
      "loss": 2.732,
      "theoretical_loss": 3.49215971819031,
      "tokens_seen": 1621688320
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025691073219658977,
      "loss": 2.7354,
      "theoretical_loss": 3.4921474609366188,
      "tokens_seen": 1621753856
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025690070210631895,
      "loss": 2.8178,
      "theoretical_loss": 3.492135204316924,
      "tokens_seen": 1621819392
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025689067201604813,
      "loss": 2.7146,
      "theoretical_loss": 3.4921229483311675,
      "tokens_seen": 1621884928
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002568806419257773,
      "loss": 2.9808,
      "theoretical_loss": 3.492110692979291,
      "tokens_seen": 1621950464
    },
    {
      "epoch": 0.49,
      "objective/train/advantage_avg": 0.4930935502052307,
      "objective/train/docs_used": 919775,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.954005718231201,
      "objective/train/original_loss": 2.9540061950683594,
      "objective/train/theoretical_loss": 3.4920984382612357,
      "objective/train/tokens_used": 1642476000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2432100474834442,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505456924438477,
      "objective/train/weighted_lm_loss": 3.1035451889038086,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0442860126495361,
      "theoretical_loss": 3.4920984382612357,
      "tokens_seen": 1622016000
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025687061183550655,
      "loss": 2.7055,
      "theoretical_loss": 3.4920984382612357,
      "tokens_seen": 1622016000
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002568605817452357,
      "loss": 2.6524,
      "theoretical_loss": 3.492086184176943,
      "tokens_seen": 1622081536
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002568505516549649,
      "loss": 2.532,
      "theoretical_loss": 3.492073930726355,
      "tokens_seen": 1622147072
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025684052156469404,
      "loss": 2.8509,
      "theoretical_loss": 3.4920616779094136,
      "tokens_seen": 1622212608
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025683049147442327,
      "loss": 2.6511,
      "theoretical_loss": 3.4920494257260604,
      "tokens_seen": 1622278144
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025682046138415245,
      "loss": 2.7917,
      "theoretical_loss": 3.492037174176236,
      "tokens_seen": 1622343680
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025681043129388163,
      "loss": 2.7905,
      "theoretical_loss": 3.4920249232598826,
      "tokens_seen": 1622409216
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002568004012036108,
      "loss": 2.771,
      "theoretical_loss": 3.4920126729769425,
      "tokens_seen": 1622474752
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025679037111334005,
      "loss": 3.0521,
      "theoretical_loss": 3.492000423327357,
      "tokens_seen": 1622540288
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002567803410230692,
      "loss": 2.999,
      "theoretical_loss": 3.491988174311067,
      "tokens_seen": 1622605824
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002567703109327984,
      "loss": 2.7147,
      "theoretical_loss": 3.491975925928015,
      "tokens_seen": 1622671360
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025676028084252754,
      "loss": 2.5379,
      "theoretical_loss": 3.491963678178143,
      "tokens_seen": 1622736896
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002567502507522568,
      "loss": 2.748,
      "theoretical_loss": 3.491951431061392,
      "tokens_seen": 1622802432
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.000256740220661986,
      "loss": 3.0978,
      "theoretical_loss": 3.491939184577704,
      "tokens_seen": 1622867968
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025673019057171514,
      "loss": 3.0689,
      "theoretical_loss": 3.491926938727021,
      "tokens_seen": 1622933504
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025672016048144437,
      "loss": 2.9109,
      "theoretical_loss": 3.4919146935092833,
      "tokens_seen": 1622999040
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002567101303911735,
      "loss": 2.7422,
      "theoretical_loss": 3.4919024489244346,
      "tokens_seen": 1623064576
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025670010030090273,
      "loss": 2.5402,
      "theoretical_loss": 3.4918902049724156,
      "tokens_seen": 1623130112
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002566900702106319,
      "loss": 2.8804,
      "theoretical_loss": 3.491877961653168,
      "tokens_seen": 1623195648
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002566800401203611,
      "loss": 2.7788,
      "theoretical_loss": 3.4918657189666336,
      "tokens_seen": 1623261184
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002566700100300903,
      "loss": 2.7017,
      "theoretical_loss": 3.4918534769127545,
      "tokens_seen": 1623326720
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002566599799398195,
      "loss": 2.6759,
      "theoretical_loss": 3.491841235491472,
      "tokens_seen": 1623392256
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025664994984954864,
      "loss": 2.7299,
      "theoretical_loss": 3.4918289947027286,
      "tokens_seen": 1623457792
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002566399197592779,
      "loss": 2.7007,
      "theoretical_loss": 3.4918167545464653,
      "tokens_seen": 1623523328
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.000256629889669007,
      "loss": 2.727,
      "theoretical_loss": 3.4918045150226242,
      "tokens_seen": 1623588864
    },
    {
      "epoch": 0.49,
      "objective/train/advantage_avg": 0.49556538462638855,
      "objective/train/docs_used": 920322,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 3.1794064044952393,
      "objective/train/original_loss": 3.1794064044952393,
      "objective/train/theoretical_loss": 3.491792276131147,
      "objective/train/tokens_used": 1644114400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.245655819773674,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0508053302764893,
      "objective/train/weighted_lm_loss": 3.3410844802856445,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 3.491792276131147,
      "tokens_seen": 1623654400
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025661985957873624,
      "loss": 2.7846,
      "theoretical_loss": 3.491792276131147,
      "tokens_seen": 1623654400
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002566098294884654,
      "loss": 2.8542,
      "theoretical_loss": 3.491780037871976,
      "tokens_seen": 1623719936
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002565997993981946,
      "loss": 2.5978,
      "theoretical_loss": 3.491767800245052,
      "tokens_seen": 1623785472
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002565897693079238,
      "loss": 2.7979,
      "theoretical_loss": 3.491755563250318,
      "tokens_seen": 1623851008
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025657973921765296,
      "loss": 2.6655,
      "theoretical_loss": 3.491743326887715,
      "tokens_seen": 1623916544
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025656970912738214,
      "loss": 2.7276,
      "theoretical_loss": 3.4917310911571855,
      "tokens_seen": 1623982080
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002565596790371114,
      "loss": 2.7441,
      "theoretical_loss": 3.4917188560586707,
      "tokens_seen": 1624047616
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002565496489468405,
      "loss": 2.67,
      "theoretical_loss": 3.4917066215921126,
      "tokens_seen": 1624113152
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025653961885656974,
      "loss": 2.9259,
      "theoretical_loss": 3.491694387757453,
      "tokens_seen": 1624178688
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025652958876629887,
      "loss": 2.7915,
      "theoretical_loss": 3.4916821545546344,
      "tokens_seen": 1624244224
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002565195586760281,
      "loss": 2.7909,
      "theoretical_loss": 3.491669921983598,
      "tokens_seen": 1624309760
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002565095285857573,
      "loss": 2.6202,
      "theoretical_loss": 3.4916576900442857,
      "tokens_seen": 1624375296
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025649949849548647,
      "loss": 2.6624,
      "theoretical_loss": 3.49164545873664,
      "tokens_seen": 1624440832
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025648946840521565,
      "loss": 2.8701,
      "theoretical_loss": 3.4916332280606017,
      "tokens_seen": 1624506368
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002564794383149449,
      "loss": 2.7714,
      "theoretical_loss": 3.4916209980161144,
      "tokens_seen": 1624571904
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.000256469408224674,
      "loss": 2.9416,
      "theoretical_loss": 3.4916087686031183,
      "tokens_seen": 1624637440
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025645937813440324,
      "loss": 2.8486,
      "theoretical_loss": 3.491596539821556,
      "tokens_seen": 1624702976
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025644934804413237,
      "loss": 2.8319,
      "theoretical_loss": 3.4915843116713696,
      "tokens_seen": 1624768512
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002564393179538616,
      "loss": 2.9029,
      "theoretical_loss": 3.4915720841525006,
      "tokens_seen": 1624834048
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002564292878635908,
      "loss": 2.8366,
      "theoretical_loss": 3.4915598572648916,
      "tokens_seen": 1624899584
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025641925777331997,
      "loss": 2.6962,
      "theoretical_loss": 3.4915476310084843,
      "tokens_seen": 1624965120
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025640922768304915,
      "loss": 2.9263,
      "theoretical_loss": 3.49153540538322,
      "tokens_seen": 1625030656
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025639919759277833,
      "loss": 2.7163,
      "theoretical_loss": 3.4915231803890423,
      "tokens_seen": 1625096192
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002563891675025075,
      "loss": 2.9596,
      "theoretical_loss": 3.4915109560258912,
      "tokens_seen": 1625161728
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025637913741223675,
      "loss": 2.8395,
      "theoretical_loss": 3.4914987322937097,
      "tokens_seen": 1625227264
    },
    {
      "epoch": 0.49,
      "objective/train/advantage_avg": 0.4949594736099243,
      "objective/train/docs_used": 921030,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.56559681892395,
      "objective/train/original_loss": 2.565596342086792,
      "objective/train/theoretical_loss": 3.4914865091924394,
      "objective/train/tokens_used": 1645752800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24503014981746674,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507415533065796,
      "objective/train/weighted_lm_loss": 2.6956162452697754,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0469889640808105,
      "theoretical_loss": 3.4914865091924394,
      "tokens_seen": 1625292800
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002563691073219659,
      "loss": 2.7256,
      "theoretical_loss": 3.4914865091924394,
      "tokens_seen": 1625292800
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002563590772316951,
      "loss": 2.8719,
      "theoretical_loss": 3.491474286722023,
      "tokens_seen": 1625358336
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025634904714142424,
      "loss": 2.9145,
      "theoretical_loss": 3.491462064882402,
      "tokens_seen": 1625423872
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025633901705115347,
      "loss": 2.7855,
      "theoretical_loss": 3.491449843673519,
      "tokens_seen": 1625489408
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025632898696088265,
      "loss": 2.8558,
      "theoretical_loss": 3.4914376230953144,
      "tokens_seen": 1625554944
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025631895687061183,
      "loss": 2.7512,
      "theoretical_loss": 3.4914254031477316,
      "tokens_seen": 1625620480
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.000256308926780341,
      "loss": 2.5632,
      "theoretical_loss": 3.491413183830713,
      "tokens_seen": 1625686016
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025629889669007025,
      "loss": 2.7316,
      "theoretical_loss": 3.4914009651441997,
      "tokens_seen": 1625751552
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002562888665997994,
      "loss": 2.8125,
      "theoretical_loss": 3.491388747088134,
      "tokens_seen": 1625817088
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002562788365095286,
      "loss": 2.4607,
      "theoretical_loss": 3.491376529662458,
      "tokens_seen": 1625882624
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025626880641925774,
      "loss": 2.7723,
      "theoretical_loss": 3.491364312867114,
      "tokens_seen": 1625948160
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.000256258776328987,
      "loss": 2.7058,
      "theoretical_loss": 3.4913520967020437,
      "tokens_seen": 1626013696
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025624874623871616,
      "loss": 2.7237,
      "theoretical_loss": 3.4913398811671894,
      "tokens_seen": 1626079232
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025623871614844534,
      "loss": 2.7704,
      "theoretical_loss": 3.4913276662624932,
      "tokens_seen": 1626144768
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002562286860581745,
      "loss": 2.6717,
      "theoretical_loss": 3.491315451987897,
      "tokens_seen": 1626210304
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002562186559679037,
      "loss": 2.9998,
      "theoretical_loss": 3.4913032383433427,
      "tokens_seen": 1626275840
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002562086258776329,
      "loss": 2.4675,
      "theoretical_loss": 3.4912910253287732,
      "tokens_seen": 1626341376
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002561985957873621,
      "loss": 2.9345,
      "theoretical_loss": 3.49127881294413,
      "tokens_seen": 1626406912
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025618856569709124,
      "loss": 2.583,
      "theoretical_loss": 3.4912666011893556,
      "tokens_seen": 1626472448
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002561785356068205,
      "loss": 2.8176,
      "theoretical_loss": 3.491254390064392,
      "tokens_seen": 1626537984
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025616850551654966,
      "loss": 3.0214,
      "theoretical_loss": 3.4912421795691815,
      "tokens_seen": 1626603520
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025615847542627884,
      "loss": 2.6979,
      "theoretical_loss": 3.4912299697036655,
      "tokens_seen": 1626669056
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.000256148445336008,
      "loss": 3.0757,
      "theoretical_loss": 3.4912177604677868,
      "tokens_seen": 1626734592
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002561384152457372,
      "loss": 2.6676,
      "theoretical_loss": 3.491205551861488,
      "tokens_seen": 1626800128
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002561283851554664,
      "loss": 2.8377,
      "theoretical_loss": 3.49119334388471,
      "tokens_seen": 1626865664
    },
    {
      "epoch": 0.49,
      "objective/train/advantage_avg": 0.4927684962749481,
      "objective/train/docs_used": 922470,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.6483607292175293,
      "objective/train/original_loss": 2.6483607292175293,
      "objective/train/theoretical_loss": 3.4911811365373957,
      "objective/train/tokens_used": 1647391200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24289631843566895,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0505115985870361,
      "objective/train/weighted_lm_loss": 2.782064914703369,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.044721245765686,
      "theoretical_loss": 3.4911811365373957,
      "tokens_seen": 1626931200
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002561183550651956,
      "loss": 2.7346,
      "theoretical_loss": 3.4911811365373957,
      "tokens_seen": 1626931200
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025610832497492475,
      "loss": 2.9228,
      "theoretical_loss": 3.491168929819488,
      "tokens_seen": 1626996736
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.000256098294884654,
      "loss": 2.7601,
      "theoretical_loss": 3.491156723730928,
      "tokens_seen": 1627062272
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002560882647943831,
      "loss": 2.5895,
      "theoretical_loss": 3.4911445182716583,
      "tokens_seen": 1627127808
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025607823470411234,
      "loss": 2.9462,
      "theoretical_loss": 3.4911323134416214,
      "tokens_seen": 1627193344
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002560682046138415,
      "loss": 2.8547,
      "theoretical_loss": 3.491120109240759,
      "tokens_seen": 1627258880
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002560581745235707,
      "loss": 2.855,
      "theoretical_loss": 3.4911079056690135,
      "tokens_seen": 1627324416
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002560481444332999,
      "loss": 2.7385,
      "theoretical_loss": 3.4910957027263274,
      "tokens_seen": 1627389952
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025603811434302907,
      "loss": 2.8212,
      "theoretical_loss": 3.4910835004126426,
      "tokens_seen": 1627455488
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025602808425275825,
      "loss": 2.721,
      "theoretical_loss": 3.491071298727902,
      "tokens_seen": 1627521024
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002560180541624875,
      "loss": 2.9196,
      "theoretical_loss": 3.4910590976720464,
      "tokens_seen": 1627586560
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002560080240722166,
      "loss": 2.6713,
      "theoretical_loss": 3.4910468972450195,
      "tokens_seen": 1627652096
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025599799398194585,
      "loss": 3.0454,
      "theoretical_loss": 3.4910346974467634,
      "tokens_seen": 1627717632
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002559879638916751,
      "loss": 2.5837,
      "theoretical_loss": 3.4910224982772196,
      "tokens_seen": 1627783168
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002559779338014042,
      "loss": 2.5678,
      "theoretical_loss": 3.491010299736331,
      "tokens_seen": 1627848704
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025596790371113344,
      "loss": 2.7687,
      "theoretical_loss": 3.49099810182404,
      "tokens_seen": 1627914240
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025595787362086257,
      "loss": 2.6955,
      "theoretical_loss": 3.4909859045402887,
      "tokens_seen": 1627979776
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002559478435305918,
      "loss": 2.9779,
      "theoretical_loss": 3.4909737078850194,
      "tokens_seen": 1628045312
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.000255937813440321,
      "loss": 3.0072,
      "theoretical_loss": 3.490961511858174,
      "tokens_seen": 1628110848
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025592778335005017,
      "loss": 2.7302,
      "theoretical_loss": 3.4909493164596954,
      "tokens_seen": 1628176384
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025591775325977935,
      "loss": 2.8845,
      "theoretical_loss": 3.4909371216895257,
      "tokens_seen": 1628241920
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025590772316950853,
      "loss": 2.9191,
      "theoretical_loss": 3.4909249275476077,
      "tokens_seen": 1628307456
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002558976930792377,
      "loss": 3.0841,
      "theoretical_loss": 3.4909127340338832,
      "tokens_seen": 1628372992
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025588766298896695,
      "loss": 2.8285,
      "theoretical_loss": 3.490900541148295,
      "tokens_seen": 1628438528
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002558776328986961,
      "loss": 2.8768,
      "theoretical_loss": 3.4908883488907847,
      "tokens_seen": 1628504064
    },
    {
      "epoch": 0.49,
      "objective/train/advantage_avg": 0.4948728084564209,
      "objective/train/docs_used": 922615,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.5114705562591553,
      "objective/train/original_loss": 2.5114705562591553,
      "objective/train/theoretical_loss": 3.4908761572612947,
      "objective/train/tokens_used": 1649029600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24496787786483765,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507326126098633,
      "objective/train/weighted_lm_loss": 2.63885235786438,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0338950157165527,
      "theoretical_loss": 3.4908761572612947,
      "tokens_seen": 1628569600
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002558676028084253,
      "loss": 2.7894,
      "theoretical_loss": 3.4908761572612947,
      "tokens_seen": 1628569600
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025585757271815444,
      "loss": 2.5969,
      "theoretical_loss": 3.4908639662597687,
      "tokens_seen": 1628635136
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025584754262788367,
      "loss": 2.6831,
      "theoretical_loss": 3.490851775886148,
      "tokens_seen": 1628700672
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025583751253761285,
      "loss": 2.9217,
      "theoretical_loss": 3.490839586140375,
      "tokens_seen": 1628766208
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025582748244734203,
      "loss": 2.8986,
      "theoretical_loss": 3.4908273970223926,
      "tokens_seen": 1628831744
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002558174523570712,
      "loss": 2.8803,
      "theoretical_loss": 3.4908152085321427,
      "tokens_seen": 1628897280
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025580742226680045,
      "loss": 2.8501,
      "theoretical_loss": 3.490803020669568,
      "tokens_seen": 1628962816
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002557973921765296,
      "loss": 2.8484,
      "theoretical_loss": 3.4907908334346107,
      "tokens_seen": 1629028352
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002557873620862588,
      "loss": 2.7618,
      "theoretical_loss": 3.4907786468272137,
      "tokens_seen": 1629093888
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025577733199598794,
      "loss": 2.5686,
      "theoretical_loss": 3.4907664608473192,
      "tokens_seen": 1629159424
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002557673019057172,
      "loss": 2.8321,
      "theoretical_loss": 3.4907542754948695,
      "tokens_seen": 1629224960
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025575727181544636,
      "loss": 2.9099,
      "theoretical_loss": 3.490742090769807,
      "tokens_seen": 1629290496
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025574724172517554,
      "loss": 2.8141,
      "theoretical_loss": 3.4907299066720743,
      "tokens_seen": 1629356032
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002557372116349047,
      "loss": 2.749,
      "theoretical_loss": 3.490717723201614,
      "tokens_seen": 1629421568
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002557271815446339,
      "loss": 2.8919,
      "theoretical_loss": 3.490705540358369,
      "tokens_seen": 1629487104
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002557171514543631,
      "loss": 2.8264,
      "theoretical_loss": 3.4906933581422805,
      "tokens_seen": 1629552640
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002557071213640923,
      "loss": 2.673,
      "theoretical_loss": 3.490681176553292,
      "tokens_seen": 1629618176
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025569709127382144,
      "loss": 2.6054,
      "theoretical_loss": 3.4906689955913457,
      "tokens_seen": 1629683712
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002556870611835507,
      "loss": 2.6834,
      "theoretical_loss": 3.490656815256384,
      "tokens_seen": 1629749248
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025567703109327986,
      "loss": 2.8699,
      "theoretical_loss": 3.4906446355483496,
      "tokens_seen": 1629814784
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025566700100300904,
      "loss": 2.9132,
      "theoretical_loss": 3.4906324564671856,
      "tokens_seen": 1629880320
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002556569709127382,
      "loss": 2.8234,
      "theoretical_loss": 3.4906202780128335,
      "tokens_seen": 1629945856
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002556469408224674,
      "loss": 3.152,
      "theoretical_loss": 3.490608100185236,
      "tokens_seen": 1630011392
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002556369107321966,
      "loss": 2.8475,
      "theoretical_loss": 3.4905959229843364,
      "tokens_seen": 1630076928
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002556268806419258,
      "loss": 2.7497,
      "theoretical_loss": 3.490583746410077,
      "tokens_seen": 1630142464
    },
    {
      "epoch": 0.49,
      "objective/train/advantage_avg": 0.4955556094646454,
      "objective/train/docs_used": 923942,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.7406790256500244,
      "objective/train/original_loss": 2.7406787872314453,
      "objective/train/theoretical_loss": 3.490571570462399,
      "objective/train/tokens_used": 1650668000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24561750888824463,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0508041381835938,
      "objective/train/weighted_lm_loss": 2.879875898361206,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0408108234405518,
      "theoretical_loss": 3.490571570462399,
      "tokens_seen": 1630208000
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025561685055165495,
      "loss": 2.6751,
      "theoretical_loss": 3.490571570462399,
      "tokens_seen": 1630208000
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002556068204613842,
      "loss": 3.0485,
      "theoretical_loss": 3.490559395141247,
      "tokens_seen": 1630273536
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002555967903711133,
      "loss": 2.8018,
      "theoretical_loss": 3.4905472204465626,
      "tokens_seen": 1630339072
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025558676028084254,
      "loss": 2.9267,
      "theoretical_loss": 3.4905350463782883,
      "tokens_seen": 1630404608
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002555767301905717,
      "loss": 2.8761,
      "theoretical_loss": 3.490522872936367,
      "tokens_seen": 1630470144
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002555667001003009,
      "loss": 2.5097,
      "theoretical_loss": 3.4905107001207414,
      "tokens_seen": 1630535680
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002555566700100301,
      "loss": 2.7662,
      "theoretical_loss": 3.4904985279313534,
      "tokens_seen": 1630601216
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025554663991975927,
      "loss": 2.7861,
      "theoretical_loss": 3.4904863563681463,
      "tokens_seen": 1630666752
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025553660982948845,
      "loss": 2.8106,
      "theoretical_loss": 3.4904741854310624,
      "tokens_seen": 1630732288
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002555265797392177,
      "loss": 2.8585,
      "theoretical_loss": 3.4904620151200447,
      "tokens_seen": 1630797824
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002555165496489468,
      "loss": 2.7196,
      "theoretical_loss": 3.490449845435035,
      "tokens_seen": 1630863360
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025550651955867605,
      "loss": 2.6888,
      "theoretical_loss": 3.490437676375977,
      "tokens_seen": 1630928896
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025549648946840523,
      "loss": 2.9555,
      "theoretical_loss": 3.490425507942813,
      "tokens_seen": 1630994432
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002554864593781344,
      "loss": 2.7135,
      "theoretical_loss": 3.4904133401354853,
      "tokens_seen": 1631059968
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002554764292878636,
      "loss": 2.8684,
      "theoretical_loss": 3.490401172953937,
      "tokens_seen": 1631125504
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025546639919759277,
      "loss": 2.7339,
      "theoretical_loss": 3.49038900639811,
      "tokens_seen": 1631191040
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025545636910732195,
      "loss": 2.8404,
      "theoretical_loss": 3.490376840467948,
      "tokens_seen": 1631256576
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002554463390170512,
      "loss": 2.7782,
      "theoretical_loss": 3.490364675163393,
      "tokens_seen": 1631322112
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002554363089267803,
      "loss": 3.0357,
      "theoretical_loss": 3.4903525104843887,
      "tokens_seen": 1631387648
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025542627883650955,
      "loss": 2.8384,
      "theoretical_loss": 3.4903403464308767,
      "tokens_seen": 1631453184
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002554162487462387,
      "loss": 2.7632,
      "theoretical_loss": 3.4903281830027995,
      "tokens_seen": 1631518720
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002554062186559679,
      "loss": 2.8525,
      "theoretical_loss": 3.490316020200101,
      "tokens_seen": 1631584256
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002553961885656971,
      "loss": 2.8775,
      "theoretical_loss": 3.490303858022723,
      "tokens_seen": 1631649792
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002553861584754263,
      "loss": 2.5572,
      "theoretical_loss": 3.490291696470609,
      "tokens_seen": 1631715328
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025537612838515546,
      "loss": 2.6638,
      "theoretical_loss": 3.490279535543701,
      "tokens_seen": 1631780864
    },
    {
      "epoch": 0.49,
      "objective/train/advantage_avg": 0.49458548426628113,
      "objective/train/docs_used": 924504,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.9936306476593018,
      "objective/train/original_loss": 2.9936306476593018,
      "objective/train/theoretical_loss": 3.4902673752419417,
      "objective/train/tokens_used": 1652306400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24467633664608002,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507023334503174,
      "objective/train/weighted_lm_loss": 3.1453697681427,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0396549701690674,
      "theoretical_loss": 3.4902673752419417,
      "tokens_seen": 1631846400
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025536609829488464,
      "loss": 3.0381,
      "theoretical_loss": 3.4902673752419417,
      "tokens_seen": 1631846400
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002553560682046138,
      "loss": 2.953,
      "theoretical_loss": 3.490255215565275,
      "tokens_seen": 1631911936
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025534603811434305,
      "loss": 2.7578,
      "theoretical_loss": 3.490243056513642,
      "tokens_seen": 1631977472
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002553360080240722,
      "loss": 2.8772,
      "theoretical_loss": 3.4902308980869874,
      "tokens_seen": 1632043008
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002553259779338014,
      "loss": 2.9615,
      "theoretical_loss": 3.4902187402852523,
      "tokens_seen": 1632108544
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002553159478435306,
      "loss": 2.6559,
      "theoretical_loss": 3.49020658310838,
      "tokens_seen": 1632174080
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002553059177532598,
      "loss": 2.8398,
      "theoretical_loss": 3.4901944265563136,
      "tokens_seen": 1632239616
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025529588766298896,
      "loss": 2.8819,
      "theoretical_loss": 3.4901822706289956,
      "tokens_seen": 1632305152
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025528585757271814,
      "loss": 2.91,
      "theoretical_loss": 3.490170115326369,
      "tokens_seen": 1632370688
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002552758274824473,
      "loss": 2.6454,
      "theoretical_loss": 3.490157960648377,
      "tokens_seen": 1632436224
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025526579739217656,
      "loss": 2.597,
      "theoretical_loss": 3.4901458065949615,
      "tokens_seen": 1632501760
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002552557673019057,
      "loss": 2.9171,
      "theoretical_loss": 3.490133653166066,
      "tokens_seen": 1632567296
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002552457372116349,
      "loss": 2.7298,
      "theoretical_loss": 3.4901215003616333,
      "tokens_seen": 1632632832
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002552357071213641,
      "loss": 2.8431,
      "theoretical_loss": 3.4901093481816057,
      "tokens_seen": 1632698368
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002552256770310933,
      "loss": 2.486,
      "theoretical_loss": 3.4900971966259267,
      "tokens_seen": 1632763904
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002552156469408225,
      "loss": 2.8698,
      "theoretical_loss": 3.4900850456945385,
      "tokens_seen": 1632829440
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025520561685055164,
      "loss": 2.8221,
      "theoretical_loss": 3.490072895387385,
      "tokens_seen": 1632894976
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002551955867602809,
      "loss": 2.751,
      "theoretical_loss": 3.490060745704408,
      "tokens_seen": 1632960512
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025518555667001006,
      "loss": 2.7597,
      "theoretical_loss": 3.4900485966455514,
      "tokens_seen": 1633026048
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025517552657973924,
      "loss": 2.8273,
      "theoretical_loss": 3.4900364482107573,
      "tokens_seen": 1633091584
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002551654964894684,
      "loss": 2.7789,
      "theoretical_loss": 3.490024300399969,
      "tokens_seen": 1633157120
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002551554663991976,
      "loss": 2.8861,
      "theoretical_loss": 3.4900121532131294,
      "tokens_seen": 1633222656
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002551454363089268,
      "loss": 2.9092,
      "theoretical_loss": 3.490000006650181,
      "tokens_seen": 1633288192
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.000255135406218656,
      "loss": 2.5955,
      "theoretical_loss": 3.489987860711067,
      "tokens_seen": 1633353728
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.00025512537612838515,
      "loss": 2.6808,
      "theoretical_loss": 3.4899757153957305,
      "tokens_seen": 1633419264
    },
    {
      "epoch": 0.49,
      "objective/train/advantage_avg": 0.4941629469394684,
      "objective/train/docs_used": 925849,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.8862476348876953,
      "objective/train/original_loss": 2.886247158050537,
      "objective/train/theoretical_loss": 3.489963570704114,
      "objective/train/tokens_used": 1653944800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24426555633544922,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506579875946045,
      "objective/train/weighted_lm_loss": 3.032478094100952,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0408108234405518,
      "theoretical_loss": 3.489963570704114,
      "tokens_seen": 1633484800
    },
    {
      "epoch": 0.49,
      "learning_rate": 0.0002551153460381144,
      "loss": 2.8532,
      "theoretical_loss": 3.489963570704114,
      "tokens_seen": 1633484800
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002551053159478435,
      "loss": 2.87,
      "theoretical_loss": 3.489951426636161,
      "tokens_seen": 1633550336
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.00025509528585757274,
      "loss": 2.9356,
      "theoretical_loss": 3.489939283191814,
      "tokens_seen": 1633615872
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002550852557673019,
      "loss": 2.8073,
      "theoretical_loss": 3.4899271403710164,
      "tokens_seen": 1633681408
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002550752256770311,
      "loss": 2.7839,
      "theoretical_loss": 3.4899149981737105,
      "tokens_seen": 1633746944
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002550651955867603,
      "loss": 2.7628,
      "theoretical_loss": 3.4899028565998407,
      "tokens_seen": 1633812480
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.00025505516549648947,
      "loss": 2.6368,
      "theoretical_loss": 3.4898907156493477,
      "tokens_seen": 1633878016
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.00025504513540621865,
      "loss": 2.8297,
      "theoretical_loss": 3.4898785753221766,
      "tokens_seen": 1633943552
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002550351053159479,
      "loss": 2.9046,
      "theoretical_loss": 3.4898664356182696,
      "tokens_seen": 1634009088
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.000255025075225677,
      "loss": 2.8124,
      "theoretical_loss": 3.489854296537569,
      "tokens_seen": 1634074624
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.00025501504513540625,
      "loss": 3.0265,
      "theoretical_loss": 3.4898421580800187,
      "tokens_seen": 1634140160
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.00025500501504513543,
      "loss": 2.7784,
      "theoretical_loss": 3.489830020245562,
      "tokens_seen": 1634205696
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002549949849548646,
      "loss": 2.7544,
      "theoretical_loss": 3.4898178830341413,
      "tokens_seen": 1634271232
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002549849548645938,
      "loss": 2.7625,
      "theoretical_loss": 3.4898057464456995,
      "tokens_seen": 1634336768
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.00025497492477432297,
      "loss": 2.8494,
      "theoretical_loss": 3.4897936104801803,
      "tokens_seen": 1634402304
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.00025496489468405215,
      "loss": 2.8077,
      "theoretical_loss": 3.489781475137526,
      "tokens_seen": 1634467840
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002549548645937814,
      "loss": 2.7043,
      "theoretical_loss": 3.4897693404176806,
      "tokens_seen": 1634533376
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002549448345035105,
      "loss": 2.91,
      "theoretical_loss": 3.489757206320586,
      "tokens_seen": 1634598912
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.00025493480441323975,
      "loss": 2.6395,
      "theoretical_loss": 3.489745072846186,
      "tokens_seen": 1634664448
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002549247743229689,
      "loss": 2.8066,
      "theoretical_loss": 3.4897329399944237,
      "tokens_seen": 1634729984
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002549147442326981,
      "loss": 2.7733,
      "theoretical_loss": 3.489720807765242,
      "tokens_seen": 1634795520
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002549047141424273,
      "loss": 2.7051,
      "theoretical_loss": 3.489708676158584,
      "tokens_seen": 1634861056
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002548946840521565,
      "loss": 2.9257,
      "theoretical_loss": 3.489696545174393,
      "tokens_seen": 1634926592
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.00025488465396188566,
      "loss": 2.6927,
      "theoretical_loss": 3.489684414812612,
      "tokens_seen": 1634992128
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.00025487462387161484,
      "loss": 2.837,
      "theoretical_loss": 3.489672285073184,
      "tokens_seen": 1635057664
    },
    {
      "epoch": 0.5,
      "objective/train/advantage_avg": 0.49434131383895874,
      "objective/train/docs_used": 926453,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.629157304763794,
      "objective/train/original_loss": 2.629157304763794,
      "objective/train/theoretical_loss": 3.4896601559560523,
      "objective/train/tokens_used": 1655583200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24444015324115753,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050676703453064,
      "objective/train/weighted_lm_loss": 2.7624242305755615,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0387312173843384,
      "theoretical_loss": 3.4896601559560523,
      "tokens_seen": 1635123200
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.000254864593781344,
      "loss": 2.8253,
      "theoretical_loss": 3.4896601559560523,
      "tokens_seen": 1635123200
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.00025485456369107325,
      "loss": 2.7446,
      "theoretical_loss": 3.4896480274611594,
      "tokens_seen": 1635188736
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002548445336008024,
      "loss": 2.6612,
      "theoretical_loss": 3.4896358995884498,
      "tokens_seen": 1635254272
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002548345035105316,
      "loss": 2.9583,
      "theoretical_loss": 3.489623772337865,
      "tokens_seen": 1635319808
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002548244734202608,
      "loss": 2.661,
      "theoretical_loss": 3.48961164570935,
      "tokens_seen": 1635385344
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.00025481444332999,
      "loss": 2.6333,
      "theoretical_loss": 3.489599519702846,
      "tokens_seen": 1635450880
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.00025480441323971916,
      "loss": 2.8313,
      "theoretical_loss": 3.4895873943182973,
      "tokens_seen": 1635516416
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.00025479438314944834,
      "loss": 2.8839,
      "theoretical_loss": 3.4895752695556475,
      "tokens_seen": 1635581952
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002547843530591775,
      "loss": 2.8491,
      "theoretical_loss": 3.489563145414839,
      "tokens_seen": 1635647488
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.00025477432296890676,
      "loss": 2.807,
      "theoretical_loss": 3.4895510218958146,
      "tokens_seen": 1635713024
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002547642928786359,
      "loss": 2.6394,
      "theoretical_loss": 3.4895388989985188,
      "tokens_seen": 1635778560
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002547542627883651,
      "loss": 2.6574,
      "theoretical_loss": 3.4895267767228937,
      "tokens_seen": 1635844096
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.00025474423269809425,
      "loss": 2.7618,
      "theoretical_loss": 3.4895146550688834,
      "tokens_seen": 1635909632
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002547342026078235,
      "loss": 3.0108,
      "theoretical_loss": 3.48950253403643,
      "tokens_seen": 1635975168
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.00025472417251755266,
      "loss": 2.9998,
      "theoretical_loss": 3.489490413625478,
      "tokens_seen": 1636040704
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.00025471414242728184,
      "loss": 2.5646,
      "theoretical_loss": 3.4894782938359694,
      "tokens_seen": 1636106240
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.000254704112337011,
      "loss": 2.789,
      "theoretical_loss": 3.4894661746678484,
      "tokens_seen": 1636171776
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.00025469408224674026,
      "loss": 3.0366,
      "theoretical_loss": 3.489454056121058,
      "tokens_seen": 1636237312
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002546840521564694,
      "loss": 2.9078,
      "theoretical_loss": 3.489441938195541,
      "tokens_seen": 1636302848
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002546740220661986,
      "loss": 2.8981,
      "theoretical_loss": 3.489429820891241,
      "tokens_seen": 1636368384
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.00025466399197592775,
      "loss": 2.7586,
      "theoretical_loss": 3.4894177042081016,
      "tokens_seen": 1636433920
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.000254653961885657,
      "loss": 3.105,
      "theoretical_loss": 3.4894055881460657,
      "tokens_seen": 1636499456
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.00025464393179538616,
      "loss": 2.7213,
      "theoretical_loss": 3.4893934727050766,
      "tokens_seen": 1636564992
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.00025463390170511535,
      "loss": 2.9045,
      "theoretical_loss": 3.4893813578850783,
      "tokens_seen": 1636630528
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.00025462387161484453,
      "loss": 2.9568,
      "theoretical_loss": 3.4893692436860126,
      "tokens_seen": 1636696064
    },
    {
      "epoch": 0.5,
      "objective/train/advantage_avg": 0.4950801432132721,
      "objective/train/docs_used": 927487,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.4462389945983887,
      "objective/train/original_loss": 2.4462389945983887,
      "objective/train/theoretical_loss": 3.4893571301078246,
      "objective/train/tokens_used": 1657221600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2451682686805725,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507543087005615,
      "objective/train/weighted_lm_loss": 2.570343017578125,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0435127019882202,
      "theoretical_loss": 3.4893571301078246,
      "tokens_seen": 1636761600
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002546138415245737,
      "loss": 2.8017,
      "theoretical_loss": 3.4893571301078246,
      "tokens_seen": 1636761600
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002546038114343029,
      "loss": 2.5197,
      "theoretical_loss": 3.4893450171504563,
      "tokens_seen": 1636827136
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002545937813440321,
      "loss": 2.9831,
      "theoretical_loss": 3.4893329048138515,
      "tokens_seen": 1636892672
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.00025458375125376125,
      "loss": 3.0343,
      "theoretical_loss": 3.4893207930979533,
      "tokens_seen": 1636958208
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002545737211634905,
      "loss": 2.7225,
      "theoretical_loss": 3.4893086820027053,
      "tokens_seen": 1637023744
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002545636910732196,
      "loss": 2.7176,
      "theoretical_loss": 3.489296571528051,
      "tokens_seen": 1637089280
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.00025455366098294885,
      "loss": 2.8172,
      "theoretical_loss": 3.4892844616739334,
      "tokens_seen": 1637154816
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.00025454363089267803,
      "loss": 2.5321,
      "theoretical_loss": 3.489272352440296,
      "tokens_seen": 1637220352
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002545336008024072,
      "loss": 2.7443,
      "theoretical_loss": 3.4892602438270823,
      "tokens_seen": 1637285888
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002545235707121364,
      "loss": 2.8416,
      "theoretical_loss": 3.4892481358342353,
      "tokens_seen": 1637351424
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.00025451354062186563,
      "loss": 2.799,
      "theoretical_loss": 3.489236028461699,
      "tokens_seen": 1637416960
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002545035105315948,
      "loss": 2.8792,
      "theoretical_loss": 3.4892239217094163,
      "tokens_seen": 1637482496
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.000254493480441324,
      "loss": 3.0252,
      "theoretical_loss": 3.4892118155773306,
      "tokens_seen": 1637548032
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.00025448345035105317,
      "loss": 2.8652,
      "theoretical_loss": 3.4891997100653853,
      "tokens_seen": 1637613568
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.00025447342026078235,
      "loss": 2.8258,
      "theoretical_loss": 3.4891876051735244,
      "tokens_seen": 1637679104
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002544633901705116,
      "loss": 2.7418,
      "theoretical_loss": 3.489175500901691,
      "tokens_seen": 1637744640
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002544533600802407,
      "loss": 2.8908,
      "theoretical_loss": 3.489163397249828,
      "tokens_seen": 1637810176
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.00025444332998996995,
      "loss": 2.6334,
      "theoretical_loss": 3.4891512942178795,
      "tokens_seen": 1637875712
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002544332998996991,
      "loss": 2.8324,
      "theoretical_loss": 3.489139191805789,
      "tokens_seen": 1637941248
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002544232698094283,
      "loss": 2.9394,
      "theoretical_loss": 3.489127090013499,
      "tokens_seen": 1638006784
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002544132397191575,
      "loss": 2.7276,
      "theoretical_loss": 3.489114988840954,
      "tokens_seen": 1638072320
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002544032096288867,
      "loss": 2.73,
      "theoretical_loss": 3.4891028882880972,
      "tokens_seen": 1638137856
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.00025439317953861586,
      "loss": 2.5509,
      "theoretical_loss": 3.489090788354872,
      "tokens_seen": 1638203392
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.00025438314944834504,
      "loss": 2.8481,
      "theoretical_loss": 3.4890786890412215,
      "tokens_seen": 1638268928
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002543731193580742,
      "loss": 2.7181,
      "theoretical_loss": 3.48906659034709,
      "tokens_seen": 1638334464
    },
    {
      "debugging/Self-BLEU-5": 0.05314325699890558,
      "debugging/distinct-1-grams": 0.8123983739837398,
      "debugging/distinct-2-grams": 1.0,
      "debugging/entropy-1-grams": 4.670931226533002,
      "debugging/entropy-2-grams": 4.941642422609306,
      "debugging/length": 399.0,
      "debugging/num_segments": 2,
      "debugging/raw_token_scores_avg": 0.008247945457696915,
      "debugging/raw_token_scores_std": 0.010600144974887371,
      "debugging/score": 0.010745774580641056,
      "debugging/score_std": 0.006251392558169144,
      "epoch": 0.5,
      "objective/train/advantage_avg": 0.4917489290237427,
      "objective/train/docs_used": 928998,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.9667258262634277,
      "objective/train/original_loss": 2.9667258262634277,
      "objective/train/theoretical_loss": 3.4890544922724205,
      "objective/train/tokens_used": 1658860000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24192939698696136,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050404667854309,
      "objective/train/weighted_lm_loss": 3.1163172721862793,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 3.4890544922724205,
      "tokens_seen": 1638400000
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.00025436308926780345,
      "loss": 2.9572,
      "theoretical_loss": 3.4890544922724205,
      "tokens_seen": 1638400000
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002543530591775326,
      "loss": 2.7763,
      "theoretical_loss": 3.4890423948171563,
      "tokens_seen": 1638465536
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002543430290872618,
      "loss": 2.8681,
      "theoretical_loss": 3.4890302979812415,
      "tokens_seen": 1638531072
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.000254332998996991,
      "loss": 2.6752,
      "theoretical_loss": 3.489018201764619,
      "tokens_seen": 1638596608
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002543229689067202,
      "loss": 2.9078,
      "theoretical_loss": 3.489006106167233,
      "tokens_seen": 1638662144
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.00025431293881644936,
      "loss": 2.8262,
      "theoretical_loss": 3.4889940111890265,
      "tokens_seen": 1638727680
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.00025430290872617854,
      "loss": 2.9826,
      "theoretical_loss": 3.488981916829943,
      "tokens_seen": 1638793216
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002542928786359077,
      "loss": 2.8396,
      "theoretical_loss": 3.4889698230899264,
      "tokens_seen": 1638858752
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.00025428284854563696,
      "loss": 3.0798,
      "theoretical_loss": 3.48895772996892,
      "tokens_seen": 1638924288
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002542728184553661,
      "loss": 2.9566,
      "theoretical_loss": 3.488945637466868,
      "tokens_seen": 1638989824
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002542627883650953,
      "loss": 2.8524,
      "theoretical_loss": 3.4889335455837127,
      "tokens_seen": 1639055360
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.00025425275827482445,
      "loss": 2.7648,
      "theoretical_loss": 3.488921454319399,
      "tokens_seen": 1639120896
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002542427281845537,
      "loss": 2.7887,
      "theoretical_loss": 3.4889093636738697,
      "tokens_seen": 1639186432
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.00025423269809428286,
      "loss": 2.8545,
      "theoretical_loss": 3.4888972736470687,
      "tokens_seen": 1639251968
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.00025422266800401204,
      "loss": 2.7274,
      "theoretical_loss": 3.4888851842389395,
      "tokens_seen": 1639317504
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002542126379137412,
      "loss": 3.0386,
      "theoretical_loss": 3.488873095449426,
      "tokens_seen": 1639383040
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.00025420260782347046,
      "loss": 2.9751,
      "theoretical_loss": 3.488861007278471,
      "tokens_seen": 1639448576
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002541925777331996,
      "loss": 2.566,
      "theoretical_loss": 3.488848919726019,
      "tokens_seen": 1639514112
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002541825476429288,
      "loss": 2.8619,
      "theoretical_loss": 3.488836832792013,
      "tokens_seen": 1639579648
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.00025417251755265795,
      "loss": 3.0797,
      "theoretical_loss": 3.4888247464763973,
      "tokens_seen": 1639645184
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002541624874623872,
      "loss": 3.1269,
      "theoretical_loss": 3.4888126607791152,
      "tokens_seen": 1639710720
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.00025415245737211637,
      "loss": 2.9003,
      "theoretical_loss": 3.48880057570011,
      "tokens_seen": 1639776256
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.00025414242728184555,
      "loss": 2.6242,
      "theoretical_loss": 3.488788491239326,
      "tokens_seen": 1639841792
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.00025413239719157473,
      "loss": 2.6592,
      "theoretical_loss": 3.4887764073967067,
      "tokens_seen": 1639907328
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002541223671013039,
      "loss": 2.9769,
      "theoretical_loss": 3.4887643241721955,
      "tokens_seen": 1639972864
    },
    {
      "epoch": 0.5,
      "objective/train/advantage_avg": 0.4963871240615845,
      "objective/train/docs_used": 929588,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.706629753112793,
      "objective/train/original_loss": 2.706629514694214,
      "objective/train/theoretical_loss": 3.488752241565736,
      "objective/train/tokens_used": 1660498400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.2464451789855957,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0508915185928345,
      "objective/train/weighted_lm_loss": 2.8443093299865723,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.044721245765686,
      "theoretical_loss": 3.488752241565736,
      "tokens_seen": 1640038400
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002541123370110331,
      "loss": 2.818,
      "theoretical_loss": 3.488752241565736,
      "tokens_seen": 1640038400
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002541023069207623,
      "loss": 2.6685,
      "theoretical_loss": 3.488740159577273,
      "tokens_seen": 1640103936
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.00025409227683049145,
      "loss": 2.7436,
      "theoretical_loss": 3.4887280782067487,
      "tokens_seen": 1640169472
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002540822467402207,
      "loss": 2.8258,
      "theoretical_loss": 3.488715997454107,
      "tokens_seen": 1640235008
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002540722166499498,
      "loss": 2.925,
      "theoretical_loss": 3.4887039173192926,
      "tokens_seen": 1640300544
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.00025406218655967905,
      "loss": 2.8967,
      "theoretical_loss": 3.488691837802249,
      "tokens_seen": 1640366080
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.00025405215646940823,
      "loss": 2.7499,
      "theoretical_loss": 3.488679758902919,
      "tokens_seen": 1640431616
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002540421263791374,
      "loss": 3.0068,
      "theoretical_loss": 3.488667680621247,
      "tokens_seen": 1640497152
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002540320962888666,
      "loss": 2.5961,
      "theoretical_loss": 3.488655602957177,
      "tokens_seen": 1640562688
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.00025402206619859583,
      "loss": 2.8665,
      "theoretical_loss": 3.4886435259106525,
      "tokens_seen": 1640628224
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.00025401203610832495,
      "loss": 3.0237,
      "theoretical_loss": 3.4886314494816166,
      "tokens_seen": 1640693760
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002540020060180542,
      "loss": 2.6296,
      "theoretical_loss": 3.488619373670014,
      "tokens_seen": 1640759296
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002539919759277833,
      "loss": 3.0681,
      "theoretical_loss": 3.488607298475788,
      "tokens_seen": 1640824832
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.00025398194583751255,
      "loss": 2.7168,
      "theoretical_loss": 3.4885952238988827,
      "tokens_seen": 1640890368
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.00025397191574724173,
      "loss": 2.667,
      "theoretical_loss": 3.4885831499392417,
      "tokens_seen": 1640955904
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002539618856569709,
      "loss": 2.8561,
      "theoretical_loss": 3.4885710765968088,
      "tokens_seen": 1641021440
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002539518555667001,
      "loss": 2.7657,
      "theoretical_loss": 3.4885590038715275,
      "tokens_seen": 1641086976
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002539418254764293,
      "loss": 2.9576,
      "theoretical_loss": 3.4885469317633424,
      "tokens_seen": 1641152512
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.00025393179538615846,
      "loss": 2.7525,
      "theoretical_loss": 3.488534860272196,
      "tokens_seen": 1641218048
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002539217652958877,
      "loss": 2.9085,
      "theoretical_loss": 3.4885227893980337,
      "tokens_seen": 1641283584
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002539117352056168,
      "loss": 2.8959,
      "theoretical_loss": 3.4885107191407982,
      "tokens_seen": 1641349120
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.00025390170511534606,
      "loss": 2.9413,
      "theoretical_loss": 3.4884986495004338,
      "tokens_seen": 1641414656
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002538916750250752,
      "loss": 2.9272,
      "theoretical_loss": 3.4884865804768843,
      "tokens_seen": 1641480192
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002538816449348044,
      "loss": 2.7413,
      "theoretical_loss": 3.488474512070094,
      "tokens_seen": 1641545728
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002538716148445336,
      "loss": 2.9855,
      "theoretical_loss": 3.488462444280006,
      "tokens_seen": 1641611264
    },
    {
      "epoch": 0.5,
      "objective/train/advantage_avg": 0.49393683671951294,
      "objective/train/docs_used": 930445,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.6322455406188965,
      "objective/train/original_loss": 2.6322453022003174,
      "objective/train/theoretical_loss": 3.4884503771065636,
      "objective/train/tokens_used": 1662136800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24403800070285797,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506342649459839,
      "objective/train/weighted_lm_loss": 2.7654330730438232,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0457526445388794,
      "theoretical_loss": 3.4884503771065636,
      "tokens_seen": 1641676800
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002538615847542628,
      "loss": 2.8298,
      "theoretical_loss": 3.4884503771065636,
      "tokens_seen": 1641676800
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.00025385155466399196,
      "loss": 2.5927,
      "theoretical_loss": 3.488438310549712,
      "tokens_seen": 1641742336
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002538415245737212,
      "loss": 3.0995,
      "theoretical_loss": 3.488426244609395,
      "tokens_seen": 1641807872
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002538314944834503,
      "loss": 2.6512,
      "theoretical_loss": 3.488414179285556,
      "tokens_seen": 1641873408
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.00025382146439317956,
      "loss": 3.0594,
      "theoretical_loss": 3.4884021145781388,
      "tokens_seen": 1641938944
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002538114343029087,
      "loss": 2.8792,
      "theoretical_loss": 3.4883900504870873,
      "tokens_seen": 1642004480
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002538014042126379,
      "loss": 2.7686,
      "theoretical_loss": 3.4883779870123455,
      "tokens_seen": 1642070016
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002537913741223671,
      "loss": 2.6846,
      "theoretical_loss": 3.4883659241538583,
      "tokens_seen": 1642135552
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002537813440320963,
      "loss": 2.9998,
      "theoretical_loss": 3.488353861911568,
      "tokens_seen": 1642201088
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.00025377131394182546,
      "loss": 2.8291,
      "theoretical_loss": 3.4883418002854194,
      "tokens_seen": 1642266624
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.00025376128385155465,
      "loss": 3.0385,
      "theoretical_loss": 3.4883297392753563,
      "tokens_seen": 1642332160
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002537512537612839,
      "loss": 2.7869,
      "theoretical_loss": 3.4883176788813226,
      "tokens_seen": 1642397696
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.00025374122367101306,
      "loss": 2.709,
      "theoretical_loss": 3.4883056191032624,
      "tokens_seen": 1642463232
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.00025373119358074224,
      "loss": 2.5096,
      "theoretical_loss": 3.48829355994112,
      "tokens_seen": 1642528768
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002537211634904714,
      "loss": 2.4255,
      "theoretical_loss": 3.4882815013948383,
      "tokens_seen": 1642594304
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.00025371113340020066,
      "loss": 2.5894,
      "theoretical_loss": 3.488269443464363,
      "tokens_seen": 1642659840
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002537011033099298,
      "loss": 2.7588,
      "theoretical_loss": 3.4882573861496358,
      "tokens_seen": 1642725376
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.000253691073219659,
      "loss": 2.8352,
      "theoretical_loss": 3.488245329450603,
      "tokens_seen": 1642790912
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.00025368104312938815,
      "loss": 2.8195,
      "theoretical_loss": 3.4882332733672063,
      "tokens_seen": 1642856448
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002536710130391174,
      "loss": 2.9174,
      "theoretical_loss": 3.4882212178993917,
      "tokens_seen": 1642921984
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.00025366098294884657,
      "loss": 2.6061,
      "theoretical_loss": 3.4882091630471024,
      "tokens_seen": 1642987520
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.00025365095285857575,
      "loss": 2.7594,
      "theoretical_loss": 3.488197108810282,
      "tokens_seen": 1643053056
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.00025364092276830493,
      "loss": 2.6762,
      "theoretical_loss": 3.488185055188876,
      "tokens_seen": 1643118592
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002536308926780341,
      "loss": 2.8182,
      "theoretical_loss": 3.4881730021828266,
      "tokens_seen": 1643184128
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002536208625877633,
      "loss": 2.6504,
      "theoretical_loss": 3.488160949792079,
      "tokens_seen": 1643249664
    },
    {
      "epoch": 0.5,
      "objective/train/advantage_avg": 0.4946185350418091,
      "objective/train/docs_used": 931444,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.4197051525115967,
      "objective/train/original_loss": 2.4197049140930176,
      "objective/train/theoretical_loss": 3.4881488980165765,
      "objective/train/tokens_used": 1663775200,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24470826983451843,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507057905197144,
      "objective/train/weighted_lm_loss": 2.542180299758911,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.044721245765686,
      "theoretical_loss": 3.4881488980165765,
      "tokens_seen": 1643315200
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002536108324974925,
      "loss": 2.7531,
      "theoretical_loss": 3.4881488980165765,
      "tokens_seen": 1643315200
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.00025360080240722165,
      "loss": 2.8824,
      "theoretical_loss": 3.488136846856264,
      "tokens_seen": 1643380736
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002535907723169509,
      "loss": 2.9266,
      "theoretical_loss": 3.488124796311085,
      "tokens_seen": 1643446272
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.00025358074222668,
      "loss": 2.8624,
      "theoretical_loss": 3.488112746380984,
      "tokens_seen": 1643511808
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.00025357071213640925,
      "loss": 2.6461,
      "theoretical_loss": 3.4881006970659048,
      "tokens_seen": 1643577344
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.00025356068204613843,
      "loss": 3.0007,
      "theoretical_loss": 3.488088648365791,
      "tokens_seen": 1643642880
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002535506519558676,
      "loss": 2.5948,
      "theoretical_loss": 3.4880766002805874,
      "tokens_seen": 1643708416
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002535406218655968,
      "loss": 2.5908,
      "theoretical_loss": 3.488064552810238,
      "tokens_seen": 1643773952
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.00025353059177532603,
      "loss": 2.9097,
      "theoretical_loss": 3.4880525059546867,
      "tokens_seen": 1643839488
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.00025352056168505516,
      "loss": 2.8957,
      "theoretical_loss": 3.4880404597138783,
      "tokens_seen": 1643905024
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002535105315947844,
      "loss": 2.9163,
      "theoretical_loss": 3.488028414087756,
      "tokens_seen": 1643970560
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002535005015045135,
      "loss": 2.8728,
      "theoretical_loss": 3.488016369076264,
      "tokens_seen": 1644036096
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.00025349047141424275,
      "loss": 2.7761,
      "theoretical_loss": 3.4880043246793466,
      "tokens_seen": 1644101632
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.00025348044132397193,
      "loss": 2.8703,
      "theoretical_loss": 3.4879922808969486,
      "tokens_seen": 1644167168
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002534704112337011,
      "loss": 3.0026,
      "theoretical_loss": 3.487980237729013,
      "tokens_seen": 1644232704
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002534603811434303,
      "loss": 2.7567,
      "theoretical_loss": 3.487968195175485,
      "tokens_seen": 1644298240
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002534503510531595,
      "loss": 3.1427,
      "theoretical_loss": 3.4879561532363086,
      "tokens_seen": 1644363776
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.00025344032096288866,
      "loss": 2.7097,
      "theoretical_loss": 3.4879441119114274,
      "tokens_seen": 1644429312
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002534302908726179,
      "loss": 3.0011,
      "theoretical_loss": 3.4879320712007855,
      "tokens_seen": 1644494848
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.000253420260782347,
      "loss": 2.8492,
      "theoretical_loss": 3.487920031104328,
      "tokens_seen": 1644560384
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.00025341023069207626,
      "loss": 3.0944,
      "theoretical_loss": 3.4879079916219986,
      "tokens_seen": 1644625920
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002534002006018054,
      "loss": 2.8948,
      "theoretical_loss": 3.4878959527537416,
      "tokens_seen": 1644691456
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002533901705115346,
      "loss": 2.6576,
      "theoretical_loss": 3.487883914499501,
      "tokens_seen": 1644756992
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002533801404212638,
      "loss": 2.5113,
      "theoretical_loss": 3.487871876859221,
      "tokens_seen": 1644822528
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.000253370110330993,
      "loss": 2.8171,
      "theoretical_loss": 3.487859839832846,
      "tokens_seen": 1644888064
    },
    {
      "epoch": 0.5,
      "objective/train/advantage_avg": 0.4943881034851074,
      "objective/train/docs_used": 932028,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.4818482398986816,
      "objective/train/original_loss": 2.4818482398986816,
      "objective/train/theoretical_loss": 3.48784780342032,
      "objective/train/tokens_used": 1665413600,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24449029564857483,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0506815910339355,
      "objective/train/weighted_lm_loss": 2.6075377464294434,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0363596677780151,
      "theoretical_loss": 3.48784780342032,
      "tokens_seen": 1644953600
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.00025336008024072216,
      "loss": 2.7244,
      "theoretical_loss": 3.48784780342032,
      "tokens_seen": 1644953600
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002533500501504514,
      "loss": 2.7454,
      "theoretical_loss": 3.487835767621588,
      "tokens_seen": 1645019136
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002533400200601805,
      "loss": 2.9094,
      "theoretical_loss": 3.4878237324365933,
      "tokens_seen": 1645084672
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.00025332998996990976,
      "loss": 2.825,
      "theoretical_loss": 3.48781169786528,
      "tokens_seen": 1645150208
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002533199598796389,
      "loss": 2.8359,
      "theoretical_loss": 3.4877996639075937,
      "tokens_seen": 1645215744
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002533099297893681,
      "loss": 2.7026,
      "theoretical_loss": 3.487787630563478,
      "tokens_seen": 1645281280
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002532998996990973,
      "loss": 2.876,
      "theoretical_loss": 3.4877755978328766,
      "tokens_seen": 1645346816
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002532898696088265,
      "loss": 2.5755,
      "theoretical_loss": 3.487763565715734,
      "tokens_seen": 1645412352
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.00025327983951855566,
      "loss": 3.0438,
      "theoretical_loss": 3.4877515342119954,
      "tokens_seen": 1645477888
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.00025326980942828485,
      "loss": 2.6767,
      "theoretical_loss": 3.487739503321604,
      "tokens_seen": 1645543424
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.000253259779338014,
      "loss": 2.9239,
      "theoretical_loss": 3.4877274730445045,
      "tokens_seen": 1645608960
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.00025324974924774326,
      "loss": 2.9508,
      "theoretical_loss": 3.487715443380641,
      "tokens_seen": 1645674496
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002532397191574724,
      "loss": 2.7155,
      "theoretical_loss": 3.4877034143299586,
      "tokens_seen": 1645740032
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002532296890672016,
      "loss": 3.0338,
      "theoretical_loss": 3.4876913858924015,
      "tokens_seen": 1645805568
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.00025321965897693075,
      "loss": 2.8994,
      "theoretical_loss": 3.487679358067913,
      "tokens_seen": 1645871104
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.00025320962888666,
      "loss": 2.9522,
      "theoretical_loss": 3.4876673308564383,
      "tokens_seen": 1645936640
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.00025319959879638917,
      "loss": 2.8318,
      "theoretical_loss": 3.4876553042579213,
      "tokens_seen": 1646002176
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.00025318956870611835,
      "loss": 2.6484,
      "theoretical_loss": 3.4876432782723064,
      "tokens_seen": 1646067712
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.00025317953861584753,
      "loss": 2.7806,
      "theoretical_loss": 3.487631252899538,
      "tokens_seen": 1646133248
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.00025316950852557677,
      "loss": 2.7733,
      "theoretical_loss": 3.4876192281395615,
      "tokens_seen": 1646198784
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002531594784353059,
      "loss": 2.5786,
      "theoretical_loss": 3.4876072039923196,
      "tokens_seen": 1646264320
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.00025314944834503513,
      "loss": 2.7265,
      "theoretical_loss": 3.487595180457758,
      "tokens_seen": 1646329856
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.00025313941825476425,
      "loss": 2.9088,
      "theoretical_loss": 3.4875831575358207,
      "tokens_seen": 1646395392
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002531293881644935,
      "loss": 2.851,
      "theoretical_loss": 3.4875711352264513,
      "tokens_seen": 1646460928
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.00025311935807422267,
      "loss": 3.0195,
      "theoretical_loss": 3.4875591135295956,
      "tokens_seen": 1646526464
    },
    {
      "epoch": 0.5,
      "objective/train/advantage_avg": 0.49495929479599,
      "objective/train/docs_used": 933210,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.307584762573242,
      "objective/train/original_loss": 2.307584524154663,
      "objective/train/theoretical_loss": 3.487547092445197,
      "objective/train/tokens_used": 1667052000,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24504445493221283,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0507415533065796,
      "objective/train/weighted_lm_loss": 2.4246318340301514,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0417574644088745,
      "theoretical_loss": 3.487547092445197,
      "tokens_seen": 1646592000
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.00025310932798395185,
      "loss": 2.7792,
      "theoretical_loss": 3.487547092445197,
      "tokens_seen": 1646592000
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.00025309929789368103,
      "loss": 2.7416,
      "theoretical_loss": 3.4875350719732,
      "tokens_seen": 1646657536
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002530892678034102,
      "loss": 2.7978,
      "theoretical_loss": 3.48752305211355,
      "tokens_seen": 1646723072
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002530792377131394,
      "loss": 2.8744,
      "theoretical_loss": 3.4875110328661902,
      "tokens_seen": 1646788608
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.00025306920762286863,
      "loss": 2.9077,
      "theoretical_loss": 3.4874990142310653,
      "tokens_seen": 1646854144
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.00025305917753259776,
      "loss": 2.7541,
      "theoretical_loss": 3.48748699620812,
      "tokens_seen": 1646919680
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.000253049147442327,
      "loss": 2.6716,
      "theoretical_loss": 3.4874749787972994,
      "tokens_seen": 1646985216
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002530391173520562,
      "loss": 2.3567,
      "theoretical_loss": 3.4874629619985473,
      "tokens_seen": 1647050752
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.00025302908726178536,
      "loss": 2.8629,
      "theoretical_loss": 3.4874509458118075,
      "tokens_seen": 1647116288
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.00025301905717151454,
      "loss": 2.4273,
      "theoretical_loss": 3.4874389302370252,
      "tokens_seen": 1647181824
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002530090270812437,
      "loss": 2.8981,
      "theoretical_loss": 3.4874269152741455,
      "tokens_seen": 1647247360
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.00025299899699097295,
      "loss": 2.8629,
      "theoretical_loss": 3.4874149009231123,
      "tokens_seen": 1647312896
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.00025298896690070213,
      "loss": 2.902,
      "theoretical_loss": 3.4874028871838694,
      "tokens_seen": 1647378432
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002529789368104313,
      "loss": 2.5588,
      "theoretical_loss": 3.4873908740563624,
      "tokens_seen": 1647443968
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002529689067201605,
      "loss": 3.0254,
      "theoretical_loss": 3.4873788615405354,
      "tokens_seen": 1647509504
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002529588766298897,
      "loss": 2.8036,
      "theoretical_loss": 3.4873668496363326,
      "tokens_seen": 1647575040
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.00025294884653961886,
      "loss": 2.7347,
      "theoretical_loss": 3.487354838343699,
      "tokens_seen": 1647640576
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002529388164493481,
      "loss": 2.8725,
      "theoretical_loss": 3.4873428276625793,
      "tokens_seen": 1647706112
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002529287863590772,
      "loss": 2.8533,
      "theoretical_loss": 3.487330817592918,
      "tokens_seen": 1647771648
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.00025291875626880646,
      "loss": 2.7687,
      "theoretical_loss": 3.4873188081346584,
      "tokens_seen": 1647837184
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002529087261785356,
      "loss": 2.691,
      "theoretical_loss": 3.4873067992877465,
      "tokens_seen": 1647902720
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002528986960882648,
      "loss": 2.8841,
      "theoretical_loss": 3.4872947910521264,
      "tokens_seen": 1647968256
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.000252888665997994,
      "loss": 2.6404,
      "theoretical_loss": 3.487282783427743,
      "tokens_seen": 1648033792
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002528786359077232,
      "loss": 2.8373,
      "theoretical_loss": 3.4872707764145403,
      "tokens_seen": 1648099328
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.00025286860581745236,
      "loss": 2.7218,
      "theoretical_loss": 3.4872587700124633,
      "tokens_seen": 1648164864
    },
    {
      "epoch": 0.5,
      "objective/train/advantage_avg": 0.493236780166626,
      "objective/train/docs_used": 933788,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.499129295349121,
      "objective/train/original_loss": 2.499129295349121,
      "objective/train/theoretical_loss": 3.4872467642214566,
      "objective/train/tokens_used": 1668690400,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24335020780563354,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.050560712814331,
      "objective/train/weighted_lm_loss": 2.625607490539551,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0382119417190552,
      "theoretical_loss": 3.4872467642214566,
      "tokens_seen": 1648230400
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002528585757271816,
      "loss": 2.9238,
      "theoretical_loss": 3.4872467642214566,
      "tokens_seen": 1648230400
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002528485456369107,
      "loss": 2.6608,
      "theoretical_loss": 3.4872347590414643,
      "tokens_seen": 1648295936
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.00025283851554663996,
      "loss": 2.6332,
      "theoretical_loss": 3.4872227544724312,
      "tokens_seen": 1648361472
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002528284854563691,
      "loss": 2.939,
      "theoretical_loss": 3.487210750514303,
      "tokens_seen": 1648427008
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002528184553660983,
      "loss": 2.7753,
      "theoretical_loss": 3.4871987471670227,
      "tokens_seen": 1648492544
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002528084252758275,
      "loss": 2.822,
      "theoretical_loss": 3.487186744430536,
      "tokens_seen": 1648558080
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002527983951855567,
      "loss": 2.6574,
      "theoretical_loss": 3.487174742304787,
      "tokens_seen": 1648623616
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.00025278836509528586,
      "loss": 2.9318,
      "theoretical_loss": 3.4871627407897203,
      "tokens_seen": 1648689152
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.00025277833500501505,
      "loss": 2.7251,
      "theoretical_loss": 3.487150739885281,
      "tokens_seen": 1648754688
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002527683049147442,
      "loss": 2.8147,
      "theoretical_loss": 3.4871387395914137,
      "tokens_seen": 1648820224
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.00025275827482447346,
      "loss": 2.6562,
      "theoretical_loss": 3.4871267399080628,
      "tokens_seen": 1648885760
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002527482447342026,
      "loss": 2.7203,
      "theoretical_loss": 3.4871147408351733,
      "tokens_seen": 1648951296
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002527382146439318,
      "loss": 2.7867,
      "theoretical_loss": 3.4871027423726897,
      "tokens_seen": 1649016832
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.00025272818455366095,
      "loss": 2.6785,
      "theoretical_loss": 3.4870907445205566,
      "tokens_seen": 1649082368
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002527181544633902,
      "loss": 2.5438,
      "theoretical_loss": 3.4870787472787184,
      "tokens_seen": 1649147904
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.00025270812437311937,
      "loss": 2.7822,
      "theoretical_loss": 3.487066750647121,
      "tokens_seen": 1649213440
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.00025269809428284855,
      "loss": 2.6451,
      "theoretical_loss": 3.4870547546257074,
      "tokens_seen": 1649278976
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.00025268806419257773,
      "loss": 2.9999,
      "theoretical_loss": 3.4870427592144235,
      "tokens_seen": 1649344512
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.00025267803410230697,
      "loss": 2.7043,
      "theoretical_loss": 3.487030764413214,
      "tokens_seen": 1649410048
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002526680040120361,
      "loss": 2.6582,
      "theoretical_loss": 3.487018770222023,
      "tokens_seen": 1649475584
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.00025265797392176533,
      "loss": 2.7264,
      "theoretical_loss": 3.487006776640796,
      "tokens_seen": 1649541120
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.00025264794383149445,
      "loss": 2.6474,
      "theoretical_loss": 3.486994783669477,
      "tokens_seen": 1649606656
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002526379137412237,
      "loss": 2.7087,
      "theoretical_loss": 3.4869827913080114,
      "tokens_seen": 1649672192
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.00025262788365095287,
      "loss": 3.1052,
      "theoretical_loss": 3.486970799556344,
      "tokens_seen": 1649737728
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.00025261785356068205,
      "loss": 2.7392,
      "theoretical_loss": 3.4869588084144185,
      "tokens_seen": 1649803264
    },
    {
      "epoch": 0.5,
      "objective/train/advantage_avg": 0.492251455783844,
      "objective/train/docs_used": 934285,
      "objective/train/instantaneous_batch_size": 16,
      "objective/train/instantaneous_microbatch_size": 16384,
      "objective/train/lm_loss": 2.696399450302124,
      "objective/train/original_loss": 2.696399450302124,
      "objective/train/theoretical_loss": 3.486946817882181,
      "objective/train/tokens_used": 1670328800,
      "objective/train/value_avg": -0.5,
      "objective/train/value_loss": 0.24252144992351532,
      "objective/train/value_max": -0.5,
      "objective/train/value_min": -0.5,
      "objective/train/value_reward_corr": NaN,
      "objective/train/value_std": 0.0,
      "objective/train/weight_avg": 1.0504578351974487,
      "objective/train/weighted_lm_loss": 2.8322670459747314,
      "objective/train/weights_max": 1.0512710809707642,
      "objective/train/weights_min": 1.0396549701690674,
      "theoretical_loss": 3.486946817882181,
      "tokens_seen": 1649868800
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.00025260782347041123,
      "loss": 2.7446,
      "theoretical_loss": 3.486946817882181,
      "tokens_seen": 1649868800
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002525977933801404,
      "loss": 3.0308,
      "theoretical_loss": 3.4869348279595758,
      "tokens_seen": 1649934336
    },
    {
      "epoch": 0.5,
      "learning_rate": 0.0002525877632898696,
      "loss": 2.901,
      "theoretical_loss": 3.486922838646547,
      "tokens_seen": 1649999872
    }
  ],
  "max_steps": 50354,
  "num_train_epochs": 9223372036854775807,
  "total_flos": 8.420619378764022e+17,
  "trial_name": null,
  "trial_params": null
}